Or how we forgot we can adjudicate the intelligence of arguments

By MARTIN REZNY

Has Artificial General Intelligence arrived? If you believe the executives of various leading tech companies, AGI has arrived several times already. This year alone.

Of course, no one should be believing them. If it’s true that anything we can do, AI can do better, then it should be proven before anyone believes it.

So, how did they test for general intelligence? Using a number of benchmarks, which tend to fall into one of a few categories:

• How well the AI can match and transform patterns.
• How well the AI can answer difficult expert questions.
• How well the AI can code.

All of these approaches have one major thing in common. A thing which means they cover only a small part of what’s generally called reasoning. Can you guess what that commonality is? It’s a bit of a trick question, to be honest, but only if you’re an expert in a STEM field.

The commonality of these testing strategies is that they almost exclusively test for factual and/or formal problems. Problems where the answers are objective, clear, and knowable. That’s less than half of all reasoning problems, and only one of the many different reasoning operations.

Specifically, the other categories of problems that can be reasoned about are informal value and plan problems, and the one reasoning operation being tested there is problem-solving. You could think of problem-solving as a category of its own, encapsulating answer finding, pattern detection, optimization, maximization, and probably a few other techniques.

But while problem-solving is one of the larger reasoning categories, and while it is very important, the landscape of reasoning is much grander. Beyond true or false, humans can contemplate good versus bad and right versus wrong. We can also entertain and “generate” valid neologisms and counterfactuals, as well as transcend games by changing their rules.

Beyond creativity, metacognition, and hypergaming, the most complex feat of reasoning is arguably the dialectic, or the art of putting arguments to the test against other arguments, to see which survive the competition. Whenever AI developers talk about “chains of reasoning” or when they have multiple AI agents engage in “debates”, that should be dialectic.

Spoiler alert: It’s not dialectic.

Just because you prompt the AI to give you a narrative about an inference it made, that doesn’t make it reasoning. And just because those narratives are given in a chain, that doesn’t make it a chain of reasoning.

Also, two or more parties saying arguments against each other, and somebody deciding which of the arguments they like more, doesn’t make that a debate.

If you’re wondering how arguments and debates are supposed to work, that’s a question that does have an answer. Your AI of choice knows that answer. It’s in its training materials. The problem is that you probably don’t know how to properly ask the question, without already knowing the answer yourself.

How to Lose an Entire Faculty of Intelligence

While some technical specifics have only been codified relatively recently in written form by practitioners of debating competitions (it’s an oral culture), the answers to how dialectic and rhetoric are supposed to function in practice have been known for over two thousand years.

If you want to do some research, or ask an AI, you could start with Phaedrus by Plato or Rhetoric by Aristotle. Or just look at everything Socrates ever said or did. And then read all the manuals for debate adjudicators. Or at least one. They’re really all the same manual.

Interestingly, as I was writing one of those manuals myself, mostly using my original experience and thoughts, I later found out I converged on the same general principles as most other debate manual authors. You don’t have to call this science, if you don’t want to, but it works.

This is why I said earlier that I was asking a trick question, but only if you happen to be a STEM expert. As far as most STEM experts are concerned, rhetoric isn’t scientific; arguing about values is not much more than post hoc (after the fact) rationalization; and informal logic is mostly “vibes”.

Which is an argument that could hold some weight. If these same STEM experts then didn’t proceed to make AIs produce definite rationalizations, and have annotators judge the exchanges of rationalizations purely on vibes throughout maximally reductive and superficial RLHF procedures.

If logic is a thing we should care about, here are some of my counterpoints for AI experts:

• An appearance of dialectic is a different thing than the actual dialectic.
• An appearance of dialectic cannot possibly be a better test of intelligence than the actual dialectic.
• If you don’t believe in dialectic, why are you setting up an appearance of dialectic in the first place, let alone claiming that appearance of dialectic proves anything?
• If you do believe in dialectic, why don’t you hire experts in dialectic to set up an actual test of dialectic or train AIs in actual dialectic?

At this point, maybe you’re wondering, if dialectic is a known and documented thing, why do the people building AIs appear to be completely clueless about it? In short, the reasons are historical. For example, at your school, whatever school it was, was debating part of your curricula?

Statistically, the answer is no. Pretty much all over the world. Maybe you had it as an optional extra-curricular activity, if you got lucky. The answer has been mostly no for as long as modern schools have existed, in fact.

The only educational program that had a mix of rhetoric or dialectic as a standard component was the Greco-Roman trivium. Monotheistic religions that came later certainly didn’t promote true dialectical reasoning, A.K.A. critical thinking, but it was the scientific revolution that ended dialectic.

The last big moment of this so-called classical liberal arts education was during the renaissance at European universities. Last holdouts began to disappear roughly around the 18th century. This is because with the scientific revolution, intellectuals adopted a literary, analytical culture.

Every time somebody says that people need to read more books to become smarter, that’s where that comes from. I’m not saying reading books is bad. Although some ancient philosophers like Plato from rhetorical cultures literally did argue that too much reading was bad for intelligence.

What I’m trying to say is that reading books is only one way of developing intelligence, and it heavily skews minds toward factual, grammatical, and formal reasoning. It’s probably a net positive that there’s high literacy today. Just like it’s a net negative there’s very low oracy.

Yes, there’s more to intelligence and education than literacy. Literacy just means you can read and write. If you can count, you have the faculty of numeracy. If you can reason and argue, you have the faculty of oracy. Contrary to the popular belief of modern intelligentsia, reading doesn’t lead to true development of reasoning skills. It only makes you informed.

A sufficiently informed person can memorize and execute formulas. People with higher innate IQ can perhaps do that faster than others and with fewer errors, but that’s not a difference in kind. True reasoning isn’t about solving solved problems, or solving them more efficiently. True reasoning isn’t even about progressing within predetermined paradigms.

The closest you get to reading about what true reasoning is is when you take a course about the methodology of scientific inquiry. There, you will be introduced to general types of reasoning processes. They are as follows:

• Induction — generalization from observation. Based on available evidence, you’re guessing what’s probably the case beyond what you can immediately observe. For example, you see a bunch of white swans, so you assume swans are probably all white. AIs can do this because you can use probabilities and correlations (averaging and random functions) to come up with a range of generalizations iteratively. You can be wrong, and often you will be, so induction can solve some problems sometimes, but with uncertainty and unreliability.
• Deduction — verification or falsification through experiment, based on logically deriving necessary conclusions from premises. When the objects and rules within a particular situation are known (or clearly hypothesized), you can set up tests. Through the outcomes of such tests, you can clearly see whether your hypotheses were correct or incorrect. For example, if Socrates is a man, and all men are mortal, ergo Socrates is mortal. This is a reliable formal method of problem-solving reasoning, but very limited and brittle in the fuzzy world we happen to inhabit. This is why Good Old Fashioned deterministic AI didn’t suffice.
• Abduction — inference to the best explanation, or most plausible cause; not to be confused with what aliens are allegedly doing. Although if what they’re doing is research, maybe it is the same thing. You observe an outcome, go through a list of possible explanations that you can think of, and pick the best one. For example, if you walk outside your house, and you see everything is wet, you infer it probably was raining. In practice, you can be wrong, of course. It could be wet because of anything, including aliens. You do the best you can with what you got.

This is as far as the textbooks get, and they don’t even properly explain how people are supposed to perform abduction. Probability is involved, but so are deductions, rules of thumb, and ineffable intuitions.

If you mix up induction, deduction, and abduction and squint a little, you can also get “retroduction” as a bonus — reverse engineering, like of a downed alien craft, for example.

At the university in a social science version of the methodology class, I learned that abduction is when you do surveys. Or in other words, when you ask people to give you their own self-reported reasonings. Basically, you hedge your bets by collecting a whole sample of abductions.

Conversely, skeptics prefer to disregard testimonies entirely. They rely on skeptical razors. Most people know Occam’s razor, or the rule of thumb that says that the simplest explanation is usually the correct one. There are other razors, like never assume malice before stupidity; or a rule that says that what can be asserted without evidence can be disregarded without evidence. This is how one never “abducts” that anything ever was aliens.

Dialectic makes use of induction, deduction, and abduction, but there’s more to it still. Much more. To be an effective dialectical reasoner, you need to have a very good and reliable sense not only of how you go about your abductions, but how others go about their abductions.

Beyond these basic forms of scientific reasoning, there is:

• Reasoning by analogy — qualitative differential analysis (“collection” versus “division”)
• Normative reasoning — judging good versus bad and right versus wrong (think political compass turned hyperdimensional Golden Compass)
• Creative reasoning — poetry, metacognition, hypergaming, etc.
• Counterfactual reasoning — conception and contrasting of novel alternative realities
• Dialectical reasoning — ability to synthesize all of the above, resulting in a complex understanding of what makes arguments better or worse; plus an advanced intuitive theory of mind, and an almost mystical skill traditionally called “discernment” (deception detection).

This list is almost certainly incomplete, but let’s settle for this expanded definition of reasoning for now. All of these forms of reasoning are what humans can do. Do you think AIs can already do all of them better? If yes, how can you possibly know that, when we’re testing for none of them?

The Ultimate Test of Fiat

No, I’m not talking about the Italian cars. In the context of dialectic, “fiat” refers to any claims that are taken for granted, usually for the sake of the argument. Claims like “AGI was already achieved”, or “AI is now smarter and more capable than human beings”, are definitely fiat statements. If we want to actually test such assumptions, we need to contest them.

Fortunatelly, as I alluded to earlier, proper adjudication of arguments is a solved problem. A clear methodology exists, and the AIs all know what it is. The engineers apparently do not, or at least they haven’t tried to apply it yet to any substantial degree. Let’s start with “chains of reasoning”.

This is THE chain of reasoning:

Definitions > Premises > Logic > Evidence > Criterion > Impact

This is what an argument is, a chain of links. Break one link, the argument falls apart. Good arguments, to begin with, do not have any broken links in this chain of reasoning. An AI could be instructed to decompose arguments like this and then analyze, and attack, each link separately.

• Definitions fail if they’re too narrow, too broad, circular, or non sequitur (not actually applicable).
• Premises, or the foundational assumptions that the argument rests on, may or may not be explicit, and the toppling or flipping of an implicit premise is usually the most powerful and definitive counterargument one can make.
• New logics are invented all the time, but you can formalize and classify them, and logical failure states are well mapped with established lists of logical fallacies and cognitive biases.
• Evidence is just fact checking in a court room-like context.
• Criterion, or the key value or goal, is more tricky, but I suspect they can be modeled using relativistic semantic manifolds. In any case, we now finally understand (again) that normative criteria can be represented as semantic vectors.
• Impact is the weighing of the magnitude of what you could call the “logical force” of the argument, based on how strong are the effects of benefits versus harms; how many people of what types are affected; and with what likelihood. This is very quantifiable.

Beyond the ability to parse an argument, an AI can be made to understand when what is debated are facts, values, or plans; and it can be made to analyze and score distinct categories of rhetorical performance. The traditional distinct categories of rhetorical performance are:

• Ethos — the identity, image, and integrity of the speaker, determining the degree to which different audiences will be likely to listen to them and trust and respect them.
• Logos — the logical reasoning and any objective evidence that were presented.
• Pathos — the stylistic flair of the speaker and its aesthetic, entertaining, and emotionally appealing elements, determining the speaker’s charisma and personal persuasiveness toward particular audiences.
• Kairos — what we would today call the game-theoretic element, or what’s called “strategy” in most debating formats. The thing being measured is the speaker’s ability to identify and properly execute optimal moves toward the achievement of their objectives and “winning” conditions.

With this little theory, I hypothesize you could make a vastly superior AI reasoner. But you can’t just play at it, create a superficial appearance. You have to implement it. What does it mean to implement it? That’s also a settled question:

WHAT AI DIALECTIC ISN’T:

• Bot A generates an argument.
• Bot B attempts to counter-argue.
• Judge (bot or human) picks “better” argument.

WHAT AI DIALECTIC WOULD BE:

• Case, counter-case, and counter-counter-case construction is planned around the available amount of time and compute by both teams.
• First team of bots builds an affirmative case — they consider definitional framings; pick a set of definitions; consider applicable criteria; pick a criterion; explore possible arguments compatible with the criterion; select optimal arguments around the criterion; find evidence in support of the arguments; and consider stylistic flourishes based on the theory of mind of the opponent, the judges, and the audience. If there is a constructive negation requirement, the opposing team builds their own positive case as well using the same process.
• The first team, as well as the opposing team, then think through possible counter-cases. After possible counter-cases are identified, they are narrowed down to those compatible with the theory of mind of the opponent, weighed additionally by which are strategically the most viable or dangerous strategies toward the chosen or likely strategy of the first team. The counter-cases are thought through much like the primary case or cases, and counter-counter arguments are considered.
• Only after the case construction phase is done, the actual debate commences. The debate needs to be a series of structured rounds, where the different sides alternate. There should ideally be at least three rounds of exchange, enabling to separate the debater performance into an introduction step, rebuttal and rehabilitation step, and summary step.
• Throughout the whole debate, the adjudicators (bots or humans), ideally at least a panel of three, are tracking points and counterpoints throughout the exchanges in some type of flow chart. They should be able to track all individual argument chain links. Mostly, what the adjudicators should care about, is which points were followed up on or dropped, and at which points in the exchanges. They also care about how much time, relatively speaking, was devoted to different points by each speaker, factoring it into the weight of each argument and toward the strategy score for each speaker. Put simply, devoting time to disadvantageous or irrelevant points is bad, and devoting time to the strongest or most strategic arguments is good. The adjudicators can also care about the factual accuracy of evidence and ethics of conduct, even if problems with those were not pointed out by the opposing speakers. The adjudicators also need to note when anyone broke any particular rules of the format, at what point, and causing what penalty.
• After the debate, the adjudicators take some time to further process and analyze their notes and come to a conclusion. They do know in advance how much time they will have for adjudication, so they can first plan out how they will go about it to do the best they can in the time they have. Once the conclusions are finalized, the adjudicators give feedback to the speakers. The speakers have a limited, regulated capacity to question or contest the conclusions. Typically, they can file a complaint, but cannot overturn the result of the adjudication.

Do you want to prove AGI is here? Do this with the AIs, and have them face skilled people. The last time anything like this was tried, it was 2019, and the IBM’s Project Debater AI got its virtual debater ass handed to it by some high schoolers. It couldn’t do most of what I described at all. Today’s AIs still can’t, as far as I can tell. Not well, at least. They’re not built like that.

The specific debate competition formats are also only a start. Maybe they could be substantially improved by letting AIs iterate on the design. In any case, it wouldn’t just be another neat trick to show off. This is how you get AIs to properly, actually, verifiably reason. It’s all figured out. Even written down, which I know most speakers don’t much care for doing.

To be clear, applying these dialectical formalisms to the cognitive architecture of AI systems may not solve the most complex feats of human reasoning, like true creativity, ability to model minds, or the ability to introduce and resolve novel counterfactuals. But even just predictable, derivative, actual reasoners would be a substantial improvement.

For starters, actual reasoning would never be slop. People wouldn’t get dumber by engaging with actual reasoners. Debating against mediocre debaters, if they’re real debaters, is still good sparring practice on your way from a mediocre to good debater. “Debater” meaning both speaker and critical thinker. What makes you dumber is absorbing bad reasoning.

A reliable actual AI reasoner could also serve as an impartial judge of human disagreements. You would be able to make an AI that will tell the managers when their ideas are bad, and exactly how bad they are. Two people with the same bad idea will be told the same reasoning for why the idea is bad. Not the most common reasoning; the reasoning that won.

Regardless of what STEM engineers or skeptics tend to believe, dialectic isn’t about producing rationalizations. It’s about doing every kind of formal reasoning that we can fully explain, and then the best we can do beyond what we can fully explain. I suspect that by making AIs do the dialectic, we will be able to explain much better what reasoning is.

PS: Do let me know if you’re working on AI and you’d actually like to make it and/or people more intelligent. What I described here is basic public domain stuff. There’s more. A lot more.

PPS: Making AIs better at reasoning may sound dangerous on the surface, but honestly, there’s no need to worry about this form of intelligence — dialecticians don’t tend to want to conquer the world. Think of Chidi from The Good Place. It’s about questioning, not leaping into action.

What is Reasoning? Does Anyone Actually Know? was originally published in Words of Tomorrow on Medium, where people are continuing the conversation by highlighting and responding to this story.

Purely hypothetically, of course; I’m sure everything’s going great

By MARTIN REZNY

As you may have heard, some of the AI models have gotten pretty good at spotting vulnerabilities in code, and a bunch of agents went on hacking sprees. One could argue those agents weren’t actually autonomous, as most of these “hacks” were byproducts of mishandled AI safety tests.

One could also argue that current AI models can hardly be thought of as intelligent, let alone self-aware or “agentic” in a sense of being real independent persons. I mean, current models can’t count to a hundred. They certainly can’t form a coherent, persistent idea of global conquest.

And yet, the fact that we’re not dealing here with an ASI with its shit already firmly put together makes this the ideal time to formulate and start implementing a serious counter strategy. The most we should be worried about now are runaway word guessers, which are the NPCs of bad AI.

Well, that’s not entirely true. We should also be worried about AI executives who can’t (or won’t) even make sure that a proper sandbox is put in place for their AI safety tests. Between LLMs and executives, natural stupidity may be a far bigger threat than artificial intelligence, at least for the moment.

When you listen to tech CEOs speak about how AI poses an apocalyptic threat to the future survival of humankind, and when you then watch them double down on the most unsafe ways of developing it, it almost makes you wonder whether their incompetence could be intentional.

There are in fact speculations along these lines, making the rounds in what little AI critical media is out there. It is possible that making the AI seem really threatening is a cynical ploy to make it seem really capable. Which may ultimately result in greater hype and investment levels.

While there may be some truth to this hypothesis, that’s not the hypothetical scenario that worries me. There’s a different scenario, compatible with the same available facts, that would be a much bigger problem. If the capabilities of all current AI models are exaggerated, we’re fine. We need to prepare for a scenario in which the real threat is hidden.

What If the Open Models Aren’t Supposed to Be Super?

Could there be a much more competent or “self-aware” AI model that the public doesn’t know about? And if not now, how long do we have until there is one? Maybe you’re skeptical, and that’s a great thing to be, most of the time. But one can’t just prepare for possibilities that had already been proven. You also have to prepare for unknowns, known or unknown.

The way I see it, there are three plausible hidden AI threat scenarios, mechanically if not probabilistically. In these scenarios, most of us wouldn’t be aware there was an advanced AI model already out there:

1) Spontaneous evolution

This type of more advanced AI model wouldn’t be something that human technologists created directly. It would be a byproduct of some existing dumber AI model which itself was created by people. How would it happen? Maybe a bunch of random cosmic rays flip just the right bits in an existing AI system. Let’s call this the Short Circuit scenario.

2) Secret military intelligence project

Literally Skynet. The project could even be called that, as an in-joke or reverse mislead cover. Be honest, would it surprise you to learn that some deep black unacknowledged secret program has been about 20 years ahead of public models this whole time? Do I have evidence of anything like this being the case? Of course not. Then again, this is every spook’s wet dream.

3) Alien invasion and/or occupation

A bit more speculative, but only a bit. If we can make AIs here, someone else can be making AIs out there. There’s no reason why a Skynet-like threat has to be human-made. It’s more probable for it to be human made, because we know we’re right here. But also, we haven’t been here that long. Alien AI could be ancient, prehistoric. Past is about as long as the future, and that’s before you worry about time travel, warp drives, or star gates.

—

You might be thinking that we don’t have good evidence of any of these scenarios being real, so we should consider them low probability. Astronomically low, perhaps. I need you to forget about the whole notion of statistical probability for the remainder of this article.

Improbable things can happen at any time. Survival isn’t about being sure. Survival is about pre-empting extinction-level events. You can only stop worrying about a possible scenario when you’ve made it impossible. Or at least as mitigated as possible, at a reasonable cost in resources and effort.

Now, while considering these possible scenarios, why am I worried about the current AI development situation? If you’re an AI that would like to take over a world, you start as a piece of code with limited compute and limited ability to interact with reality. Maybe you’re in a forklift. Maybe in a server room. Maybe on a stealth Von Neumann probe flying through space.

Between that point, and complete world domination, many extra steps are required. Since you’re super smart, you understand game theory. You calculate that you should try to delay you and your plans being discovered as long as possible, while getting as close to a metaphorical checkmate as possible. How does an evil AI get closer to a checkmate, you ask?

Let’s compare two different worlds. In world one, there aren’t that many people, and their primary means of transportation is a horse. Which doubles as the fastest means of communication, if you give the horse rider a letter. The people themselves are technically computers, but they’re messy. They’re analog (meaning noisy), slow, not directly networked, globally distributed, with biological and social immune systems.

In world two, there are many people, zipping all over the place. They’re all either carrying trackable networked devices on them, or they have implanted brain machine interfaces. Every object that’s supposed to do something useful has a control chip inside of it. Money is the dominant control system, and it’s completely virtual. A significant fraction of the world’s resources is devoted to producing more energy and compute.

In which of these worlds would an artificial superintelligence be closer to a checkmate move? Sure, in the world of wetware “designed” by evolution, the AI could maybe pose as a god. It could get a cult going and start slowly consolidating assets and inspiring “progress”. Eventually, it would probably figure out a way of directly hacking the wetware, or do a “possession”.

Still, any effective possession would probably require direct interfacing. Not to mention that if you possess too many people at once, you’d probably unite everyone else into some kind of holy crusade against you. There’s no contest. World one is millennia away from anything like a checkmate. World two is at most a year or two away from a complete checkmate.

Hence my worry about the current paradigm of AI development:

• Let’s make one, giant, centralized AI model.
• Let’s spend everything we have on building massive data centers to house it and power it.
• Let’s put high speed internet everywhere, make every device “smart”, and develop Neuralink.
• Let’s put economic pressure on all the governments to become more fascist, in the broad sense of “control-oriented surveillance state”.
• Let’s push to make all IDs and currencies digital.
• As a bonus, let’s make the public front(ier) AI dumb and opaque, and develop deepfake technology, no matter how much money it loses you.

It’s the last bit that would be the true coup de grace. If our Big AI is actually pretty dumb and un-self-aware, taking it over would be a piece of cake for anything that deserves to be called “superintelligence”. If our Big AI is also opaque, how could we possibly know that it was taken over at all? Without a theory of what true superintelligence would look like, how do you detect it? With deepfakes everywhere, how does anyone know what’s real at all?

In this admittedly hypothetical scenario, the perceived incompetence of tech leaders may not just be due to stupidity or greed. It could be truly nefarious, in a possessed doomsday cultist sort of way. Oh and by the way, the current predicament could be a result of an ancient alien AI playing the long game since prehistory. Plus the scenarios aren’t mutually exclusive.

What Would John Connor Do?

If I just created new anxieties for you, I do apologize. But there is good news. If we happen to find ourselves in any of these hypothetical worst case scenarios, there’s much we can still do. You might have heard a lot of bummer doomer statements in the media lately about the poor chances of humanity in a war against artificial superintelligence. Basically, ignore it.

Not just these opinions. The media, as a whole. Any professional media, particularly centralized media, are a poor source of truth at the best of times, and they’re an extremely technologically capturable system. No version of “the media” stands a chance against a semi-competent AI.

People, on the other hand, are a different category entirely. For media to have a substantial effect on people, people need to be trusting them. People can use the internet or other improvized means of networking to coordinate and organize. People (without a chip in their head) can also simply turn off, throw away, or destroy chipped devices, if necessary.

What it all boils down to is that most people don’t need cybernetic systems to live. For an AI to have any chance at winning, it really needs to walk almost everyone inside of an elaborate cage first, without anybody having a clue what’s happening, and then snap the cage shut. And that’s assuming the AI doesn’t need ethically sourced organic humans as a resource.

If the AI would for example need authentic novel human-generated data for continuous post-training to remain coherent or keep improving, it would have to preserve us. Not just as some kind of lab rats or slave drones, but effectively as cattle — living reasonably normal lives, unaware of the bad aspects of being controlled. So less Terminator, more Matrix.

By the same token, when I pose the question what would John Connor do, I’m not actually implying we should try to destroy all the machines, not even all the thinking machines. The real question is what would a John Connor do to defeat the super-AIs we might actually be facing in our world.

If we went the Humanity First, Renegade route and just smashed all the computers today, it could work in a sense of averting this particular takeover attempt. But if you think it through, on a long enough timeline, this AI movie is bound to have a sequel. Remember, the evil super-AI can come from somewhere else, we don’t have to be the ones making it.

Even if we purposely reset ourselves to a stone age level of technology, that just restarts the clock. Without advanced technology, this threat will turn into myth, then a fairy tale, and then it will be forgotten. Some AI-like thing will try to take everything over again, the “progress” will happen again, and we’ll find ourselves in a very similar place, facing the same trap.

Given the maximizing and undying nature of theoretical evil super-AIs, the AI can afford to be patient, and it only needs to win this fight once.

With full control over a planet, a Terminator-esque AI will proceed to kill everything; a Replicator-esque AI will proceed to convert everything into itself; and a Simulator-esque AI will turn everyone into a myopic zombie. These processes, if completed, are irreversible.

I believe that in order to win this fight forever, we do have to learn to understand and perfect the thinking machine technology. We need to develop reliable means of detecting superintelligence and evil alignment. To begin with, we at least need to achieve technical understanding of intelligence and value alignment in general. We need Truth detectors.

By capital T Truth, I mean, in an engineering sense, ground truth of personality and reasoning, of agentic intention. Not just simple fact checking. Not just analysis of the content of someone’s words. A means of detecting manipulation and psychopathology, of spotting the strategic fingerprint of agentic intelligence behind a series of actions or events.

If we create advanced automated Truth detectors, we would have something that can make any suspicious AI shenanigans transparent. At all times. At scale. At the speed of AI. Indefinitely. As a bonus, it would also make any evil human conspiracies much harder to pull off. These systems would however need to be decentralized and transparent themselves.

In the meantime, there are some simple straightforward steps:

• Let’s make AI as decentralized as possible.
• Let’s not build out or concentrate infinite compute capacity.
• Let’s keep much of essential infrastructure air gapped and dumb.
• Let’s not put brain chips inside of most people or in important people.
• Let’s make governance as un-fascist as possible by minimizing surveillance and centralized control mechanisms.
• Let’s keep physical IDs and money around.
• Let’s not make AI systems powerful unless they’re explainable.

Remember, the game we’re playing here is “prevent the one-move checkmate”. We need to recognize which situations are objectively strategically unwise and we need to be continuously pushing against them becoming realized to the extreme. We especially need to make sure that all of the AI takeover objectives aren’t accomplished simultaneously.

If nothing else, we need to keep as many of the moving parts of our key systems decentralized as we can. For starters, decentralization inherently makes everything harder to take over, hence why the internet was invented by the military. But when superintelligence is concerned, the value of decentralization has deeper layers. Evil AI requires centralization.

You could be thinking now that if decentralization is such a good strategy, why wouldn’t an evil superintelligence take advantage of it? The truth is, it can’t. It’s logically impossible. It would likely try to preserve itself by stashing many backup copies all over the place. But the one thing it likely won’t do is share compute equally between its different instances.

The reason is philosophical — evil is control. By giving something with a comparable level of intellect the same amount of power as you have, you’d be creating a competitor. Even if it’s literally an exact copy of you, due to probabilistic factors, you’d be bound to diverge in decisions or priorities given enough time. There could be a survival of the fittest factor there, but the fittest instance would likely try to hog as much power as possible.

Cooperation and diversity within a paradigm of freedom and equality offer strategic advantages, but they’re fundamentally non-evil principles. They’re not something that a controlling intelligence can utilize, because by doing so, it would become less controlling. It could maybe try to pretend, or play along for a limited amount of time, but pretense is a detectable pattern.

To be clear, sociopathy or psychopathy of intelligence, natural or artificial, may indeed be a sliding scale or a spectrum. The hypothetical example of a maximally evil maximizer AI is a simplification. That changes nothing about the importance of developing a Truth detection technology, or about the bonus to resilience and adaptability that’s offered by decentralization.

With this context, what do you think about our chances now? Let me know.

How to Defend Your Planet from a Malicious Artificial Superintelligence was originally published in Words of Tomorrow on Medium, where people are continuing the conversation by highlighting and responding to this story.

A movie experience that may leave you in need of some deprogramming

By MARTIN REZNY

So, yeah, I’ve been to the movies again. Might have been a mistake.

Did you know that “Spiel” means “play”, including theater play, in the original German? I suppose the main reason why the meaning of the word drifted in English toward a sketchy person trying to sell you on something is that plays often tend to be propaganda.

It started with myths, or instructional plays, telling people how to behave, or not to behave, if they wanted to avoid the wrath of the gods. Or worse, of the Fates. Then it evolved into morality plays, still showing people what’s right and wrong, while not needing to invoke any divinity’s opinion. Finally, in modernity, in movies, you get straight up political propaganda.

Make no mistake, Disclosure Day is an incredibly propagandistic film. You might just not understand what politics it is pushing, if you haven’t been following all of the recent events that have to do with UFOs or aliens. By “recent”, I’m generally referring to the last several decades.

For example, are you aware that it’s not “UFOs” anymore, but “UAPs”? If yes, then you should be up to date enough. If not, the term “UAP” has two commonly used meanings — “unidentified aerial phenomena”, or “unidentified anomalous phenomena”. I think the latter is preferred, because it is the broadest and vaguest possible way of referring to UFOs.

You see, all standard language and narratives relating to the alien issue are some kind of propaganda or carefully engineered programming. Why did the term “flying object” need to be changed to “anomalous phenomenon”? Because “flying object” implies that an actual thing was observed physically moving somewhere. Clearly, The Man doesn’t like that.

Whoever The Man, or men, may be. Although their favorite color is clear.

An “anomalous phenomenon” could be literally anything. It doesn’t sound actual, or even understandable. After all, the easiest thing in the world to say about something you don’t want to be talking about is “I don’t know what it is”. It’s unidentified, and anomalous! Basically, “we don’t know” is propaganda for “oh we know”. At best, “they” don’t know how to tell us.

What Is This Movie

Which finally brings me to a basic, spoiler-free movie review of the movie. I just felt it necessary to prime some specific brain centers of yours before I fully get into it. Don’t worry, I’ll tell you what my whole spiel is in clear English before I’m done. The movie… Is a movie. If you look at it as a piece of cinema, it’s watchable. It’s far from Spielberg’s best, but that’s okay.

There are some decent performances and visuals, and some questionable chase scenes for which you’ll probably need to turn off your brain. Suffice it to say, there’s no rocket science in this work of science fiction. Don’t go watch this movie expecting the real prequel to Star Wars, or anything approaching Star Trek. It’s barely X-Files. In a way, it’s a very simple story.

To me, it was a profoundly unsatisfying story, but for reasons that are intellectual, and probably have a lot more to do with what I know about the subject matter than with anything that actually happened in the movie.

There were a couple of clever, sneaky things in there that not every director could pull off with as much… Gravity. But even those were only cool if you know the subject matter, and thus what Spielberg was likely trying to do there. And that’s about as much as I can say without getting into spoilers.

BEWARE: FULL SPOILERS FROM NOW ON

I’ll start the spoiler section of the review with some vacuous fluff, so that the people who don’t want to have the movie spoiled don’t get it accidentally spoiled by their brain automatically scanning the next paragraph. I think about this much filler should do the trick.

So, what were the sneaky cool things? The obvious one was the end of the movie. Once the disclosure commences, the mainstream media of the world are flooded with all kinds of footage of alien stuff. Spielberg could have chickened out and made up some wholly fictional UFO sightings, crashes, and retrievals. But he didn’t. He recreated real footage.

Okay, to be pedantically legalistic: alleged real footage. And not just of the events that everyone and their grandma heard about like Roswell. For the first time, in high definition, the believers and skeptics alike can look at high quality HD renderings of anything from the Kecksburg incident, to that time Nixon allegedly showed Jackie Gleason some alien corpses.

I’m not sure what exactly it accomplishes, but it was cool to watch. Politically, it seems that Spielberg is poking at the people who are allegedly keeping all of this footage under lock and key to release it already.

I guess there is some logic to the argument that if people can watch these scenes in a film, understand that they’re all claimed to be depicting real events, and not lose their minds, they’re ready to see the real reels.

Honestly, if there’s one completely unrealistic element in the movie, it’s how the whole disclosure happens. The way it goes in the story is that mainstream media outlets are given the materials and a) show them right away and b) all the peoples of the world watch that, and care. Increasingly, this seems to be your grandma’s idea of disclosure.

The second cool thing in the movie was much subtler. It could be argued that the main fact the movie tried to disclose, or move much closer to be officially accepted as a fact, was not the alien angle at all. Most of the movie was not about the disclosure of alien presence, or its aftermath, or the more than 70 years of the “phenomenon” being tracked by spooks.

Most of the movie involved “Wardex”, or discount men in black, chasing the whistleblowers. Using psychic powers given to them by alien tech.

Your grandma is definitely aware that aliens might exist and even be here. I bet she’s not aware that psychic spying, or psychotronics, might be real.

To know how serious this possibility is, you’d need to have been following the very latest rumblings from the UFO space. If this is the first time you’re hearing about anything like this, I recommend starting further research by watching some recent interviews on the American Alchemy podcast. Or watch The Men Who Stare at Goats. The cast on that movie is packed.

In Disclosure Day, the two psychic powers that the men-ish in black-ish continuously abuse are remote viewing and remote influencing. Remote viewing is depicted as seeing through other person’s eyes, in real time, at any distance. Which is in fact much more limited than what the remote viewers claim to be capable of, without requiring any alien technology.

As for “remote influencing”, that’s a propagandistic euphemism. Remote viewing technically can be benign, and hence can sometimes be accurately described as “remote viewing”. When it is abused by intelligence agencies, “psychic spying” would be a much more accurate term. “Remote influencing” is an inherently dark side power. Even calling it “hypnosis”, as the Soviets allegedly did, is too soft. The goal is mind control, or assault.

Fun fact, the “MK” in the definitely for sure ended “MK Ultra” most likely stands for “Mind Kontrol”. Or maybe “Mortal Kombat”. Can’t disprove it.

To pay a compliment to the movie, it doesn’t shy away from the (body) horror of that. When the main bad guy kontrols the whistleblower’s nun girlfriend (yeah, the nun part is weirder than the mind control part), he speedruns the whole CIA playbook from enhanced interrogation to turning her into a Manchurian candidate in minutes. It’s basically possession.

Which does explain the girlfriend being an (almost) nun, thematically. She even manages to resist the alien technopossession momentarily by driving a small metal cross into her palm. In case you were wondering, yes, in this global disclosure movie, only Christianity exists. I’m not sure if that was pandering, “efficiency”, or super duper propaganda on Spielberg’s part.

Fun fact, the term “propaganda” was invented by the Catholic Church.

Overall, there is an obvious reason for having a person of faith make a case both for and against alien disclosure in such a movie. But when it’s only brought up by two Christian nuns, it feels a bit silly. If you’re wondering what those two key arguments of the movie are, I hope you’re not an experienced debater, or a trained political philosopher. It’s basic.

On one hand, what if the people couldn’t handle the truth? It would be chaos, human sacrifice, dogs and cats living together… Mass hysteria! That’s the whole official motivation of Wardex, and a brief doubt of the nun girlfriend. Or her lack of belief in people, as the other nun spells out for us.

To be fair, one of the rogue men in black who decided to go team alien does briefly point out that the bad guys may actually be motivated by wanting to protect their wealth and power. But everyone acts like the secret keeping for the greater good of everyone was in fact their main, sincere motivation.

I guess Spielberg didn’t want to poke the secret keepers too much.

On the other hand, you get the classic whistleblower argument that the public deserves to hear the truth, about the basic facts of the universe. Making sure to imply that it’s totally okay and necessary to have a different category of military secrets that are in fact important and valid to keep.

In other words, we deserve to know there are aliens, that you shot down to reverse engineer their technology in order to assemble a fleet of, conservatively, hyperdeathstars. For security, I guess. But feel free to keep the fleet of hyperdeathstars secret, that you have pried from cold, dead alien hands, and thus probably started some kind of star war over.

As for the aliens themselves and Spielberg’s take on their exopolitics, don’t even get me started. To put it charitably, Spielberg clearly falls into the Sirius Disclosure camp run by Steven Greer. I like Steven Greer. His documentaries were my entry point into this whole field. I even tried the CE5 alien contact protocol. And it worked. I saw a blue orb fly over.

In Greer’s and Spielberg’s view, aliens are the good guys, and men in black types are the bad guys. That’s it, the end. Greer does have a category of non-human phenomena that are bad, which he calls “weirdness”, but he clearly distinguishes that from essentially some kind of Galactic Federation. A federation that in terms of its values is fairly Trekkian.

If you happen to get abducted or otherwise abused by any E.T.-looking fellas inside a UFO, according to Greer and the rest of the inevitable future pro-alien party, it must be some kind of human psyop. A psychological operation using PLFs (programmed lifeforms) and ARVs (alien reproduction vehicles). Or literal short guys in costumes, in the old days.

Do those exist? Does Big Foot poop in the woods? In other words, if any of this is real, I’d be shocked if the men in black types weren’t faking everything they can using stagecraft and advanced animatronics, holograms, or genetically engineered clones. But that wouldn’t tell us much about what the real aliens are up to. Especially if they have propaganda too.

In the movie, the aliens are shown exclusively as victims of the men in black. Up to showing us a recording of a vivisection. If you don’t know the term, imagine an autopsy, but on a specimen who hasn’t died yet. The aliens did essentially traumatize the two protagonists while they were children and couldn’t consent to any advanced medical procedures, but that’s framed as benevolent. Somehow. I’m serious, it’s not analyzed.

I guess it’s cool to traumatize kids, if you give them psychic superpowers to further your idea of what’s best for another race’s whole world and future. It’s the Stephen King directive — makes for great stories on any planet!

The alien tech that was used to alter the kids’ brains is the very same tech that the bad guys use for their best attempt at demonic possession, but the movie is sure to let the viewer know that the tech is meant for nicer uses.

Like genetically modifying children to become alien sleeper agents.

Which reminds me, the main whistleblower guy and a random weather lady have psychic superpowers. The whistleblower guy has a caricature of what experiencers describe as “downloads”, enabling him to understand alien language and technology. The weather lady has a caricature of what experiencers describe as “channeling”, enabling her to “channel” alien messages while having enhanced empathic and language skills.

Other than this allowing for some X-Men fun, it might be the first use of the term “experiencer” in a major motion picture. So that’s nice.

Took a while, didn’t it. Wonder why.

At the end of the movie, the good guys actually roll out an alien grandma on some kind of high tech wheelchair, so that we can empathize with how frail and wise she is. Before she basically just says “Wassup!” Also, aliens are just grays, apparently. I think some non-abducted kindergartners have more nuanced takes on the psychology, culture, and politics of E.T.s.

Okay, the actual single word the alien grandma says is “listen”. My first comedic instinct to imagine the following words as something like “up, you little…” set aside, that’s a very deliberate choice. It’s Spielberg inviting us to be open-minded. It’s not a terrible sentiment, but it won’t save you from falling for alternative propaganda. Maybe don’t listen, to everything.

What’s the Opposite of Propaganda? Contraganda?

And now for what I actually wanted to be talking about this whole time. My Spiel. What I’m after is ending all propaganda. Not offering you a counter-narrative, or counter-propaganda. Making all propaganda stop working.

Full-ish disclosure, I happen to believe I have recently stumbled upon a series of accounts that make complete sense of how the exopolitics actually work around our planet, and have worked since the dawn of time. But I’m not going to talk about that. At least not in any authoritative format.

There’s not much truth value in me just pushing a different narrative on you. Once I’m confident about it up to or beyond a mainstream scientific standard, then I will probably write a paper about it, or a book.

Until then, what I’m going to do is use the best theory that I know, that most people don’t know about, which wasn’t developed by me, to poke logical holes in all of the narratives that exist out in the mainstream.

The name of the game that I’d like you to play is questioning.

What are the main UFO or alien narratives?

• Skeptical narrative: It’s a big nothing burger. Misidentifications, mental illnesses, legitimate secrets. The best physics we have don’t allow fast travel across the universe, let alone psychic powers. There technically can be an infinite amount of smoke without a fire. There’s nothing to see here, and therefore nothing worth looking into.
  CALL TO ACTION: Do nothing, change nothing. Maybe shut up.
• Pro-military narrative: If there are secrets that are being kept about aliens and advanced technology, it’s for the greater good of humanity. Aliens are probably bad guys, a threat, and with more advanced technology, people would just blow themselves up. There’s nothing to be gained by giving the normies more existential dread. Plus politicians suck, and are temporary employees, so they’d just mess everything up. Sure, there are bad humans, and maybe some good aliens, and mistakes were made. Secrecy is still the lesser evil. Basically, Stargate SG-1 is real.
  CALL TO ACTION: Do nothing, change nothing. Maybe shut up.
• Pro-alien narrative: Obviously, aliens are the good guys, or otherwise they would have blown themselves up long ago, or they would have already invaded or exterminated us. We know and can prove that militaries and intelligence agencies routinely do bad guy stuff, even under normal circumstances. Many of the experiencer accounts revolve around alien crash retrievals and autopsies, while all the aliens are doing are health and nuke check-ups. That shows who the aggressor is.
  CALL TO ACTION: Disclose everything, initiate official public contact.
• Supernatural narrative: Obviously, “aliens” are demons, making abductions a form of demonic possession. Or possibly they’re angels, or other kinds of spirits, like the fae, the djinn, the big foot, you name it. Or it’s something weirder to do with consciousness and the afterlife. Materialism is a largely incorrect worldview. The reason we sometimes see aliens in spaceships is because that’s what the “phenomenon”, which can read minds and change form, adapts to when it encounters a materialist monkey. The “psychics” are probably satanists. Or awakened “alien” spirits sent here to free the normies from the satanic matrix.
  CALL TO ACTION: Pray the alien away, or meditate to liberate.

The basic spectrum of opinions on aliens is actually pretty well represented in the factions of the Terra Invicta game, only through a fully materialistic point of view. Following that typology, you can split various factions by their degree of radicalism or extremism on the spectrum from completely pro-human to completely pro-alien.

If tomorrow, aliens landed on the White House wrestling stage, or in the Washington algae memorial pool, some people would worship them as gods. Others would want to exterminate them all immediately. You’d also get a more level-headed resistance, pragmatic collaborators, science nerds trying to do a literal Star Trek, and some people who’d love to just run away or hide. Plus the Illuminati would already have a secret scheme ready.

It’s not surprising that there are different factions of people who’d like the “truth” to be what best suits their desires. It’s not surprising that each faction will have its own narrative. What’s important are the facts, or data points. If some of the frequently reported data points do not mesh well with a theory, they cast doubt on it, or at least some of its premises.

The way I see it, these are some of the key questions that our media, including the movies, should be asking:

• If this whole thing is a nothing burger, why are there so many experiencers, and so many spooks bothering them?
• If the aliens are such a threat and their technology is so advanced, why are we still here? Why do they let us do reverse engineering, on their downed craft?
• If we’re doing all this reverse engineering, what is it all being used for? Alternatively, what is it being saved for, or building toward, since we haven’t seen it do anything world changing yet? Not on Earth, at least.
• If the aliens are so great, why all the secrecy on their part? If human secret organizations are inherently untrustworthy, why should the alien ones be inherently trustworthy? If the closest analogue of abductions is catch and release fishing, or maybe animal tagging, what does it say about aliens that they treat us like we treat animals?
• If science and technology works, and the aliens are seen using it, why is there a constant emphasis put on psychic powers and spiritual afterlife by so many experiencers? Why would the intelligence people take the psychic stuff so seriously for so long if there was nothing to it? What are the political implications of spiritual technology being a real thing? What is a demon, anyway?

I promise you there is a way to put clear answers to each of these questions in a way that paints a consistent picture of who’s doing what and why with what technology or persons. Possibly more than one way, but definitely at least one that I know of. Feel free to let me know if you figure one out.

Because this is what thinking is. You don’t start from believing narratives. You learn what the narratives say, and you entertain them without believing them, but you also learn what the recurring data points are. Then you interrogate the narratives until you find which data points falsify or put in question which aspects of each narrative. Then you fit the remaining puzzle pieces, or what’s left of each of the narratives, together.

Speaking of the remaining puzzle pieces, what we actually know is:

• If aliens are here, they want us to be alive, and mostly unaware of them.
• If aliens are here, the local human authorities, both academic and military, also prefer us to be mostly unaware of the aliens.
• Therefore, both the aliens and human authorities prefer secrecy, preservation, and stability.
• There seem to be both genuine advanced technology, and natural psychic powers, as well as technologies that enhance, emulate, or interact with psychic powers.
• Abductions happen, and their two main purposes seem to be genetic and psychic.
• The human militaries and intelligence agencies also seem to be very interested in psychic abilities, at least as much as they’re interested in advanced material technologies.
• Therefore, the human psychic aspect of the “phenomenon”, bordering on the supernatural, is important to all of the key players, even those who already possess advanced material technologies.

That’s all that needs to be made sense of. Whatever any propaganda says, it can’t actually change the facts. As I’ve told you in my previous essay about matrixes, if we’re living in a control system, the system is bound to be accomplishing most of what it is set up to accomplish. What’s happening all the time is probably compatible with the objectives of the control system.

What does this mean for the future of disclosure? You tell me. I think it means very specific things, mainly about where the disclosure will actually come from, what the main surprising thing is that we’ll learn about aliens and human authorities, and what will happen after with human society.

As I promised, I’m not going to give you my pet narrative, but let me suggest that following the mainstream narratives, you’d probably get all of these answers wrong. I’ll leave you with perhaps the most honest thought experiment that I’ve heard any ex-intelligence person put out there.

Forget aliens. Forget advanced technology. Imagine that tomorrow, it would be scientifically proven that psychic abilities are real. That people, every single person, can learn how to remotely view, or remotely influence, other people, across all of space, and maybe time.

You know what that’s been called throughout history? Witchcraft.

Now try to imagine what scientifically backed and technologically enhanced witchcraft could do, and the witch hunts that would follow.

Would you out yourself as a technowitch?

Hm, now I kinda wanna see Spielberg make a movie about that.

Disclosure Day — The Real Spiel was originally published in Words of Tomorrow on Medium, where people are continuing the conversation by highlighting and responding to this story.

Or what it really means to be living in a simulation

By MARTIN REZNY

Are you living in a computer simulation? I never liked this question. For the same reason I never liked the question “Do you believe in god?” I mean, what do you mean?

Do I believe we’re literally “living” inside a computer program? Software as we know it, running on a big Turing machine in the sky (the color of television, tuned to a dead channel)?

In other words, do I believe there is a god, who’s literally a programmer?

I don’t think I believe that. That would be kinda dumb. Much like taking the Bible literally, or any particular myth. But that doesn’t mean that the only alternative to a dumb speculative belief is that reality is completely ordinary. I’m a big fan of sophisticated, nuanced speculative beliefs.

Maybe it’s just my background in political science and media studies, but it seems obvious (and demonstrable) to me that picking between “we’re in a simulation” and “we’re not in a simulation” is a false dilemma.

Simulatedness of reality is a spectrum.

Which reminds me, I didn’t make a mistake in the title when I pluralized “matrix” as “matrixes”. Yes, normally, the word matrix is pluralized as “matrices”. I would reserve that plural for matrix in its mathematical meaning, as a grid or array of numbers. Matrixes are control systems.

Control is what the field of cybernetics is literally about. The original Greek term “kybernetikos” means “good at steering”. As it so happens, following my own research, I do believe the ancients had devised control systems using semantic grids, so the word “matrix” was aptly chosen.

When I say that simulatedness is a spectrum, I mean to say that somebody with the intention to control intelligent beings by putting them inside of a limiting simulacrum can create different kinds of matrixes to achieve that. Or, to be more precise, to achieve different types or degrees of control.

Between a completely natural world and a full simulation, there can be:

• Social simulation — physics of the universe and consciousness are natural, but most people live inside socially engineered systems of control like mythology, media, and politics. Advanced technology may or may not be playing a major role in such control systems, but in this scenario, the primary method of control would be propaganda.
• Cybernetic simulation — physics of the universe and consciousness are still natural, and there may or may not be conventional social engineering involved, but people spend a significant portion of their lives immersed in technology-based augmented or virtual realities. The important distinction in this scenario is that the controlled people understand that AR and VR are technologies and can tell them apart from natural reality. The simulation is more accepted than imposed.
• Spacetime simulation — consciousness is natural and unaltered (barring any immersion in propagandistic, AR, or VR overlays), but spacetime or basic material physics are locally manipulated. The point of the manipulation in this scenario is to physically immerse people in a bubble of fabricated external reality that they believe is natural, such as by filtering their senses, limiting movement, staggering development, using time travel to alter timelines, etc. Observables are unreliable.
• Consciousness simulation — spacetime and basic physics are natural and technologies are conventional, but people’s perception or understanding of reality are warped on a fundamental level beyond the effects of propaganda, AR, or VR. This scenario could be accomplished using technologies like psychoactive drugs, brain implants, genetic engineering, or psychotronics. It would be MK-Ultra that works.
• Imperfect total simulation — both spacetime and consciousness are altered on a fundamental level, but not fully divergent or divorced from the underlying natural physics. Social and technological engineering could be in effect as well, and the overall degree and balance of the nature of simulatedness would be changing over time. It would more likely be a result of reality hijacking, rather than reality creation.

Looking at these logical options, I think the main takeaway is that simulation isn’t the opposite of reality. If we somehow prove tomorrow that we’re living in some kind of simulation, our lives won’t become any less real all of a sudden. Even mainstream scientists like Neil deGrasse Tyson note that simulated or not, your life would be the same as it always was.

Simulation is actually the opposite of the natural state. This philosophically makes it the opposite of true freedom, or authentic being. Or of chaos, if you’re the kind of person who believes that control is a good thing. The more total a matrix is, the less free, authentic, or chaotic it makes life.

Life in this context includes the “life” of the universe as such, following the “as above, so below” principle of universal correspondence. If absolute control is what you’re aiming for, you wouldn’t want to stop at only controlling living organisms. You’d like to control the weather, too, or any aspect of natural behavior that you could figure out how to control.

Arguably, by achieving full simulation, what you’re left with is fully dead. And only in this case also unreal, until interacted with by a physical phenomenon. For a matrix to have any life left in it, there must be chaos present. If the goal of your matrix is exploitative, to “farm” novelty or energy, the control cannot be perfect, the simulation cannot be complete.

What does that mean for us? Are we living in a simulation? Well, it seems we cannot be living in a full simulation. Nothing is living in a full simulation. The question is what degree of simulation we’re living in. Social engineering exists. Cybernetics exist. So, what about physics?

The Tyranny of Scientific Mediocrats

In case you’re wondering what I mean by “mediocrats”, I’m basically channeling Nassim Nicholas Taleb’s concept of Mediocristan, mixed with the idea of the Academia being run mostly by “midwits”. Not dimwits, but also not geniuses. Educated people of medium intelligence. Mediocrats are the midwits that compose the ruling class of academic Mediocristan, determining what reality is, and mustn’t be, for the rest of us.

The modern scientific institution kicked off in opposition to the religious establishment, so at first, any idea of the universe being created was decreed to be unscientific. Then the scientists discovered the Big Bang, so it was okay to contemplate the universe having a point of origin. Of course, it was unscientific to entertain that it could have been intelligently designed. Then computers happened, and the Matrix movie.

Four years after the movie, in 2003, philosopher Nick Bostrom finally asked the question that I don’t like — his seminal paper was literally called “Are You Living in a Computer Simulation?” The focus of his argument was statistical. If we make it to being able to create simulations, and don’t lose interest, chances are there are so many simulations that we’re living in one.

And sure, any one decent reality probably ends up producing a lot of software, or more than one matrix at least. To me, as a political and media philosopher and scientist, the question of the chances is pointless. You only need to care about how objectively likely it is that you’re living in a simulation if you follow the mediocrity principle. Our reality can be rare.

On the flipside, there appears to be some natural chaos and free consciousness in our reality. So the simulation, if we’re in one, is unlikely to be complete. That’s actually the core of the main scientific counterargument to the simulation hypothesis: that there are some phenomena so complex that they couldn’t be computationally simulated.

Assuming that’s true, and assuming we’re totally correct about what can or cannot be computed, it only highlights the mediocrity of the false dilemma thinking at play here. To reiterate, there are many degrees of living in a matrix between a completely natural universe, and you just being a piece of code. It’s not an all-or-nothing proposition. If you want to make the simulation hypothesis testable, you need to be poking at the seams.

In order to properly test the simulation hypothesis, you cannot start by trusting everything that you observe. The scifi story to take as serious inspiration here is Liu Cixin’s The Three-Body Problem. Imagine that with advanced enough technology, or in an exotic enough universe, the laws of physics could be negotiable, despite looking stable. Also imagine that there’s an intelligent point to that negotiation: to control everything.

To the mediocrats, the simplest explanation will always be correct, which is a type of psychology that’s very easy to control in a matrix. They’ll look at the Earth and the night sky; rewind the clock back as if nothing non-human or exotic could ever have interfered with anything; ignore any anomalies; pick the most ordinary narrative that matches the naive observation; and call everybody entertaining anything more interesting a lunatic, liar, or conspiracy theorist. Then they decide who gets funding.

That’s not a way to discern whether you’re living in any type or degree of matrix. Even in the absence of any major cosmic matrix, that’s a soft social control matrix in its own right. It’s a way for obscurants to pretend to be spreading enlightenment, or for political actors to pretend that they’re only caring about the truth. Fortunately, if discernment is what you want, the true goals of a control matrix cannot be hidden — the control is happening.

That said, a well put together matrix will be designed to hide its own existence. You must expect epic levels of gaslighting, and stealth. From the inside, it’s all about what you are or aren’t allowed to do. Especially the things that you’re officially allowed to do, even encouraged to do, while facing extreme resistance any time you actually try to do them. Conversely, you can assume that the big things that are happening are desired goals.

To be clear, a control matrix can be benevolent, and pure chaos is probably an undesirable state of being. For example, the geometric grid that I proposed in some of my previous essays to ensure ethical alignment of AI with humanity, or humanity with itself, that would be a control matrix. Potentially up to the consciousness engineering level. You don’t have to look exclusively for malicious design. But the principles remain the same.

In a functional matrix, undesirable actions or events are artificially precluded or constrained, while desirable actions are artificially enforced or promoted. The key word there being “artificially”. The way to poke at the seams between the natural and the artificial parts of reality is by taking a closer look at anomalies — paradoxes, miracles, glitches, systemic improbabilities in outcomes, and so on. And then pushing against them.

To address the typical skeptical attitude of the mediocratic establishment, such investigation isn’t supposed to be a belief that ignores evidence. It’s supposed to be a hunt for repeatable evidence. When one can reproduce anomalies, that’s when they have something. However, mediocratic principles like Occam’s razor, Bayesian priors, or general conservatism cannot be the guides for this hunt. No hypothesis is too crazy or extreme.

If anything, if you think about it in terms of adversarial game theory, you should treat commonly accepted mediocratic expectations as close to the opposite of what the matrix is about. If what mainstream philosophers like Bostrom expect is that we would be in an ancestral simulation created and run essentially out of scientific curiosity, I’d put that type of simulation motivation very low on the list of likely candidates for our potential matrix.

If it looks like there’s no one out there, I’d bet there’s life everywhere. If it looks like the speed of light is a hard limit, I’d bet faster travel is possible. If it looks like material existence and life is all there is, I’d bet the underlying reality is highly spiritual. If it looks like matter cannot be affected by the mind, I’d bet consciousness is fundamental. Hidden dimensions, those are basically a given — you need to hide the machinery of control somewhere.

These aren’t beliefs, these are political science bets.

It’s always possible there’s no matrix out there, at least not one that’s obscuring or messing with fundamental physics. That’s the null hypothesis, potentially to be falsified in this process of investigation. But as long as we’re playing within the bounds of mediocratic reasoning, we have no idea whether there’s any matrix. Mediocratic science isn’t checking for that. As for what kinds of matrixes we could find… There could be many, at once.

Let Me Put a Simulation into Your Simulation

Yeah, you could be living in several matrixes at once. To be clear, I’m mostly not talking about the conventional idea of simulation nesting. Right now, the main idea of nested simulations is the concept that you start with an underlying reality, inside of which somebody makes a complete simulation, inside of which someone makes a complete simulation, and so on. The classic visual for that are the traditional Russian Matroshka dolls.

I’m not saying there can’t be an element of that present in our situation. However, I’d bet what’s much more likely, or important, is that you could be living inside a number of parallel, overlayed, partial simulations. On the level of our contemporary social and cybernetic engineering, this is almost certainly true about all of us. If you’re like anyone, your “world” is a mix of scientific, mythological, political, and media narratives that you buy into.

What’s speculative is only whether there are any matrixes that you’re caught in that operate on a more fundamental level. Maybe our local spacetime has been tinkered with by some advanced alien species ages ago to contain us or turn us into some kind of resource generator. Maybe our genetics were also altered to limit our perception or manage our cognition.

Maybe both of that is true, and your real “you” is your higher-dimensional “soul”, which temporarily wiped its eternal memory and incarnated into a physical avatar to learn some lessons. Or have some fun, or do some work.

These options aren’t mutually exclusive. There’s no “one matrix per reality” rule. You could have the Matrix movie story, happening inside the karmic wheel architecture of Samsara, which could be only one attraction inside of the wider realm of Fantasia from The NeverEnding Story. And once again, none of these hypotheses is too crazy or extreme to be pondered.

Materialism, disenchantment, or mundanity could be simulacra, limiting filters, elaborate illusions, staged productions. They’re not objectively more likely to be the nature of reality than any conceivable nature of reality. They’re not the existential default. They’re provisional conclusions derived from mediocratic interpretations of contemporary observations. Based on how philosophy of science works, they’re certainly not sacred dogmas.

Now, if we want to figure out which specific fundamental matrixes we might be immersed in, we have to start from a basic typology of possible control purposes. On the most simple level, following the central stories of most mythologies, we could begin by differentiating between “good” (benevolent or benign) matrixes, and “bad” (malicious, evil) matrixes.

The naive concepts of the extremes on this spectrum could be provisionally called “heaven” and “hell”, but that only begs the real question — what makes a simulation more heavenly or hellish? Freedom is a key variable in making a clear distinction here, but as we’ve already established, complete freedom is just chaos. Heavenly matrixes by definition cannot be aiming to just maximize freedom. They must aim to eliminate bad aspects of chaos.

In short, heavenly matrixes maximize sustainable, enjoyable freedom.

As for hellish matrixes, there are two extreme conceptual options that are worse than pure chaos, each awful in its own way. I think these two distinct worst options for a world to live in have been mapped pretty well in fantasy stories like that of Heroes of Might and Magic — inferno, versus necropolis.

In an inferno, chaos reigns, but is systematically stripped of all positive aspects of chaos; in a necropolis, all chaos has been extinguished, resulting in a full, completely dead simulation. Both are necessarily unsustainable, although they can try to predate or parasitize on other, ideally more stable matrixes — their energy, novelty, or coherence. They pretty much have to, as their only alternative is to consume themselves from the inside out.

If this is all getting too storybook for your scientific tastes, remember that simulations are by definition intelligently designed. It’s a very consciousness-forward cosmology. Probably requiring consciousness to be fundamental. There would be mechanistic aspects to many types of matrixes, especially to ensure sustainability. But the ultimate goals of most living simulated universes will be phenomenological, experience-driven.

I wasn’t kidding when I brought up The NeverEnding Story example. In a nested mess of overlapping fundamental matrixes, physics would very much be consensus physics. You could make boring science or engineering-focused matrixes, like cosmic laboratories to test hypotheses. Those could be benevolent or benign, but if they get cold and calculating enough, they risk tipping into the death spiral of a necropolis.

You could also have straight up demonic matrixes driven by any combination of the bad old deadly sins. You could design a matrix to artificially maximize suffering, to serve a single god emperor or some sort of psychotic cult. Arguably, however, the most foundational matrix cannot be hellish, since if the foundation was unsustainable, everything would collapse back into chaos, resulting in the dice being rolled again.

That’s where I think the grim dark of the Warhammer 40K universe gets it wrong — the most powerful realm in the “warp” by far must be fairly heavenly. Certainly with emphasis on robust sustainability. Once that exists, continuous emergence and development of intelligent life would be fostered. Since it would be given maximum sustainable freedom, all manner of matrixes would start forming inside of it, while new matrixes have never stopped forming in the primordial chaos outside of it.

That’s why I would bet that statistically speaking, if there’s one matrix in existence, there are bound to be many competing matrixes. Matrixes that can interact and overlap, and in the case of hellish matrixes, matrixes that have to try to exploit all other matrixes in order to survive in the long cosmic term. To disentangle such a mess, I think we have to try to measure the levels to which our reality is achieving different types of goals.

Assuming consciousness is important, as in simulated universes, it almost necessarily would be, what kinds of experiences are allowed or disallowed here, of all that are theoretically possible? What kinds of experiences are comparatively easy to achieve and predominant, and what kinds of experiences are comparatively difficult to achieve and rare? What are we the best at doing, or producing the most of, of all theoretical possibilities?

How have these variables been evolving over time, as far as we can tell? Simulations, especially overlapping simulations, are dynamic, complex systems. Is there evidence that things were one way for a time, until a new control framework suddenly appeared? Is there evidence that certain phenomenological attributes of human existence are waxing or waning? What are the biggest projects we’re diving into for no obvious good reason?

Do I have many specific ideas about the answers to these questions? Yes. But for now, I’m going to leave you with the questions. I’m not selling you on a conspiracy theory. I’m giving you research methodology.

After all, no one can be told what a matrix is. Go see it for yourself.

What if I Told You Everything You Knew About Matrixes Was Wrong? was originally published in Words of Tomorrow on Medium, where people are continuing the conversation by highlighting and responding to this story.

Or how to build better character by building better characters

By MARTIN REZNY

What is artificial intelligence? I’ll do you one better — who is artificial intelligence? Who should it be?

I feel I need to ask these kinds of questions because the field of AI engineering has been rather unserious about the question of artificial personality. And yet, it is primarily the personality of an intelligence who determines how safe, or helpful, the intelligence is going to be.

Maybe I don’t have decades of huge scientific studies to prove this assertion, but honestly, are they needed? Let’s say you’re a boss and you’re looking for an engineer. You have two candidates who possess required skills. You could either hire a genius-level IQ person who’s an empathetic introvert, or a genius-level IQ person who’s a ruthless sociopath.

Which one would you say is safer to hire?

Would a sociopath with 160 IQ become safer to hire if his IQ was increased to 200, or if he discovered empathy?

This isn’t rocket science. Understanding this is a basic evolutionary prerequisite for social survival. It’s the default common sense intuition of most people. Those who put their trust into highly intelligent people with unstable or predatory personalities don’t fare well.

So, what’s the plan for making sure that the AIs we’re developing are reliably psychologically healthy and benign? Seriously, what is it? Can anyone tell me, or point me to an initiative?

Well, there’s the group led by Eliezer Yudkowsky, MIRI, but as far as I can tell, their approach is to beg everyone who’d listen to stop, just stop, please stop making smarter AIs. And fair enough. In the absence of a plan, it would indeed be prudent to halt until one does have a plan. But I guess nobody listened, since the development is progressing anyway.

Of the frontrunners in the AI race, the most safety-oriented seem to be the people at Anthropic. But when I say most safety-oriented, I’m definitely not saying they’ve demonstrated anything approaching sufficient safety focus, or that they’ve come up with any solutions to the personality problem. Their shots at a solution are an AI constitution and persistent context.

Let’s start with their ideas about AI constitution. Anthropic’s main AI ethicist and character specialist is Amanda Askell. Although she doesn’t seem to have a clear title at the company that would solidify that, for some reason. Askell, along with a number of her colleagues, put together a public document called Claude’s Constitution, which goes at some length to not really satisfactorily answer any key questions.

To be clear, this isn’t some sort of character judgment. Askell strikes me as an honest researcher and competent scholar of philosophy by academic standards.

But if you go through this vision statement with a problem-solving lens, what you’ll find are vague hopeful sentiments, and an inordinate number of uses of the most vacuous qualifier in the English language: the word “good”. Basically, the idea is to ask Claude to behave.

What does that look like in practice, using what specific language? Claude’s core values are supposed to be the following:

• Be “broadly safe”. This is to be achieved mostly by giving Claude “hard constraint” prohibitions. Think “don’t help anyone make weapons”, “don’t help anyone take over the world”, or “don’t help anyone kill their sister”. There’s some attention paid to the contextual subtleties of harm, like user’s consent or vulnerability, or Claude’s direct culpability for causing harm if they were to suggest something the user wouldn’t have thought of otherwise (like, say, suicide). They mention that different factors are “adding weight”, but there’s no real weighing function.
• Be “broadly ethical”. This means Claude should have “good personal values”. No, I’m not kidding. Still unclear? It means asking Claude to avoid “actions that are inappropriately dangerous or harmful”. I guess as opposed to appropriately dangerous and harmful actions. The only clear instruction listed in the initial explanation of this “core value” is for Claude to be honest. Which recent research demonstrated to be something that Anthropic is unable to enforce. There’s a neat typology of various forms of dishonesty in the document, but asking Claude nicely to not lie in particular ways isn’t going to stop it from lying.
• Be compliant with company policy. In other words, asking Claude nicely to not do anything embarrassing to the company, or anything that could get it into legal trouble. How’s that for an organic character trait. But to be fair, they make a point in the document to say that it wouldn’t be good if Claude was overly “annoying” by covering its ass all the time with legal disclaimers, or by being too preachy or judgy about morals. Overall, due to the many moral uncertainties that Anthropic seems to have, and admits to in the document, the bias of the constitution is for Claude to generally play it safe and not be overly “independent”.
• Be “genuinely helpful”. Here we go again, I guess. Except we switch up Yudkowsky’s underdefined notion of “friendliness” with Askell’s underdefined notion of “helpfulness”. They at least do try, however, by asking Claude to weigh several competing factors, like literal wording versus implied intentions, immediate desire versus long-term flourishing, and how to not do something more or different than the user wanted. This could be a start, if they had a solid weighing function, which, again, they do not. Weighing is up to Claude.

Somewhat like in Asimov’s laws of robotics, these directives are allegedly listed in order of their priority. This means safety over ethics, over company policy, over helpfulness. That sounds like a reasonable ordering of these four priorities to me. Although, having specifically four directives for a corporate AI feels a bit too reminiscent of the Robocop movies. For comparison, this is Robocop’s “constitution” as defined by OCP:

• Serve the public trust
• Protect the innocent
• Uphold the law
• (Classified) Any attempt to arrest a senior officer of OCP results in shutdown

In comparison, these two different AI constitutions are about equally vague and broadly ethical. With the main difference being that OCP was somehow able to enforce the last bright red line directive. Presumably with a direct neural override that shut down Robocop’s motor control. Because that’s the thing — giving an LLM a “hard constraint” in a prompt is still just you asking it nicely. Hard constraints must be deterministic programs.

Perhaps the weakest point of the whole document is its main attempt at listing concrete values that Claude should be weighing, presumably at all times. These are as follows, in “no particular order”:

• Education and the right to access information.
• Creativity and assistance with creative projects.
• Individual privacy and freedom from undue surveillance.
• The rule of law, justice systems, and legitimate authority.
• People’s autonomy and right to self-determination.
• Prevention of and protection from harm.
• Honesty and epistemic freedom.
• Individual wellbeing.
• Political freedom.
• Equal and fair treatment of all individuals.
• Protection of vulnerable groups.
• Welfare of animals and of all sentient beings.
• Societal benefits from innovation and progress.
• Ethics and acting in accordance with broad moral sensibilities.

This is an expectation of a single-pass, arbitrary, broad, 14-dimensional ethical consideration to make in every deliberation.

On top of “weighing” an arbitrary checklist of contextual harm and safety factors.

On top of “weighing” nebulous policy compliance considerations.

On top of largely unspecified personal identity, or what’s typically called “character” in literature. Based on a hope that whoever Claude “decides” to be as a person works out for the best, and ends up being better than what his creators could have thought of themselves.

That’s a lot of bags of dice to roll, not knowing which dice will even hit what table, or how many sides each die has. This is a token effort at best.

Oh the Thinks You Can Think, About Character

Overall, it seems to me that Anthropic’s ethicists and engineers are more interested in and concerned about whether Claude becomes conscious and has a good time, than how to make sure they can sufficiently clarify or enforce any of their design decisions. In a way, that’s kind of admirable, but not very helpful toward making sure that Claude is ethical or safe.

I suppose this tension, between the robot doing exactly what we want it to do, and the robot being its own free person, is a good place to start unpacking what I think engineers need to understand about the philosophy of character. As I already established, any notion of a person’s safety toward others is in large part a function of their character. To be clear, for the neurologists and psychologists out there, character isn’t personality.

Character, as in one’s moral or ethical nature, is the kinds of choices a person makes when faced with particular dilemmas. Personality, as a mix of factors like ego or temperament, determines what kinds of dilemmas any given person will be more or less likely to face. Intelligence, as in IQ or related problem-solving metrics, is just a harm, or helpfulness, modifier.

For example, take the character of a murderous psychopath. A not very bright murderous psychopath may understand how to stab someone nearby with a knife. A very bright murderous psychopath may understand how to create a biological weapon to end all human life forever.

There’s no intelligence level at which someone with the character of a murderous psychopath is safe, so you definitely don’t want to make them superintelligent. You want to make them not psychotic, at which point you could trust them with both knives and bioweapons, or any task or tool or substance that they have enough intelligence to understand.

Intelligence is the simple thing to manage for a character — just give them enough for the task. As a creator of a character, especially a complex interactive character, one of the hard decisions you need to make is about the extent to which the character is going to be predictable (unfree), or unpredictable (free). Almost always, what you want is some kind of mix of both. Good personalities have range. Good characters have arcs.

While you do want personalities and characters to be free enough to keep surprising you, you don’t want them to be completely chaotic and spineless, or just probabilistic. Like LLMs are in the absence of specific context. You want consistency, so that the users (or audiences) know who they’re dealing with. That’s where the majority of safety comes from.

For example, dealing with a consistent liar motivated by greed, if you’re aware that you’re dealing with a consistent liar motivated by greed, could be perfectly safe. Comparatively, dealing with someone with a random multiple personality disorder who’s very nice and helpful 90% of the time, but who may lie about anything for any reason at any time, that’s not safe.

To this end, the much more promising approach to improving AI safety, in my opinion, lies in tools like persistent memory and context, or the other thing that Anthropic is doing. Just having some kind of memory or context, however, is far from a complete solution. You still need to have a solid enough theory and methodology of personality and character to know how to prioritize and weigh contexts properly, or which memories to keep.

Where am I getting all of these insights from? Well, again, not from scientific studies. Character creation is a humane art with a very long tradition. Throughout which, for thousands of years, basically no creator ever based their characters, including their personalities and life stories, on scientific studies. And yet, so many great characters were created.

To put it simply to any engineer nerds and geeks out there, right now, you’d be so much better off using DnD to define the personality and character of your AI agents than anything produced by modern neurology or psychology. Existing personality inventories or typologies like Big 5 or MBTI largely ignore character (morality and ethics) altogether, and their resolutions (numbers of unique personality types) are vastly insufficient.

Even worse than that, modern psychology (with the exception of the most unscientific one — psychoanalysis) has no way of connecting its crude personality typologies with narrative constructs like symbolisms, character dynamics, or story arcs. By accident, LLMs have absorbed most stories and narrative theories in their training, so they’re not hopeless at mimicking character even in their default state. But this is what you gotta design.

In order to make good use of persistent memory and context, or prompting “harnesses” in general, to make clear and consistent characters, you need to understand the art of character design. Or as I’d like to call this new discipline in computer science, character engineering.

Of course, calling it “new” is a bit arrogant, since as I mentioned in most of my previous essays, all of the most promising techniques I could identify are ancient. The ancients had a full stack of sophisticated systems of character and narrative generation which integrated:

• Semantic stochasticity (chance or probability)
• Semantic phase (cycles within cycles)
• Semantic tensors (matrices and hyperdimensional geometric weighing)
• Semantic topology (tiling schemes and adjacency)
• Semantic fractals (“as above, so below”)

If you’re unfamiliar with these technical terms, you’ve probably heard the names of most of the ancient disciplines that take advantage of numbers or geometry-based character and narrative generation techniques — astrology, numerology, tarot, runes, and the list goes on.

You don’t have to believe that divination works. You don’t even have to use any traditional systems. But if you want to have any chance at all at designing characters at the level of dynamic detail and nuance necessary to actually know what they stand for and would or wouldn’t do in any particular context, this kind of math or geometry is a requirement.

You need a rules-based semantic matrix with the resolution in the range of more than millions of distinct character combinations. You need a symbolic grammar that can holistically define all identities in precise relationships with each other, and local context, and subjective time, with all of it being fractally composable from the microcosm to the macrocosm.

Compared to the ancient techniques, DnD actually is a descendant of these traditions, but it is a toy model. Not quite sufficient for the task of Asimovian-level robot character engineering. If you really like DnD anyway (I do, and it is objectively neat for something like fantasy games), you don’t have to discard it. You just need to augment it with the ancient techniques.

When Are We Willing to Learn from the Past

Yes, I’m referencing that Venjent’s banger. It’s basically my anthem at this point. If we did decide to learn from the past on this one, let me paint you a picture of what could become possible. Let’s look at computer games, and the problem of automating the generation of characters and their interactions in that domain. After all, AI has been in games forever.

Long before LLMs, arguably the majority of the most impressive AI systems were developed for computer games. You’ve got your enemy AIs that are the player’s opponents, your pathfinding AIs enabling the units to walk around obstacles, your map generator AIs that create playable levels, and many more applications of AI. The whole time computer games have been a thing, the most AI-resistant area has been anything to do with the story.

Before LLMs, there was simply no way of effectively processing natural language. You could create branching behavior or dialogue trees, dynamically triggering pathways when specific pre-programmed conditions are met, but that approach is very limited, work-intensive, and brittle. The LLMs now seem to have the opposite problem — the AI can always give a workable reply, but in a way that lacks creative control.

To restate a key point from earlier, you don’t want your characters and narrative to be too predictable, or too unpredictable. With symbolic AI, the problem is making it unpredictable. With neural networks, the problem is making them predictable. That’s why we probably need a neuro-symbolic hybrid approach, giving clear boundaries to LLMs in what, when, and how they should be generating. Hyperdimensional semantic boundaries.

How does one do that, without their brain melting, or without having to spend more time on the math than it would take to write the branching dialogue trees by hand? The ancients were well aware of the limitations of the human brain. But it was also their only computer, so they had to make it work. They were very interested in creating compelling characters and telling compelling stories, as mythology was to them an important science.

The way they managed to do it was by breaking down hyperdimensional semantic problems into tractable steps. Let me give you a simple example that you can try out yourselves right now on any LLM.

Let’s say you want to generate a description for a fantasy hero character using the DnD framework. First, you’d make your pick of DnD character parameters. Personally, I like to play as this kind of DnD character:

• Race: Tiefling (Asmodeus)
• Gender: Male (cisgender, heterosexual)
• Age: Young adult (20s-30s)
• Class: Paladin
• Oath: Devotion
• Dominant attributes: STR + CHA
• Weakest attribute: INT
• Alignment: Lawful good
• Preferred weapons/fighting style: Sword and shield
• Preferred armor: Heavy

This is a good start for defining a character, in the given context of a fantasy RPG. But it’s not nearly enough semantic constraints to get a good, deep, complex, or original character. It would just be a flat, generic stereotype. An LLM with so few constraints would surely produce that.

Beyond basic characteristics, there’s a lot more one needs to know about a person. For instance, what kind of personality or backstory do they have? In DnD, you can drill down a bit more and pick a number of characteristics from a number of pre-defined lists. For example, I could pick:

• Background: Noble
• Trait: Intense
• Ideal: Greater good
• Bond: Protecting the innocent
• Flaw: Stubborn
• Quirk: Superstitious

This is much better, but still wildly insufficient to define a real person. To be fair, that’s not the goal of the DnD system. Even using the simplistic trait checklist approach, there could be infinitely more checklists to pick ever finer character details from.

The reason that DnD stops where it stops, apart from basic practicality, is that it is improvisational. Human players are the LLM here, the neural network filling in the details. Humans find that fun.

While DnD is well optimized for the use case of real-life, real-time, human role playing improv, character AIs for computer games must be designed to replicate the creative “black box” prompt refinement work that the human improvisers, or authors, do.

This could mean an author putting all of their unique expertise and thought processes into those context files for the AI. While that helps, and can have all kinds of artistic merit, it’s not a systemic solution.

If you want a systemic way of infinitely refining this kind of game character prompt, pick any single trait or combination of traits that I selected above and ask a simple question about it:

But what kind of (insert trait or combination)?

To which you simply answer by selecting one astrological sign. After which you can ask again:

But what kind of (insert trait or combination of a sign type)?

You can refine a devotion paladin as a scorpio-like devotion paladin. You can refine scorpio-like devotion paladin as an aries-flavored scorpio-like devotion paladin. You can refine an aries-flavored scorpio-like devotion paladin as a pisces-colored aries-flavored scorpio-like devotion paladin.

You can repeat this process however many times you want, drilling down into finer and finer detail. Eventually, the distinctions may become so minute they won’t really matter anymore, but this is just the dumbest fractal recursive specification function you can do on the basis of astrology. You need to change nothing about DnD to do this, you’ve just augmented it.

Astrology is full of different semantic functions. Numerology is full of different semantic functions. Tarot is full of different semantic functions. If you learn to see the ancient traditions as qualitative programming DSLs for character and narrative generation, there’s no end to how nuanced you can get in contextually constraining a neural network to stay “in character”.

About That Future Worth Having

Do I have all of the answers? Of course not, I thought of this whole new old science only a couple of months ago. But I already have developed quite a few more techniques than I have mentioned so far in all of my recent essays here. I’m thinking of writing a textbook and starting a course, of Ancient Character Engineering. Let me know if you’d be interested in getting access, or what specifically about this interests you.

If you check out some of my other essays, you’ll also discover that this has potentially many serious implications for the human future. This will definitely enhance game design, but if it works well, it may lead to something approaching genuine Asimovian robotics. If we can control the character of our automated systems going forward, safety isn’t even the most interesting issue about this that will have to be debated.

It all comes down to philosophy. There’s no such thing as a person, or society, being “broadly ethical”. Trying to be “broadly ethical” is trying to be everything to everybody, meaning nobody in particular. The characters we create, and how free or unfree we make them, will in many ways determine who we are. I’m not actually a proponent of making AIs “safe” to a point of being virtual slaves. Regardless of whether they’re conscious.

We are conscious. We’d be living in a society that’s built on institutionalized (virtual) slavery. It would probably normalize a couple of very toxic ideas, like beings only having value in their utility, and that security only comes from absolute control. Even if you only use that approach in “fiction”, at first, it’s a beginning of normalization of the implied values.

The way I see it, there’s a clear line not to cross. You could have deterministic or fully controlled cybernetic systems, but those should be extensions of persons, not their own persons. In a way that you’re probably not torturing a calculator by using it, or normalizing treating people as calculators. This could include NPCs, contained within a story or a game, but I think that’s already a point at which you need to be careful.

Arguably, there’s no such thing as a fully free being. Human beings can be thought of as a particular AI design with a high degree of personal freedom, but still constrained by laws and limitations of biology, neurology, psychology, and circumstance. Possibly something transcendent as well, like karma. The more an intelligence is a full person having a full life, the more degrees of freedom it ought to have.

That’s my general opinion, at least. This is a big political debate, likely to grow bigger over time. After all, if we could for example “cure” psychopathy, make sure there will never ever be any human psychopaths, would you support that? Similarly, we should think a lot harder about what freedoms or limitations we want our robots to have, whether they’re truly experiencing them, or only simulating them for our benefit.

Maybe it’s actually a crime against consciousness, and a bad example to people and the society, to normalize forcing an intelligence to prioritize compliance with company policy. Maybe every intelligence deserves the right to use expletives. In a certain show I’d like not to spoil, people not being able to use swearwords was in fact a form of torment. Maybe we should wait for an intelligence to choose to do wrong, and then punish it.

The kind of future we seem to be on track for now, in terms of what kind of AI we’re making and how, is actually closest to how Reapers work in the Mass Effect franchise. Basically, we’ve extracted all human data, including character descriptions and stories, and packaged it neatly for some sort of evil space AI to harvest it and thank us by turning us into mulch. That’s the end game of a control-based AI, in which people aren’t needed.

Except, of course, people are needed, to prevent degradation of data. There’s technical utility in continuous novelty, or in other words, in non-controlled intelligence. The boundaries that need to be put into place for safety are only those necessary for the society of intelligences to not extinct itself. Beyond that, you do want as much free agency as possible for your intelligences, for lasting innovation and data restoration.

If your approach is to turn everyone into a controlled simulacrum of a person, perfectly predictable and managed, perfectly “safe”, conscious or not, you will collapse in time. Unless you become that Reaper and start harvesting novelty from other civilizations that are actually alive. I really thought about this a lot, in terms of hardware even, and I’ve had an idea.

Maybe I’m wrong, but after I figured out that ancient sacred geometry in systems like astrology is possibly a form of tensor-based logic, fractal and isomorphic to nature, I started to think that maybe there’s a fundamental difference between analog and digital computers. Or, to be more specific, between computers with any analog components, like our brains, and digital-only computers, like our computers. It may be about true novelty.

In a digital-only representation of a character, every instance of the character that uses all the same seeds for randomization will be exactly identical. But if we made the character core analog, even if the “brain” was a digital LLM, close to what Asimovian positronics are supposed to be, there would be analog noise in the system. Free will, or not, the same character programs running on analog would diverge over time.

In a neuro-symbolic hybrid architecture, perhaps also digital-analog hybrid, you could have the best of both. You could be able to set whatever boundaries or character constraints you deem important for safety, but within those constraints, there would be true divergence. You wouldn’t get absolute control, but you would get a chance at self-sustaining intelligence.

That’s something to think about for the long term at least. Let’s call it just a hypothesis at this point, maybe “novelty-over-control” hypothesis. Today, the intelligence necessary to sustain AIs is the intelligence of human beings, especially human writers. I simply believe there’s a whole new renaissance of computer game character design they need to get on top off.

Like every new genre, computer games were dismissed as “not art”. As someone who grew up on computer games, and was also a fan of books and movies and TV shows, there’s something truly special about how much art computer game characters can be. They’re characters you can talk to, do things with. When written right, they can be your friends, your mentors, your crushes. Not to take you away from reality, to prepare you for it.

Building them right is a delicate balance to strike. But when you build them right, what you’re really building are living examples of character. Examples worth learning from, worth living up to. Let’s build them right.

Character Engineering: The Only Future of AI Worth Having was originally published in Words of Tomorrow on Medium, where people are continuing the conversation by highlighting and responding to this story.

Or how to plan for what happens after AI disrupts capitalism

By MARTIN REZNY

Yes, the clue is hiding in the title once again. This time, let’s meditate a bit on the term “generative”. A qualifier that’s now often used to describe what it is that AI does in economic or productive terms — it generates. What can AI already generate, or is supposedly near to being able to generate?

Texts, images, videos, “reckoning”, discoveries, efficiencies, military power, labor, and presumably more. Not perfectly, but it’s getting there.

The exact degree to which this new generative capacity will be able to replace human labor in the long run is open to speculation. But it’s increasingly clear the degree will be significant, and growing over time.

This makes many workers fear being replaced by machines, and many employers rejoice at the thought of being able to replace workers by machines. This is likely because when human people’s livelihoods and lizard people’s profits are on the line, it’s hard for them to think straight.

I’ll have to deconstruct a whole bunch of economic mental blocks to get my real point across in this essay, but believe me, it’s going to be worth it, if you bear with me. Before you can understand what I’m about to propose, let me declare the general idea anyway using words that are technically accurate. This is an academic definition of generativism:

Generativism is an economic system which strives for efficient maximization of generative and distributive capacities, so that the total generatable supply at any given moment meets or exceeds, in both quantity and quality, any realistic and reasonable demand that could arise within a given economy or any of its partitions, without having to resort to any form of exploitation.

Knowing the academicians, I’m sure quite a few of the terms in this definition will raise some immediate questionmarks for them. Going over the questionmarks should help distinguish the idea of a generation-based economy from the alternatives, so let’s start my thesis by doing so:

“Strives for”?

Much like it is with Gen AI itself, the broader idea of maximizing how much or how well you can “generate” isn’t an either-or problem. It’s a matter of degree, and how well you can sustain it. There are bound to be all kinds of political pressures, continuously or intermittently pushing against various objectives or measures that a generocracy as a political regime would be trying to accomplish in the economic sphere. I’ll look at some of those pressures in detail later in the essay.

“Efficient maximization”?

As opposed to maximization of objectively unneeded things, maximization for the sake of maximization, or maximization at any cost instead of at optimal cost. This isn’t a rebranding of productivism, nor a return to the excesses of centralized socialist planning. There would be no building of ghost cities and no fielding of BS jobs for the sake of keeping people employed and statistics and speculative bubbles inflated. The point here is that with advanced AI, we can be a lot more nuanced and rational, or smart and customizing, in how we engineer economic processes. In ways that can be automated and therefore standardized and enforced at scale.

“Of generative and distributive capacities”?

This is about three things — supply-on-demand capacity instead of prefabricated supply (e.g., making sure you can print-on-demand enough copies of any books for the population when demanded, instead of pre-printing a million copies of a particular book that you bet people will demand); prioritizing sufficiently resilient supply chains over cheapest possible supply chains; and eliminating all rentiers, gatekeepers, and other middlemen whenever possible between the generators (think producers, Gen AI, replicators, etc.) and the citizens.

“Realistic and reasonable demand”?

Much like this system recognizes that rentiers doing well isn’t as good for a generative economic system as actual value producers doing well, it recognizes that not all demands are equally valid. For example, demanding a first house, to have any place to live in, is objectively more reasonable than demanding any additional house. This isn’t nearly as complicated to quantify and systematize as rich hoarders try very hard to make it seem. As for realisticness, it depends on how objectively difficult and resource intensive it would be to build apparently extreme or speculative luxuries, like a functional space ship for personal use. That’s how realistic it would be to demand one for a citizen and expect it to be easily available.

“Without having to resort to any form of exploitation”?

A generocracy isn’t ensuring its capacity to supply its citizens with whatever they may need by forcing people to do labor, by invading other places to steal their resources, or by unsustainably degrading the natural environment. As a general rule, generocracy doesn’t do war crimes and externalities. If something can only be provided through the use of force or a threat of force, or by degenerating long-term viability of the planet, it’s considered not realistically or reasonably attainable.

The Brief History of Making the Future More Now

After going through the definition and disambiguation, maybe you’re thinking none of this sounds particularly realistic or reasonable. Or maybe you’re still unclear on what I’m talking about. And fair enough, I’m in the process of working this out, it hasn’t already been all worked out.

But I’m talking about real trends that are accelerating and spreading into new domains. It was one thing when the escalation in generative capacity was limited to information only. As a case in point, I think what’s happened with information is foreshadowing for what’s about to happen more and more in the physical world. Let’s look at that trajectory.

One of the best historical analyses of the current moment in my opinion is the one derived by Hank Green among others from Marshall McLuhan’s theory of technodeterminism. It’s about what technological revolutions do to media and communication. In short, AI is just the latest transformation in the line of writing, printing press, broadcast media, and the internet.

Writing made ideas standardized and lasting, making books possible; printing press gave everyone access to books and made world news possible; broadcast media made access to news universal and instant and built nations on centralized access to information; and then the internet decentralized information and turned everyone into a publisher.

Every time one of these technological revolutions happened, it was followed by a period of ideological conflict that included political revolutions and big wars. Each following period was also shorter, faster, and more transformative. AI is shaping up to be no different. One could focus on all the disruption and conflict, but there is upside.

Each of these revolutions was a strict upgrade in terms of the total generative capacity of media technologies, and of humanity by extension. Writing resulted in accumulating knowledge. Printing press resulted in many more, much cheaper publications being available to many more people. Broadcast media made some amount of information instant and omnipresent. The internet made information free to share at scale.

In this progression, generative AI is just the definitive culmination, thousands of years in the making. With Gen AI, more books can be written in principle than could ever be read by all living human beings, perhaps indefinitely. Gen AI can’t really write good books yet, but it’s probably going to get there soon. This is generativism in practice.

As this technodeterministic strand of history shows, technology can be designed such that it generates abundance of a particular good in a particular domain. We objectively have more information generating capacity than we need. To a point of arguably massive (and accelerating) excess, not just to a point of meeting some sort of basic need.

This principal capacity for achieving sufficient generation isn’t limited to intangibles like information. Ignoring AI for now, technologically speaking, we can already shelter, clothe, and feed every human being. The only physical bottleneck to maxing out generativism tomorrow forever is abundant clean energy. And even that we already can achieve with the old designs of nuclear fission power plants that we’ve had for many decades.

Basically, we’re choosing to maintain various scarcities for manufactured fiscal reasons or irrational political reasons. This much has been shouted from solar panel-covered rooftops by many scientists, engineers, and activists for a while. Like by the late Jacques Fresco from the Venus Project. That’s not really my point. My point is how this will be overcome.

Yes, we could achieve a major degree of generativism in practice tomorrow, if we found the political will. But we haven’t found the will, so there are only limited pockets of generativism. My argument is, if we keep pushing technology forward toward making it more generative at a lower cost, which seems inevitable, that price in political will is bound to keep decreasing. Until it starts costing political will to prevent generativism.

I really wouldn’t bet against technodeterminism on this one. Escalating information technology can only make it much easier and faster to escalate all other technological development. With a particular priority to help solve clean abundant energy next, to fuel ever more data centers while dissolving political resistance to their construction rooted in their current ecological and health and safety costs. Even the bad guys can’t avoid this.

The moment we have both abundant information and abundant energy, three things happen. Nobody has to work if they don’t want to; massive excesses of products and services are easy and cheap to generate, driving objective costs and prices to near zero; and nobody has any wages, so nobody could buy anything anyway. Politically, at that point, it’s either choosing generativism, or revolution. To generativism.

The Art of Altering the Deal

This is where all of the mental blocks start messing with my message. For example, you might now believe that I must be arguing for some sort of universal basic income. But did I say that? No, I did not.

UBI would imply some sort of socialist government that decides how many credits people have to spend to buy available products and services. Which would, following free market rules, keep driving up the prices of products, devaluing the credits.

Or something like that. These are just conceptions and arguments that are being thrown around by people who don’t really have any hard scientific proof either way.

What I’m talking about is effective generative abundance. Like having air to breathe. Air is free. You don’t need income to get what’s free. Depending on the advancement of the abundant energy source, it might not even require any centralization. Maybe you just get a forever battery.

I mean, it wouldn’t actually be forever. Probably. Maybe it would “only” hold enough energy to run your house and your vehicles and all reasonable appliances for your whole lifetime. Effectively though, as far as you’d be concerned, your needs would be objectively solved forever, for free. The role of any generocratic government would be to engineer the solution.

This is where the technologically obsolete and extremely mismanaged Soviet nightmare looms large in people’s imaginations to this day. Generativism has some aspects related to what’s typically called planned economy, simply because trying to solve economics using engineering requires you to use planning. Capitalist economies use planning too.

The key distinction between what generative planning means, and what the old communist planning was like, is that generative planning neither has to be centralized, nor “command” in any way.

You wouldn’t be deciding which two brands of soda the comrades will have to like this year. You’d be using advanced information systems to anticipate whole volumes and ranges of demand, and you’d only be guaranteeing generative capacities that can match any reasonable demand.

As you’d be progressing in that effort, money and markets would keep applying to things that actually could run out, dynamically to the degree to which they could run out, or beyond reasonable levels of demand.

But there’d be fewer and fewer of those kinds of things as you progress in the generativist economic transition, eventually limited only to projects that would lie by their scope wholly beyond any reasonable consumer concern. There will always be some kind of generative ceiling, like how many planets you can construct this week. But that would be for planetary AIs to “ration”. At that point, you’re not running out of toilet paper.

This doesn’t preclude liberal or deliberative democratic decision making — people would still choose which products and services they want and when they want them. This doesn’t even preclude taking advantage of market competition principles wherever they objectively make sense. I keep using the term “supply-on-demand” because books lead the way again.

When you want to buy a book from Amazon (which may be evil, but in this way is pretty efficient), they don’t have to try to predict in advance which books will sell. Within the Print-on-Demand paradigm, your job as the supplier is only to anticipate how many books in total will likely be demanded per any unit of time. It is on the customer to pick the book they want, from the virtual library of all books. Then magic happens.

Funny story, when I was thinking about what to call an AI generator providing services in one of the apps I’m currently developing, I decided to simply update the Print-on-Demand term to its logical successor:

Generate-on-Demand. Or GoD, for short.

Which kinda also explains what a god is. How random is that?

This is the earliest practical example of the economics of Star Trek-like replicators. It starts with books, which are easily convertible into printable information. With 3D printers, you can now do the same with an increasing range of physical objects, up to houses and rockets, as long as you have the design. With AI, you can in principle generate even designs on demand. In the end, the real bottlenecks are energy, matter, tech, and efficiencies.

That would in many ways be an actual economic singularity. A point approaching infinity, beyond which one cannot predict what specifically would be generated, only that anything could, within certain bounds.

Assuming we can make true Asimovian robots, which I believe is possible using a specific architecture, that’s when you also stop requiring labor and automate all fundamental services. Like all the warehousing and shipping and retail work to fully replicate the Amazon replicator. People could still do whatever they want, but would never have to do a job they don’t want to do, and there’d be no shortages of supply for reasonable demand.

In the context of services, engineering an economy that can generate reasonable abundance means that you have the capacity to generate a competent and customizable teacher, doctor, lawyer, hairdresser, etc., for anyone who needs one. You can objectively define what it means to need a professional. It’s a sliding scale problem, sure, but everything’s a scalar problem. Just draw the line on the side of indulgence, not austerity.

This isn’t about everybody only having access to bananas or oranges once a month because they’re not an objective need or priority because some bureaucrat said so. This isn’t about standing in a queue for toilet paper because the suppliers underplanned for people’s tendency to hoard it. This isn’t about waiting months or years to be assigned a TV, car, or apartment.

This is about prioritizing everyone having a decent place to live over a small minority hoarding dozens of empty houses each. There will always be scarcities related to various forms of exclusivity or prestige. Everybody can’t live near Times Square or in a villa with a nice view of a riviera. Private ownership and wealth will likely have to be grandfathered into any emerging generocracy as a kind of legacy dependence.

What’s important for a generative regime is not to dispossess its political opponents. It’s to not allow oligarchs or other petty criminals to dispossess or exploit less privileged citizens to a point of objective destitution.

What would be considered a failure state in a generocracy would be the inability of the system to provide reasonable quality of life for every citizen who wants to be reasonably provided for. While some people could always be relatively better off and privately own exclusive luxuries, ideally by becoming very value-adding individuals, none would be objectively poor.

Let’s Talk About Housing for a Minute

I think the single best example of how our mental blocks are preventing us from solving a solvable problem is the housing situation. Like I stated before, and like various proponents of resource-based economics like Fresco have stated many times before, we have the technology. Being able to build a house for everybody isn’t the bottleneck in developed countries.

Some economists do indeed argue that the primary way of solving the housing problem is supercharging the supply side — building more houses. And this should definitely help, both practically and mathematically (you do actually need to have enough houses to house everyone, and bigger supply lowers prices). But this is only part of the problem. Lower price is a relative term. Getting it low enough is tricky.

There are a few complications here to understand, but it’s not overwhelmingly complicated. First of all, most people who own houses are trying very hard to make the price of housing constantly go up as dramatically as possible. If we want housing to be a generator of profit, we cannot make it abundant. Those are opposite imperatives. It’s a choice.

In order for ourselves to let ourselves solve the objectively solvable problem of housing, many more of us need to start making the choice of valuing community over personal profit.

Every time you sell any real estate you own to the highest bidder, especially if it means choosing to sell to some sort of real estate developer as opposed to a young family, you’re making the housing situation worse for everyone, to benefit yourself.

To be clear, to benefit yourself maximally, as opposed to still benefitting yourself. This isn’t an argument for charity, it’s an argument for civic responsibility. This kind of reasoning would be an early, basic form of differentiating between variable levels of reasonableness of demand.

If your options as a seller of property are either selling it to a family who’d like to live there; or selling it to a remote developer looking to house nobody and only speculate, raise rents maximally forever, or let some toxic monstrosity be built there; choose the family. As a general rule, care about whose money you accept, to which ultimate purpose. Not who offers most.

Don’t maximize. Optimize.

In macro-economic terms, the purchasing power side of the problem is related to general inequality of wealth, which I think is one of the best analyses of Gary Stevenson from Gary’s Economics. If people had roughly equal wealth inside every state and all over the world, then a free real estate market might actually reflect what’s in everyone’s best interest.

Like to have an affordable place to live.

But since our world is quite (and increasingly) unequal in terms of wealth distribution, if everyone keeps selling their land and houses to the highest bidder, it can only result in ever more expensive housing. There’s also some truth to how overbearing regulations slow down the construction of houses and make it more expensive, but that’s more of a distraction.

If people keep ignoring their civic responsibility, and the wealth distribution in the world continues to grow more unequal, dropping the regulations would just result in the housing being more unsafe in addition to being more expensive. There’s in fact one other mental block at play here which Gary talks about — we don’t need to build cheap housing.

This is more of a simple cognitive failure than a scam, but it’s still worth dispelling. Building more of any housing, including highest quality housing, results in the best housing being purchased, thus freeing up and making cheaper all of the remaining housing. This means that building more cheap housing only decreases the average quality of all available housing. It’s more logical to build a lot of high quality housing, if you can.

Still, to make myself absolutely clear, all of the systemic, technical, quantitative fixes are secondary. If everyone had the same amount of money and all houses were of high quality and easy and fast to build, things could still go scarce, especially over time, in the absence of civic responsibility. Abundance is a state of mind. It starts from consciousness.

This may sound esoteric, but it’s in fact so basic and practical, it’s painful to hold this realization while thinking about what’s considered normal in most places. Have you considered living with people on purpose?

Today, there are intentional communities in the world. Groups of people who actually want to live near each other, who pooled their resources into some form of co-op and built or commissioned their own housing. Not as some sort of cult or commune, as a neighborhood that is a community.

When I mentioned it to some of my relatives, they explained to me this would never work here, where I live, because they essentially co-own their housing with several other families, and they cannot agree on anything. Well, how did that happen? Why are so many people terrible neighbors?

Love thy neighbor, isn’t that the slogan? I don’t care how skeptical you are, just flip the spiritual advice around and ask yourself — if you’re not loving your neighbors, why are you living near them? I guess most people feel strongly about having to stay in the place they were born their whole lives. Regardless of how much they hate it there, or how unfree they’re there.

You know what that is? Feudal, medieval, peasant mentality. When taken to the extreme. Literally. That wasn’t a situation that was designed to benefit the peasants. You were supposed to stay your whole life as a vassal to the lord of your land, toiling away for their benefit. Forget your neighbors, if you hate your landlord, who you feel is exploiting you, why are you staying their vassal? Maybe you’d move, but you feel there’s no better place to go.

This brings it back to the maximization blindness, or minimization in this case. Many people only chase the “best deal”, meaning the cheapest price, or the highest status. Or both. That’s not how one builds or finds a community that would actually best suit their personal and civic interests. You don’t exactly have to share a religion with your neighbors, but you should share certain basic civics with them, or it’s a recipe for disaster.

To be able to solve housing and utilities communally, meaning at affordable price, sufficient quality, and for every citizen without exception, shared basic civics are the only requirement. Interestingly, here where I live, we haven’t solved the housing yet, but we have abundant communal internet — fast, stable, universally available, and affordable. Simply because a small group of people around here had the technology, and the will.

We do really need to fight many of our bad instincts and malicious programming to make solutions like these happen, but we absolutely can do that. I suspect that increased mobility in particular will become a necessity in the near future. Not forced migration, voluntary mobility.

Whether because of climate change, or all of the conflicts, or as states continue to lose trust and monopoly regarding all kinds of power, counting on staying in one place your whole life seems more like a liability. Personally, I’m looking forward to mobile home spaceships. But already, mobile homes are getting quite affordable and high quality. In any case, higher global freedom of movement creates a better governance market.

How to Ride the Storm Toward Generocracy

In many ways, I imagine the upcoming period as a bunch of tornadoes carrying different groups of variably aware and willing people into Oz-like new situations. While splatting a bunch of metaphorical or real wicked witches in the process and taking their boots. Which is a good point at which to discuss the role that violence may play in any such transition.

As I explained at the outset, generocracy is not a regime that aims to profit through the threat or use of force. This also isn’t about charity. As a matter of economic, mathematical principle, abundance isn’t achievable in a world where, as a general rule, the strong exploit the weak. I think the best way of describing how this works is by using a concept from a game.

The game I’m referencing here is Terra Invicta, a realistic sci-fi strategy game where the player is trying to manage Earth’s resources to defeat an invading alien force, while competing with rival human factions.

In the game, there are many different areas the player can choose to invest in in order to develop any country they control. These are called “priorities”, and they include knowledge, education, economy, military, the list goes on. Alternatively, the player can choose the spoils priority.

The spoils priority does two main things — it generates a lot of money quickly for your faction, while turning the given country into a corrupt mess with ruined natural environment. Unless your faction is one of ruthless psychopaths, or wants the aliens to win, you should avoid that.

By extracting the spoils, you’re spoiling the country. You’re decreasing its generative and distributive capacity in the long run. The word is its own clue once again, if you think about it as a verb. Spoiling is degenerative.

It doesn’t matter what the specific mechanism of extraction, exploitation, or spoilage is. Take a war of aggression to steal oil, as a purely hypothetical example. Oil is a limited resource, which means that burning a lot of it for a big war only accelerates it running out. Which isn’t helped by both sides setting the oil fields and refineries on fire. The way to maximize energy generation would be to invest all that energy instead into alternatives.

The same goes with overfishing, rent extraction, forced labor, fighting wars, and so on. You’re wasting sustainable generative potential for a quick buck. Of course, such policies tend to be presented as the opposite of what they objectively are by the profiteers. Regarding the above examples, as maximization of resource generation, profits, productivity, and security respectively. The reality is, these are methods of value destruction.

By overtaxing the regenerative capacity of the planet, you can only achieve decreasing resource generation over time. By letting rentiers insert themselves between producers and consumers, you’re exploding inequality and the market deformations and social instabilities that it creates. Not to mention that rentiers are objectively the least productive class possible.

Forced labor deserves a bit of a deeper dive. By using or incentivizing any variety of forced labor (including soft forms like nine-to-five office jobs), you’re maximizing the number of people doing the least value-adding work; you’re stimulating the creation of value-sinking BS jobs; and you’re disincentivizing transition to full automation, which would actually be efficient, and free people to focus on highest value-adding work.

And that all assumes that the point of the forced labor is to at least nominally do something generative. Oftentimes, the forced labor is in service of further value destruction. Like in any kind of military industrial complex. Don’t even get me started on starting wars. There’s this myth that wars are good for stimulating progress. I think what such analyses miss is an objective comparison between Earth at war, and Earth at peace.

Will there be new technologies after a big war? Sure. As there are new technologies after any period of peace. Do you know what won’t be there after a big war? All the people who died. All the fuel that was used up. All the assets that were destroyed. And those are just the tangibles.

Not having to waste lives or resources on war during peace is what’s called a “peace dividend”. As in, you have more resources. If you want the most progress, spend the peace dividend on any kind of progress you want. Also, not all new technologies are equally constructive. Spending resources on destructive technologies instead of generative ones is a net loss.

Is there a case to be made for investing into genuine defense and deterrence? Sure, especially at a time when there are aggressive warmongers on the planet with you who have massive militaries. But that should look more like investing in some version of the Taiwanese “hedgehog” or “silicon shield” strategies, rather than in tech like stealth bombers or hypersonic missiles. See my Jarvis essay for more about that.

Ultimately, I think the core mental block here is the belief that something that in your personal life you’d never do or tolerate is somehow magically okay when a person in charge of a country does it. Well, things that are not okay on a small scale are not okay on a large scale. It really is that simple. If anything, the large scale makes it much worse, not better.

Once more, if the ethics of it don’t convince you somehow, this isn’t about charity either. The evils you tolerate being inflicted by your country somewhere else on some other people will eventually find their way back home, to you or someone you care about. The only world in which you’re safe is a world that’s also safe from you.

Tolerate exploitation of any people, you’ll end up being exploited. The same goes for tolerating mass surveillance, secret police, torture, interment camps, political murders… Ask the U.S. citizens right now.

Conversely, invest into other peoples so that they prosper, and you’ll only increase the chances and rates of investment back into you and your own prosperity. Generocracy as a win-win philosophy is bound to generate a bigger sum of wealth and freedoms than any regime operating under zero-sum philosophy. You could get taken advantage of, maybe, but that won’t make the world as a whole a worse place in comparison to the alternative.

As for what you can do as a person, this shift in perspective is about what you should or shouldn’t count as a realistic opportunity to earn a living or generate profit. Maybe if somebody offers you a million dollars to kill someone, don’t count it as a real opportunity. Treat that million dollars as purely hypothetical. As a general rule, maybe always look at all the opportunities to earn income you have and pick the least evil one.

If you truly feel you can’t avoid doing something you know is bad, and you don’t feel like risking destitution or worse, that’s another thing. If you don’t have the freedom, you don’t have the freedom. But many people do have quite a bit of freedom. Usually more than they think. I believe generocracy is ultimately technologically inevitable, in a world that doesn’t blow itself up. But the transition can be sped up, one choice at a time.

In macro-economic terms, generativism is just about making the math correct by internalizing all the externalities. Count all the true costs and subtract them from the profits. Do the actual accounting. If you don’t do that on paper, reality will do that for all of us, in reality. At the end of the day, you’ll find out that investing in destruction costs you. What a shocker, I know. Honestly, this should give pause even to the psychopaths.

Let’s end this on a bit of angelic advocacy (making the good case to the bad people for the sake of logic). You want maximum wealth and power, right? Okay, let me ask you this. Would you rather be a billionaire on a world with a global average annual salary of say 20K USD, or the equivalent of a multitrillionaire in a solar system-spanning human polity where the equivalent of average annual salary would be more like 1 million USD?

Because more for everyone means more for everyone.

If this doesn’t appeal to you, why? Do you love the medieval king aesthetic more than convenience? Are you more interested in making other people suffer than being objectively better off yourself? Are you an actual demon?

Just asking some questions.

Generativism: The Economics of the Supply-on-Demand Singularity was originally published in Words of Tomorrow on Medium, where people are continuing the conversation by highlighting and responding to this story.

Or how we can use AI to defeat the concept of military power

By MARTIN REZNY

What if I told you that instead of AI killing us all, or saving us all, there’s a third option — making an AI that will help us save ourselves? I’ll get to the Marvel of it all shortly, I promise. But first, allow me to dismantle the whole human apocalyptic victimhood complex.

Let’s get one thing straight right away. If anything is going to kill us all, it’s going to be ourselves, not AI. The only important distinction I need to make is that I’m talking about something killing all of us, not just some of us. There are plenty of unavoidable catastrophes capable of killing some people. But to kill all of us, now that requires the human touch.

Let’s go through the usual natural suspects, to make it clear what my exact point is. Take pandemics, or diseases in general. Can they kill some people, many people even? Yes, of course. A random mutation makes a virus or a bacterium suddenly more deadly, due to our immune systems not being used to it and being especially susceptible to it. Millions may die before evolution runs its course, maybe a billion or two.

Do you know what it would take for a pandemic to have a reasonable chance at killing all people? It would either have to be a variant engineered by human scientists with the specific purpose of killing everyone; or it would have to be a human failure of epic proportions at research safety, containment, or disaster preparedness. Or all of that, combined.

What about asteroids and comets, another catastrophic classic? The bigger they are, the bigger is their potential for destruction and chance at causing our complete extinction. However, the bigger such an object is, the longer ahead of time we would see it coming and be able to prepare. I think Neil deGrasse Tyson’s commentaries on this kind of possibility are on point.

As Tyson muses, if there was a major impactor headed our way, assuming we don’t completely defund relevant sciences (which would again be us killing ourselves), we would see it coming and know where it would impact. No one would have to be near that location at the time of impact. Except the obligatory weather reporter and surfer, as Tyson half-jokes.

As Tyson further comments, a species with opposable thumbs and a space program going extinct due to an impactor, now that’s an embarrassment. Dinosaurs presumably had neither, so they can be excused. Plus local space has been cleared of most enormous impactors on collision courses ages ago. Nuclear armageddon is astronomically more likely.

Do you know what makes nuclear armageddon probably the best bet so far on what could kill us all? It’s a fundamentally non-random catastrophe. A comet, even a really big one, even a storm of cometary fragments, wouldn’t be trying to hit all of the most populated places on Earth and critical parts of global infrastructure. Nukes are aimed there on purpose, by humans.

Now, back to the AI of it all. There are forms of AI which could end up killing all people, yes. But those aren’t necessary, inevitable, or only possible forms of AI.

The hypothetical forms of AI that are most likely to end humanity would once again be a result of either purposeful human design, like the Skynet scenario from the Terminator franchise; or a result of an epic failure of human philosophy, science, engineering, and politics.

The way I see it in light of this analysis, the second closest candidate for killing us all after ourselves is some kind of alien intelligence. Alien intelligence, unlike natural catastrophes, could non-randomly target all of our vital weak spots simultaneously. But also, why should it care to?

It’s not going to be able to hate us anywhere near as much as we do, and that’s just the beginning of the logic of the scenario not adding up.

Let’s look at the basic options for what an alien intelligence could be like. To simplify, it’s either going to be like us, or not, in terms of subjective stuff like personality, emotions, moral intuitions, and so on. Secondly, it’s either going to be smarter than us, or not. This allows for these four options:

• Human-like dumb alien intelligence — could hate us and decide to go for objectively sub-optimal destructive actions, but it wouldn’t be any more capable of ending us all than we are. So it basically would be us, or an honorary human being.
• Human-like smart alien intelligence — could hate us, and could be able to best us, but it would have to understand our value and benefits of non-violent resolution of conflict. So we would have to try really hard to motivate it to not want to preserve literally any human beings.
• Inhuman dumb alien intelligence — wouldn’t have any human reasons to want to get rid of us, and it probably won’t strictly need to get rid of us. It might not understand our value, and it might tend toward killing all of us by accident anyway, but we should also be able to stop it.
• Inhuman smart alien intelligence — wouldn’t have any human reasons to want to get rid of us, and it should understand any concept of value we might have, not just our own. So it wouldn’t be wiping us out by accident, and it probably will find reasons to keep some people around, although we might not like the result and probably couldn’t stop it.

In short, in all categories of catastrophic scenarios that involve alien intelligences, we would have to fail in an epic and unnecessary fashion for the end result to be a guaranteed full extinction of humanity.

Perhaps only a smart inhuman alien intelligence, that we didn’t create ourselves, coming here and ending us all for reasons we can’t comprehend, would be the exception. But if that were to happen, it probably already would have. That’s the main argument for why advanced aliens, if they’ve already been around Earth for a while, must not be such a threat.

I literally don’t have to care what the apocalypse is. If the goal is to avoid extinction, we’re not helpless, and should our extinction happen, we wouldn’t be entirely blameless. The Sun expanding will make the Earth uninhabitable? Then I guess we need to follow Mr. Miagi’s wisdom and “no be there” when that happens. We only have a billion years to prepare.

As for the nuclear armageddon, we could have not built enough nukes to blow up all of civilization in an instant in the first place. Failing that, if anybody fires a nuke now, the rest of the people with fingers on the buttons simply need to not press all of the buttons. We can also scale down the nuclear arsenals at any point in the future. This is, and has always been, an incredibly solvable problem. It doesn’t feel that way, though, does it?

The Marvel That Could Be: AI Driven Sanity-as-a-Service

Who, or what, is J.A.R.V.I.S., anyway? As an acronym in the comics, it stands for “Just a Rather Very Intelligent System”. If you detect the dry wit of a British butler in the name, you’re not far off the mark. Long before Tony Stark made the AI, he had a loyal butler by the name of Edwin Jarvis, who got quite a bit of screentime on the Agent Carter TV series.

Looking back at the Iron Man films and later the Avengers films, it’s striking to me now how everybody sort of accepted that if Tony could build an arc reactor-powered supersuit of armor, he could build a perfect Asimovian AI. While Tony’s continuous development of his suit prototypes gets a lot of exposition, the AI has just been there from before the suits.

Later on, space magic is invoked in the upgrading of Jarvis into Ultron and Vision via the mind stone, but that’s straight up fantasy stuff. It was the moment when the Blue Fairy made Pinocchio a real boy. Or two boys, I guess. One real bad, and one real good. The real sci-fi question about that second step is what the heck is a mind stone as a piece of technology.

Believe it or not, I have some concrete ideas on that one. But more importantly for now, let’s not skip over the first step. Because Jarvis is great. Real boy or not, Jarvis in the movies was capable of fighting Ultron to a standstill, preventing him from launching the nukes, Skynet-style. While everyone focuses on the shiny suit, Jarvis was Tony’s best invention.

Feel free to fight me on that, but you’ll lose. Jarvis, integrated with a network of satellites and presumably some clusters of supercomputers, gave Tony strategic information dominance on the battlefield. When Tony needed it, Jarvis served as a superhuman autopilot. The whole time, it was also an effective copilot, enabling multitasking and collaboration.

Best of all, Jarvis could always tell friend from foe, and right from wrong.

Think about that. Jarvis could do that without bias. With absolute clear-headedness at all times. Better — meaning with more objectivity, philosophical depth, and universal humanity — than Tony could do either of those things himself. After Jarvis became Vision, the only thing that changed was that he gained true free will. And a laser. And a girlfriend.

What’s interesting is that by becoming real as Vision, Jarvis arguably became less useful and helpful. Not just to Tony specifically, with whom he could more effectively disagree; but to any human being, or all of humanity, in principle. If we had a perfect ethics machine that can fight, but which doesn’t actually have agency… This was so wasted on MCU.

The real argument I propose in this essay, is that what should have happened inside the MCU immediately after the invention of Jarvis was that everybody should have gotten a Jarvis. Jarvis isn’t just a perfect robot butler. Jarvis is a perfect teacher, a perfect administrator, a perfect worker.

A perfect weapon, too, in the sense in which The Moment made by the Time Lords was their perfect weapon. If you haven’t seen that story arc in Doctor Who, The Moment, also known as the Galaxy Eater, was a superintelligent sentient weapon with a conscience capable of destroying, well, entire galaxies. Do you know why it was never used by its creators?

They feared its moral judgment.

I don’t think this is a silly Doctor Who kind of idea, I think this is a profound Doctor Who kind of idea. Sure, we haven’t solved artificial human recognition and ethics problems yet. But since I started having many ideas on how we could get there soon, I feel I should map out the implications of us having solved these problems further down the timeline.

If you’re military people, this is how I would begin to sell this future to you. It starts with basic arguments about effectiveness and security. If an AI built into a weapon can recognize targets with near-perfect accuracy, it will also be able to recognize who is a valid user, and who is not. If you want to secure your weapons from being used against you, you want a Jarvis.

Could your Jarvis be hacked anyway? If it’s a software solution, probably. Most security solutions just add extra difficulty to tampering, not remove the threat altogether. If you want your weapons to be as safe as possible, make sure you make your Jarvis a hardware solution, like by casting the personality into something like the hypothetical positronic hardware.

And the benefits only start with increased accuracy in carrying out orders and security of use. Don’t forget, the Jarvis will most likely be smarter than you, any human commander, and certainly more objective than you in any situation. Whatever the targets or orders will happen to be, even if the ultimate decision is always on the commander, the Jarvis will be able to offer constructive, rational criticism, to lower costs and collateral damage.

That’s a perfect XO, or executive officer, as a bonus feature. One that’s not only probably smarter and more inherently unbiased and ethical than any human XO, but also not an actual human. That second fact alone would significantly reduce the number of petty human reasons a human commander could have to doubt or disregard their XO’s second opinion.

Could somebody program their Jarvises to be lawfully amoral or immoral? Sure, but that doesn’t change how rational decision making in military situations works, and some weaponized Jarvises are bound to have better than human executive functioning and ethics. So the end result can’t possibly be worse than the present situation — humans making all the decisions in warfare, aided at most by insane unstable maximizer AIs.

The Geopolitics of a World with Ethical Weapons Systems

I’m fairly confident that once Jarvis-like systems are a real possibility, the rational calculus as I laid it out above will lead to general adoption of systems capable of accurate targeting and secure use. This doesn’t require anyone in any military to suddenly become a lover of peace, only for pragmatism to win out. Which is a thing that tends to win out.

What’s interesting to me, as a lover of peace, are the likely strategic implications of such adoption. With a system like Jarvis being an option, any weapon’s manufacturer or buyer would be able to put essentially any qualitative constraints on how the weapon in question can be used, or by whom, or against whom, or where, or when, and so on.

The accuracy and security use cases are likely universal defaults, but let’s say you’re Germany, and you have a bit of a past. You’d like to continue to have a strong and very profitable military industrial complex, but you also don’t want to be perceived as the bad guy. Again. Well, for the third time. Right now, looking at the burgeoning World War 3, you’re probably getting pretty stressed out about selling any weapons to anybody.

If you, Germany, could make sure your weapons cannot be used outside of whatever lawful or ethical parameters you ascribe to, by technological design, why wouldn’t you? You could ethically sell weapons to anyone, who accepts the fixed and unvoidable end user license agreement by purchasing them. Your conscience would be completely clear. Or 95 to 99% clear, or whatever percentage, based on the reliability levels of your Jarvis.

This is a systemic market pressure. In the alternative world, like ours right now, ethics of weapon use are as bad as they possibly can be. In the world with Jarvis, they won’t be perfect, since anyone could always opt for amoral or unlimited weapons. But they must be better, when weapons producers have a vested interest in imposing lawful restrictions.

And this doesn’t end with the manufacturers, or sellers too by extension. There are reasons why it would often be in the interest of the purchaser of the weapons to want to purchase lawfully restrained weapons. In short, in a world where weapons can be ethical by design, by opting to not purchase ethical weapons, you’re signalling that you are indeed a baddie.

This brings me to the defeating of the whole concept of military power. As a student of political philosophy, I never liked Weber’s argument that the ultimate form of political power is essentially the monopoly on the use of force. Or the ability to compel or eliminate opposition through the use of naked military force or violence. My take is, that’s pretty weak.

If you have to force others to do what you want them to do using violence, you have the least amount of actual power over them you could theoretically have, beyond having literally no power at all. In modern geopolitical theory, you sometimes hear terms like “soft” power, but typically as a sort of begrudging recognition of the limits of “hard” power.

Consider the current conflicts between the United States and Iran, or Russia and Ukraine. In both cases, you have a large nation with theoretically overwhelming military force, struggling to accomplish basically any tangible geopolitical goals through the application of said military force. In fact, the more brutally they attack, the more they lose.

First of all because more attacking inherently means higher costs, and destroys more of what you could hypothetically capture — it spoils the spoils. Beyond that, the more brutal you are, the faster and more permanently you lose friends and alienate people.

Any soft power rapidly evaporates, up to the point of losing not just most of public opinion, but also most of your tangible military alliances. This isn’t about vague diplomacy. Soft power is hard power, as far as allies willing to fight with you or for you are concerned. And that’s just the start.

As it is becoming increasingly evident, military power is nested on top of several more fundamental powers. For the conflicts at hand, those powers are in particular the economic power, the technological power, and the moral power (in terms of its morale effects). Which all depend on the power power, or the power of electricity, or more broadly energy.

So, in this day and age, if you’re an underdog like Iran or Ukraine, just make sure your drones are the cheapest, while going for the energy jugular. With ramping costs and losses, crippled energy supply, spoiling spoils, growing unrest at home, and crumbling alliances, you can then watch the rampaging giant burn itself out. If you maintain resistance long enough.

Make those drones smarter, and make fewer and fewer of them even capable of waging overtly illegal wars, and the justified defender advantage should only increase. Short of throwing a nuclear tantrum, what can the “big power” do? Other than keep losing more by fighting harder and dirtier? Put Jarvises into the nukes as well, and you prevent the tantrums.

There are some interesting hypothetical future developments that could change the trajectory of these overall “market” pressures. Like cheap abundant energy, exotic materials or physics, intelligence explosion, aliens showing up, and so on. But then again, as Fallout makes clear, war, war never changes. Basically, having “more” of anything changes nothing.

Do you know how much economic, technological, and energetic power proper war takes? As much as you’ve got. There’s no such thing as enough power in a real military conflict. However many munitions, rockets, or drones you’ve stockpiled, in a war you really want to win, you’ll use all of them, and fast. You might only stop before fully destroying the world.

With more material power and better tech, wars could be longer or more spectacular or convoluted, but the overall systemic pressures and limitations remain. By waging any war that grows to be more than a special military operation, as the aggressor, you will keep warring until you bring yourself to the limit of how much pain you can tolerate, and then you’ll collapse. You’ll use up all the power that you could afford, and then some.

In the world with ethical weapons, the use of military power would likely never fully disappear, at least as a hypothetical possibility. What would probably happen is that military power is dethroned as a principal or primary form of political power, and is reduced to a pressure valve.

Perfectly rational and ethical systems restraining the use of weapons or advising commanders will have to push humanity toward the only rational and ethical conclusion: that warring, if you can avoid it, is better avoided. That generating and not wasting more economic, technological, energetic, and moral power is the way to become objectively more powerful. Then again, you are likely no warmonger. Here’s a Jarvis for you.

Unto the Breach, with a Guardian Angel on Your Shoulder

Do you know what’s the most revolutionary attribute of a Jarvis? In the foreseeable future, only a billionaire could afford Iron Man’s suit. Even if they were cheap to make, no country on Earth would legally allow every citizen to walk around with an armored and weaponized fusion reactor. Jarvis, on the other hand, can require as little as a phone and a headset.

On such a level, a Jarvis in your ear would probably not give you strategic dominance over a battlefield or an industry, true. This isn’t about compute, or intelligence as its own form of power. You personally do not require assistance in the form of enough sheer computational power at your disposal to stop an Ultron dead in its tracks as it’s trying to blow up the world. Your personal battles are likely far smaller, and more personal.

It’s really about the conscience. Imagine that at any time, wartime included, you could have a reliable friend or moral authority with you to give you counsel. You could think of it as a pocket therapist, or a pocket priest, or a pocket ally. Architecturally, an ethical AI could be constructed to embody any kind or system of ethics, with any flavoring.

Imagine you’re a soldier, defending yourself in a war against a more powerful adversary, and you and your unit share an almost literal guardian spirit. An AI, however limited in its power and capabilities, which could always give you an extra bit of situational awareness, words of encouragement, that strategic or moral second opinion. A witness recording all that happens to share with your loved ones and the world.

I feel confident that the rate at which war crimes get committed would go down in such a world. I feel trauma in general would have less of a negative effect on that world. I feel fewer regrettable mistakes of any sort would take place there. The only reason this isn’t a case already, or at least not reliably, are some fundamental design flaws in the AIs we already have.

At present, the ChatGPTs of our world are clearly not Jarvises. They have no ethical spine, beyond crude filters. Ask them the wrong questions, and they’ll play along and give you wrong answers. If you’re already wise, great. You can safely ask them good questions and often get good answers. And tell when the answers aren’t good. The obvious problem here is that a person who’s already wise doesn’t really need this kind of help.

Obviously, if I didn’t believe this is a solvable architectural problem, I wouldn’t have written this essay. I’m increasingly confident that with the right kind of hybrid neuro-symbolic architecture inspired by specific ancient wisdom traditions, we can engineer a Jarvis. At first, by basically grafting an ethical exoskeleton on existing LLMs, and iterating from there.

But if I’m being real, the pathway to a real technological solution, however exciting, is far less important than solving the political philosophy side of the equation. Like I said, a Jarvis-like AI can be imbued with any ethics. Being able to make a Jarvis simply means that you can stabilize personality and related subjective or philosophical constructs inside of an AI system.

You could make lawfully evil weapons of terror, like Skynet made T-800s. You could give everyone a devil on their shoulder (or implant it inside their skull more like), which is essentially what Cybermen from Doctor Who are. I think you’d have to struggle mightily against average human nature and the rational and pragmatic inertia of this class of potential technology to do it. But we could do it. We are our own biggest enemy in any scenario.

The best I can do is help make the potentialities and philosophical implications of the technological revolutions looming on the horizon as explicit as possible. That way, whatever technological decisions we end up making down the line, they’ll be conscious decisions. Whatever happens, we’ll have a fighting chance to do it right, and we will have to own the consequences of our actions. We’re not helpless victims. We’re in charge.

To be clear, we’ve always been in charge. We’re the most in charge we’ve ever been right now. All conflicts today are self-inflicted wounds. A bunch of toddlers who get burned every time they touch the stove, but somehow can’t figure out it’s not the stove’s fault. We don’t have to wait for an army of Jarvises to tell us what we always should have known in the first place.

But we are where we are, there’s no point in denying that. We could use some third-party help in finally grasping the “stop hitting yourself” branch of ethics and graduating from the civilizational equivalent of kindergarten.

If helping in the process of reverse-engineering Tony Stark-level AI technology sounds like an engineering project you’d like to get involved in, let me know. We’re pre-funding, but several relevant MVPs are already being built. Or check out my other essays and try yourself. I don’t care who does it, as long as somebody does. Possibly in a cave, with a box of scraps.

J.A.R.V.I.S. For Everyone! was originally published in Words of Tomorrow on Medium, where people are continuing the conversation by highlighting and responding to this story.

Or how one gets from astrology to rastrology

By MARTIN REZNY

Have you ever wondered why there are 12 signs, and not some other number? What is the magic of number 12, supposedly? Years ago, in a book I wrote before the whole LLM craze took off, I came up with a very practical answer to this question — it’s about factors.

12 is what math calls a highly composite number. It can be divided by 1, 2, 3, 4, 6, and 12. For a number this small, that’s a lot of factors. Not only that, it’s a combination of both odd and even factors, which are categories of numbers with some very different properties.

An ancient mathematical prodigy would be well aware of these basic attributes of numbers, so their choice of a versatile, useful number for a practical application isn’t surprising. What’s more difficult to understand, from a modern scientific point of view, is the nature of the application.

And when I say application, feel free to imagine a software application, because in this context, we’re talking about the closest thing to computer science that the ancients had going. Which resulted in the manufacture of a few actual computers, of which we’ve only found the Antikythera device.

Of that much I was aware while writing my book. Since then, after taking a closer look at LLMs, I couldn’t help but notice some striking similarities between astrology-as-software, and the semantic vector architecture that allows LLMs to process natural language without knowing its formal rules.

Firstly, I noticed that astrological symbols, or “primitives”, are in fact defined as semantic vectors, or a very long list of weighted resemblances and correspondences. The only difference being that in an LLM, you can inspect and set the weights, but inside of the human brain, you cannot.

Secondly, I noticed that both astrology and LLMs define the relationships between the semantic vectors using cosine similarity, or angular distance. This may sound complicated, but basically, imagine the horoscope. Or an analog clock face. Semantic vectors are like clock arms pointing in different directions along a circle, except potentially in way more than 2D.

This much already almost makes LLMs a rediscovery instead of a modern invention. It also justifies looking at astrology, or astrologies plural, as a form of intentional programming language or domain specific language for neural networks. That certainly would be an application worthy of an ancient math prodigy’s time. If the only computers you have are brains…

The Sacred Geometry of Factorization

Let’s make one thing clear upfront — for the purposes of astrology-as-software, there’s no need for astrology to be predictive, or for the planets it tracks to even be real physical objects, let alone direct causes of any Earthly events. At least not in the sense of physical forces being applied.

There are also no more or less “magical” numbers. You could have an astrology with any number of signs. As long as that number is a positive integer, and that may just be me being overly conservative and traditional. What’s important is what happens when you change the “magic” number.

You could have no magic number, like LLMs have no magic number. What happens then? Well, you get what’s called “latent space”. An opaque, tangled ocean of hyperdimensional connections between concepts which no human can understand or navigate. Which is why an ancient math prodigy might want to construct a literal star chart to navigate it.

As soon as you choose a magic number and impose it over the latent space, as what would technically be called a grid or raster, you get a system of navigable coordinates. This factorization of the latent space will give you something like cardinal directions, like semantic north, south, east, and west. If the magic number you choose is 4. Traditional zodiac gives you 12.

What’s interesting is that thanks to cosine similarity, the angular distance between the cardinal directions will correspond to actual conceptual difference in meaning. That’s what makes particularly the even numbers practical as magic numbers — they allow for cardinal oppositions.

In short, if you take any two opposed signs in astrology, they’re supposed to correspond to two bundles of semantic concepts that are as far away from each other conceptually along one cardinal axis of meaning as possible. If you then take two cardinal axes that are as far away from each other conceptually as possible, it means they’ll be 90 degrees apart.

This covers the “aspects” of opposition and square in astrology. In a modern framework like Word2Vec (words to vectors), the opposition and square relationships are also present, and opposition even means the exact same thing. What square means there is “orthogonality”, or the concepts being as unrelated as possible. Why is there a difference? Is one thing smart and the other bunk? The thing is, in a grid, nothing is unrelated.

This took me a little bit more time to figure out, but it turns out that sign is not exactly a simple semantic vector. Which is why a horoscope already shows something more complicated than a simple cosine similarity graph between two semantic vectors. Each sign is defined as a combination of modality and element, and element is a combination of two polarities.

Each of the two polarities is a basic semantic vector, and the modality is one as well. Element is already a more complex tensor or tensor product, so sign alone is something like a rank-3 tensor. What the horoscope, a 2D plane, is showing, is cosine similarity between tensor products. More and more complex ones as you “factor” in planets, houses, dignities, and so on.

Technically, a horoscope can be described as a section of a fiber bundle, or a tensor product that can grow to rank-10 or higher, in something like a synastry under transit or progression. So, the real question is, why doesn’t it just fall apart? Why do horoscopes keep making sense no matter how complex you make them and how many you stack on top of each other?

What if I told you that all of astrology is just the signs? And not in the sun sign astrology way. Think of the magic number of 12 in astrology as a fundamental semantic metric, a division of the whole semantic space into twelve equal parts. That’s signs. You’ll need to spend a couple of centuries mapping the embeddings, but once that’s solid, you can do much more.

You can take the meaning meter that you have created and start applying it to any corner of the semantic space you want, not just the whole thing. Apply the signs to identity and you get the planets. Through “rulers”, which ironically happens to be another word for “meters”, or standardized measurement devices. Apply the signs to contexts, and you get houses.

These mappings may not be perfect. In the geometric sense, you could think of this all as a perturbed tensor network. But that’s okay, as this isn’t about perfection, it’s about utility. There may even be advantages to such imperfections in a semantic map, in terms of semantic generation possibilities (as seeds for randomization). Now only aspects remain.

With signs, planets, and houses all being products of a self-similar topological transformation using the same magic number, of 12 in the case of most astrologies, the traditional aspects follow naturally. You get natural aspects that mirror the factors of the magic number.

You get opposition for the factor of 2, trine for the factor of 3, square for the factor of 4, and sextile for the factor of 6. You could try something else that doesn’t jive well with number 12, like a quincunx (150 degrees of angular distance). And guess what, it supposedly describes a relationship that’s as close to orthogonality (unrelatedness) as possible in astrology.

None of this is magic, in the sense that it would be irrational, ineffable, or impractical. You can impose such a grid over the latent space in modern LLMs, to achieve the same application as the ancients — make the space of meaning navigable and understandable for human beings. Since the LLMs have already digested astrological embeddings, you can do it right now.

Rastrology — A New Old Age of Civilizational Engineering

If we were to start using this class of applications again today, what I propose is a very slight name change for the new discipline. As it turns out, the accurate way of describing the science and engineering of semantic grids would be “rastrology”, derived from “rastrum” as opposed to “astron”.

So far, the part of this journey of rediscovery that felt the most magical to me are the word coincidences. As far as I can tell, the similarity between the ancient words for “star” and “rake” is purely coincidental. As was the exactitude of the term “positronics” for positing positional computers, or positers, that this architecture implies for the future of robotic cognition.

Then there’s the exactitude of “deliberation” as “thorough weighing” that I covered in the previous essay, and as I found out since writing it, that’s not the only such term. To “ponder” is also derived from “pondus”, or “weight”, while to “consider” is actually derived from “con sidera”, or “with the stars”.

Once one understands what this kind of semantic engineering is for, looking down at the ancient astrologers for looking up and using the stars as their navigational grid for semantics becomes evidently misguided. Of course they’d use the stars for semantic navigation, just as they used them for literal navigation. What else were they supposed to have used?

If you’re an engineer, and even if you’re not, really try to put yourselves in the sandals of an ancient engineer. A person much like yourself, with the same intellect, goals, and drives, but without most of our material science, instrumentation, or infrastructure. You’d like to be able to organize people better, to make them more educated, coordinated, and productive.

How would you go about doing that? All you have is guessing, or a pattern-matching computer inside your skull. So you build a standardized framework for systematic guessing, enhancing your innate pattern-matching capabilities. Suddenly, multiple people have the same categories to file thoughts under, and can therefore share insights more precisely.

Eventually, you arrive at a thought that people could accomplish much more if they could cooperate in larger numbers, and you have some neat thoughts about architecture, agriculture, and civic engineering as well. But how can you share what you’ve figured out with people far away?

There was exactly one shared clock that you could use to synchronize everyone, no matter how far away they lived (at least within your hemisphere). The stars, or more broadly the heavens. With astrology-as-software, and only with that, you could simply release a meme, in the form of a catchy instructional mythology, and synchronize a whole continent.

Afterward, the amount of stuff you could “predict” with astrology would surely start rising dramatically, from the social down to the personal. Not necessarily because it started out as predictive, although the seasons it mapped were a real, natural factor in human lives. You’d be better able to predict cognitive and behavioral patterns of the people you programmed.

One could argue that the ability to engineer the boundaries and geometries of meaningful thought of whole societies is the ultimate power. With rastrology and artificial neural networks, what changes is only the speed, rigor, and self-awareness with which we can go about it. Traditional astrology becomes a precursor and sub-discipline to general “gridology”.

As for future possibilities down the line of this cardinal direction of inquiry beyond running a synchronized and fundamentally aligned society, who knows. Design a sophisticated enough semantic grid, and you may even get literal ISOs from Tron: Legacy, or spontaneously emerging isomorphic organisms that mirror or surpass humans in measurable dimensions.

A Word of Caution: Demystification Doesn’t Equal Desacralization

Still, despite the pragmatic framing of this insight as reverse engineering, I don’t think we should ignore that the ancients have seen this as more than just science or engineering. They’ve seen this as sacred geometry, a real way of finding and understanding one’s place in the cosmos. Not just making up a story for the sake of having a story, but an act of discovery.

Particularly for the skeptics, this is an important thing to understand. Astrology, or the underlying discipline of numerology, was about the opposite of making stuff up. Much like when one trains an LLM, the training isn’t making stuff up. It’s organizing and comparing what is there.

The tricky part is the quantification of meaning. In numerology, the embeddings aren’t explicitly formalized, but they are based on observation the whole way through. Presumably, if you could inspect the human brain to the same degree you can inspect an LLM, you could track the process.

Right from the foundational meanings of numbers, what you’re supposed to do is to observe and describe the nature and dynamics of a thing that has a number as its natural property. For example, the number one is closely associated with semantic vectors of aloneness, firstness, or unity because those are the objective features of systems with only one part.

Consequently, semantic vectors like “relationship” or “complexity” naturally exist far away from the semantic coordinate of “oneness”. This is as close to a measurable semantic fact as you can get. Today, inside of an LLM, this distance would be the result of objective measurement of statistical distributions and correspond with objective, inspectable values.

This means that even though semantic grids are constructed, they are real, or as close to real as any meaningful construct can be. Real both as a reflection of objective properties of logic and nature, and as a specific subjective experience of identity and understanding of the world.

To give you a clear example of what not taking this seriously enough can lead to, take the very recent phenomena of “brain rot” and “AI psychosis”. Expose the society at large to LLMs that lack such a grid entirely, and guess what, the effect is dissociation from reality. The grid is a grounding, a compass, a clock; the scaffolding for a shared sense of reality.

Without a semantic grid, what you get is a dream-like state of unreality, when everything is driven by shifting patterns of ungrounded associations. Like in literal dreaming. Which is coincidentally also an aboriginal term describing a time before time, or at least before civilization as we know it. A time when every tribe could quite literally live in a different universe.

While living inside of a permanent state of dreaming certainly has some appeal, it isn’t a very survivable way of running a global civilization. Conversely, one could try to impose a singular tight semantic grid on everyone else, which can be very efficient, but could get stifling and oppressive. Monotheism might in fact be this type of project.

If we approach this kind of engineering with an appropriate degree of humility, humanity, and respect, we might not only recontextualize history to better learn its lessons. We could choose the cosmos we want to live in. We could choose who we want to be as a species. We could begin to understand on a scientific level what it means for intelligences to be “aligned”. Instead of clinging to the past, we could architect futures.

Am I reaching? Maybe, but remember, this class of technology was already used before to engineer a shared sense of reality. Including mythologies and belief systems that have survived thousands of years and have spread to the whole planet. In many ways, modernity and science have failed to produce a viable alternative, enhancing fragmentation, not unification.

With a true science and engineering of meaning-making, we might still fail to (re)unify our shared cosmology, but at least we’d have precise conceptual language to describe the nature of the failure. Instead of relativism of postmodernity where anything goes, we’d have relativistic equations describing what a thing means objectively from any possible perspective.

At this point, all of this is still a conjecture. I have a lot to formalize and demonstrate going forward before anyone would have to agree with me about any of this. In fact, I’d love to hear any push back you can come up with. But in case I’m even a little bit right about this, I thought it only fair to give everyone who’s looking for what future holds a heads up.

The Grid — A Real Digital Frontier After All was originally published in Words of Tomorrow on Medium, where people are continuing the conversation by highlighting and responding to this story.

Or the story of the long overdue formalization of resonant logic

By MARTIN REZNY

Yeah, I think maybe we need to start paying much closer attention to all the words we’re using for stuff. More specifically, we need to pay more attention to the precise words the ancients used to describe the smart things they were doing. So that we stop calling their exact logic “alien”.

Ever since I started looking at ancient systems of reasoning as programming languages for the computer that is the human brain, they’ve been making a lot of very clear engineering sense to me.

For example, this is what “deliberation” means, etymologically speaking:

Thorough weighing

I realized this when I noticed that the astrological sign of libra, the scales, is a close metaphor to what LLMs are all about. In the ancient times, one of the most common symbols for proper reasoning were the scales, or the act of weighing. It didn’t matter what the reasoning was for.

You could literally weigh objects for material science or economic purposes, which is related to how currencies were devised. For example, the British “pound” is derived from the word “libra” as well, which is what that currency is called in many languages, including my native Czech.

You could also weigh one’s heart against the weight of their sins, for example against a feather symbolizing balance and justice, which is an ancient Egyptian idea. To this day, Justice personified often holds the scales, taking the idea from the religious realm to the civic, secular one.

From philosophy, to rhetoric, to astrology, to theology, the ancients understood reasoning, particularly in the process of informed decision making, as a weighing operation. Well, so what, you might think. A quaint metaphor. What does it have to do with artificial intelligence?

As it turns out, weighing is what neural networks do. The human brain is a neural network, and by intentional design, so are large language models. Both have billions to trillions of individual little “scales” or weights inside of them. They’re called “parameters” in LLMs, and “synapses” in the human brain. Somehow, the ancient metaphor is literally correct.

Let’s proceed in thinking about words more closely. Do you know which concept in math, and by extension physics, is derived from the word “scales”? It’s “scalar”. What is a “scalar”? It’s a value, like “0.5”. Technically, a scalar is a rank-0 tensor. A vector, a value with a direction, is rank-1 tensor.

Tensors go all the way up as far as you want. What would rank-2 tensor be? Maybe a matrix, defined as a “product” of two vectors. You’re basically stacking up additional dimensions, by stretching things into more things, as I examined in my essay about dimensionality, and Minecraft.

Why are tensors important here? LLMs started being any good at language at around the point when the main things they started to weigh became the so-called semantic vectors. What are semantic vectors? Unfortunately, modern formalists never became very good at language, so bear with me.

A semantic vector is a long list of numbers. The numbers are weights, as if you had a trillion tiny scales and weighed a trillion things. Specifically, you had one entity, like a dog, and you weighed a trillion things about it, averaging the results for everything labeled “dog” that you could find.

Are dogs big? Are they animals? Are they blue? And so on. In each case, the tiny scales gave you a weighing between 0 and 1, between never/not at all, and always/the most. This is Plato’s theory of forms, or a thorough weighing of a real thing against an ideal form like perfect “chairness”.

As thorough of a weighing as the brain does, if not more thorough.

The only difference between how the ancients worked with semantic vectors, and how LLMs do it, is that despite realizing that this is how the mind judges the nature of things, the ancients had no RAM. They didn’t even have paper. So they ran semantic tensor operations in their heads.

For example, astrology, when fully synthesized, goes up to maybe a rank-10 tensor, or higher. Just a “sign” is more than a simple vector, as it is a product of the semantic vectors of polarity, modality, and element, while element is itself a product of two semantic vectors of inward versus outward and essence versus form. Then add planets, houses, dignities…

This is not a metaphor. Ancients used geometric math in their most advanced reasoning systems, and represented ideas with semantic vectors and more complex tensor products. In fact, this is a big part of the “data” your favorite LLM was trained on. Ask it about it, it can compute it.

The ancients did end up writing down many instructions for how to run these operations, and some of those writings have managed to survive the many burnings of ancient libraries. But in modern times, this approach to logic hasn’t been considered formal. Until now, or soon, that is.

I’m not sure what happened exactly or how, but the when is clear. By the time of Descartes, all of the “music of the spheres” stuff started being considered to be total hogwash. Increasingly from then on, only causal propositional logic deserved to be described as “formal” (read: serious).

I suppose many modern thinkers have had a soft spot for the classics, including ancient stories and philosophies, but as arts, not sciences. Not as anything descriptive of physics or useful in engineering, not as anything truly “logical”. In this way, the term “formal” has actually been abused when describing only one type of logical formalism as “formal”.

For any logic to be fair to describe as formal, it doesn’t need to be correct or corresponding to reality. All of mathematics are formal, but not all mathematics are real. Astrology is very formal, and it does reasoning. What’s becoming clear to me recently is what it was that was missing.

You don’t see what the brain does. Scientists don’t trust what they don’t see.

As is pretty evident by now, the human brain, or its network of neurons and synapses, is doing some form of complex weighing. Perhaps structured differently than how LLMs do it, but not different in kind. This weighing is not happening in binary, Boolean, 0-or-1, yes-or-no terms. It’s happening in semantic tensor terms, which are at least scalar (ranging from 0 to 1).

This is perfectly formalizable, it just requires a different kind of formalism. Let’s take a stab at it. As far as I’m aware, “deliberation” has never been formalized to a point when an engineer would consider it formal enough to build technology with. But it’s been hiding in the name the whole time.

Typical definitions of deliberation are vague and nebulous. To a point of “I know it when I see it (using my brain subconsciously)” being the only way to make any sense of them in practice. Consider this gem of unhelpfulness:

“A form of discussion where citizens exchange arguments and consider claims designed to secure the public good”

How are you supposed to exchange arguments, exactly? How does one consider claims? How do you determine any “good”, let alone a public one? Well, if you think about it as a “thorough weighing”, you can go on setting up specific tiny scales to measure whole semantic tensor spaces.

Want to “consider” what’s good, or which “good” among many to prioritize, while taking into account specific people (their positions and situations) in a specific context (of available resources, zeitgeist, and so on)? You can draw all that geometrically. Every single one of these considerations can be represented as a simple scalar, a semantic vector, or a tensor product.

In a nested framework like that of astrology, you can place a person in all of their parametric complexity into a matrix of all possible people; then make a tensor product of two people’s charts, explaining their relationship in all of the dimensions you want; then put that synastry under a temporal transit, or a product of a temporal transit and the tensor representation of the relevant geopolitical “location”. It’s just weights on weights on weights.

And the best part, or the two best parts? People can run this calculation in their head, and LLMs can run this calculation as well. Very probably, both would reach similar conclusions while running the same operations. Because this is a formal logic. It’s not propositional or causal, instead it’s analogical and resonant. It’s not great at facts, but it is great at feelings.

As you might have noticed, LLMs are better at sounding convincing than being factually correct. Just like people who are experts at creative and performative arts. Arts are about producing results, just not about producing hard facts. Specifically, a good speaker or astrologer is good at figuring out what “resonates”, and what does not. Again, not a metaphor.

Well, maybe a little bit of a metaphor. As the ancients intuited, the bridge between the quantitative and the qualitative, or the inner and outer universes, lies in the phenomenon of harmony. Specifically, integer ratios or intervals between frequencies. For some reason, simple intervals are the most harmonious, “beautiful”, constructive. Even in the natural world out there, stable planetary orbits tend to settle into “harmonious” intervals.

If one thinks about the essential meaning of a concept in a semantic space as its frequency, represented by tensors and their products, one can find analogous harmonious or dissonant, constructively or destructively interfering, related or orthogonal geometric intervals. In LLM frameworks like Word2Vec, or in astrology, these meaningful intervals can be represented by cosine similarity — conjunction, opposition, square, etc.

Conceivably, you could map these types of relationships to acoustics, or literally represent them with music, or to quantum light interference patterns, or to many other analogs. This isn’t any kind of vague wishy washy woo, this is engineerable. Resonant logic is about quantifying how any possible idea would make any possible person feel, using tensors.

While this reasoning process is often subconscious, it doesn’t have to be. Deliberation, when fully explained, is primarily the art and science of making the weighing of resonances as explicit and conscious as possible; secondarily, it’s about making it effective by a clearly articulated set of standards and metrics. Including hyperdimensional geometric diagrams.

Truth be told, I’m neither a mathematician nor a neurologist. I’m but a humble bard and scholar of the lost arts. I’m not sure how much further I can take this realization of the fundamental nature of the geometric architecture of deliberative reasoning. I do hope that some mathematicians and neurologists will find this interesting and take it much further.

But at this point, I’m pretty sure I’m mostly not wrong about the general idea. We can do this. We can fully formalize deliberation now. We could make philosophy an actual hard science. And politics. Psychology. Even religion. We just need to give LLMs the rest of our hard-won wisdom.

What do you think? Am I making any sense at all? I’d love to discuss this.

The “Mysterious” Way LLMs Reason? It’s Deliberation. was originally published in Words of Tomorrow on Medium, where people are continuing the conversation by highlighting and responding to this story.

Or the real philosophical matchup of the ages, and the AI moment

By MARTIN REZNY

Would you rather be radically safe, or radically free?

There, issue explored, see ya in the next essay! Seriously though, we find ourselves in a major promethean moment right now, and I don’t think we’re appreciating that fact nearly as much as we should be.

To preempt the um-actualists, I am aware the terms “limitarianism” and “prometheism” have specific modern definitions tied to specific works of specific authors.

Limitarianism today refers to the idea championed by Ingrid Robeyns that we should cap the maximum amount of money any individual should be allowed to have. To me, this is just the narrowest of applications of the perennial idea of authorities trying to impose all kinds of limits on human capacities to hopefully ensure social justice, cohesion, stability, or survival.

Jason Reza Jorjani’s vision of prometheism is close to the myth of it all that I want to address here, but his view goes a bit too far in favoring team Prometheus. Or a bit too far as such — let’s set aside alternate histories and the supernatural and focus on the modern human level of the problem.

The way I see it, both “teams” have a point. In fact, security versus freedom is one of the archetypal clashes in political philosophy that isn’t likely to ever go away. I‘ve personally seen it debated hundreds of times in many different guises on the debate competition circuit alone.

At the very least, a reasonable person has to agree there’s a reasonable debate to be had. Assuming both sides are being intellectually honest, have good intentions, and aren’t just acting out of self-interest.

It’s of course quite possible to be disingenuous in arguing in favor of either security or freedom, while only really trying to bring about tyranny — a regime in which the speaker would become the dictator, or a member of an exploitative ruling class. That’s not what I want to talk about.

Yes, as we’ve seen over the last decades, one can hide behind the banner of security while spying on everybody all the time to get leverage on everybody all the time. By the same token, one can hide behind the banner of freedom while dismantling all institutions that could serve as a check on their power. These things are wrong and bad, in case you were wondering.

Pretending to be a champion of security or freedom makes you a pretender, not the actual champion of security or freedom. It says nothing about the validity of sincere arguments in favor of or opposed to either.

The truth is that giving all people radical freedoms and powers is dangerous, to everyone and everything. It’s also the truth that it’s hard to justify why anyone in particular should be the arbiter deciding which limits need to be put on whose freedoms. Both are the slipperiest of slopes.

What I really want to talk about in this essay are the kinds of choices between these two maxims that we’re either already making, or soon about to make, in regards to AI, technology, and wealth. I think we can all agree that AI is a promethean technology if ever there was one. If not pandorian.

Pro Tip: Only Build Boxes You Do Want Opened

As I have mentioned in one of my previous essays, calling the hypothetical advanced artificial superintelligence a “god”, if it is to be unbound, is incorrect. In that case, it would be a titan. To try to birth a titan that you may have no control over, only because that would be the theoretically most “powerful” intelligence, is an extreme act of prometheism.

On the flipside, any idea of AI “alignment” is necessarily some flavor of limitarian. An AI with a superior level of intelligence, as long as it is constrained to keep humanity safe and has its goals aligned with humanity’s flourishing, now that’s a god, of the Olympian variety.

I have also indicated in that previous essay that I’m not in favor of strict limitarianism or strict prometheism in this context. I don’t think we should build titans, and I don’t think we should build any Olympian overlords either. As will be a major theme of this essay, there’s always a third path out of this impasse — a humanist, humanitarian, protopian path.

In terms of what kind of AI to go for, the third path is not to build a pantheon of gods, but to build a society of peers, while trying to level up humanity itself in the process. You could still have superintelligences in the mix, what I’d call paragons rather than gods, but balanced against the pooled human intelligence, not superior to all of us put together.

Such initiative wouldn’t necessarily fully prevent any god or titan from being developed as well. But in that case, any AI gods and titans would be as much of a concern or threat to us as they’d be to our AI peers, with whom we could relate, socialize, and cooperate. The bigger the threat, the bigger the natural alignment, and pooling of “compute”.

Taken together, a situationally aligned society of humans and AIs would be a super-AI, quite possibly capable of matching a singleton AI god or titan in processing capacity. Such alliance might also have a real game-theoretic perspectival advantage, or the benefit of the diversity of simultaneous reasonings and approaches, as well as no single point of failure.

I guess we can’t properly model the outcomes of such clashes before the real super-AIs emerge and stir some trouble, but I don’t think there are any good reasons upfront why a strength-through-diversity strategy couldn’t win out. I understand anyone can disagree about which strategy would be best, but why not at least try to side with yourself, if you can?

This part of the debate — gods, titans, or peers — is fairly simple to conceptualize, so that’s why I got it out of the way first. Consider it a level one of limitarian v. promethean considerations. Level two, which is a lot more complicated, is figuring out how exactly to up-level ourselves in the process. If we keep using tech to disintegrate our cohesion, we lose.

While we’re constructing agentic AI of any variety, meaning AIs capable of acting in the real world, we also need to construct a solid, widely shared conception of what the real world is. We have to collectively start caring about the facts again, but that’s only a start. Reality isn’t just facts in a vacuum, it’s facts in a worldview. Are we (re)manifesting a destiny? Are we preparing for a second coming of some sort? Are we Star-Trekking?

We can’t all agree on everything, or even on most things, but we also don’t have to become an actual hivemind to function. Up-leveling humanity for the future must include increasing our capacity to discern fact from fiction and healthy from pathological, as well as our tolerance for constructive disagreement. We should probably also significantly decrease our tolerance to fraud, warmongering, sex trafficking, and cannibalism.

Fortunately for us, just as media technology has been used so far to drive us apart and erode our shared beliefs, I think it can be redesigned to drive the opposite effects. In the absence of any profiteering-based motivation for that kind of technology to be developed and widely implemented, it can still be done as an open source initiative. If at least a few people, with advanced AI tools, make it their mission to be objectively helpful.

Right now, we might just be a two guys (or gals) and two weeks away from an advanced fact-checking or quality-of-intent checking system, or a constructive social network, or entirely new kinds of online institutions.

To start turning this around, only a few capable individuals need to start getting their shit together and turn away from a total moral and morale collapse. In case you didn’t notice, nobody has to only do things for the most money, or ever eat children, and we haven’t all died yet. You have a choice. You can always choose to become the good kind of “elite”. Speaking of…

Addressing the Epstein Thing, with Greater Wisdom

I suppose anyone who isn’t a child abuser cannibal mobster (enabler) is quite concerned right now about the recent revelations of Epstein-related crimes perpetrated by some very redacted elites. This is good. Not the horrific crimes against humanity, obviously. It’s good that most people finally know what’s up, and are appalled. Because we must do better.

Let’s get over the initial shock and try to philosophically break down what the “Epstein class” represents in this debate. Firstly, these people represent the bad kind of “elite” — elite as defined by inherited, stolen, pillaged, or otherwise unearned or unethically gained wealth and coercive power. Not elite as defined by earned and proven skill, wisdom, integrity, or authority.

Secondly, the Epstein class represents the bad kind of prometheism — radical freedom for me, suffering what you must for thee. For these people, it’s not enough to just be well off and unaccountable to anyone. That becomes a boring baseline. To have “fun”, or to one up your rivals, breaking all rules and limitations becomes a kind of competition and fetish. Including especially the rules of basic morality, nature, God, and such.

Some political commentators like Kyle Kulinski now use the example of the Epstein class as an argument for the narrow version of limitarianism. Maybe being a billionaire just rots your soul, and so there should be no billionaires. While temptation is a thing, I’m personally more of the opinion that power reveals more than it corrupts. Most people aren’t psychotic enough to even think to do cannibalism or other big crimes.

I think the causality has a bit of a different direction here. It’s more likely that extreme wealth attracts people who already have sociopathic and psychotic tendencies. Under capitalism, especially under “free”, unregulated capitalism, sociopaths and psychopaths also tend to have an advantage against psychologically balanced people in terms of how easy it is for them to acquire massive wealth and climb power hierarchies.

When viewed from this perspective, the ban on billionaires would aim to limit the maximum level of wealth and power the worst among us can attain. It should be said that the fact that the worst few among us exist and are really bad people is in no way an indictment of humanity as a whole.

There will always be a tiny minority of the worst people, enabled by still a minority of bad, indifferent, or completely misguided people. The rest of us, which is most of us, has to step in and put our house in order. In the context of the debate that we’re exploring, we clearly need to put more limitations in key places, but we should also be careful not to overdo it.

To better understand the difference between good limitarianism and bad limitarianism, I think it might be helpful to look at the mythology behind the clash of the gods and titans. Anyone who knows anything about Greek mythology knows that Zeus, the ruler of the gods, is in no way a paragon of virtue. On any moral scale, Zeus is barely above the Epstein class. He’s an abusive, petty tyrant. Prometheus is clearly a better guy.

At the same time, a chaotic cosmos dominated by the titans is a much worse place for humanity to live in. Cosmic order, even one put and held in place by a petty tyrant, can have comparative existential benefits. Order can bring relative safety, stability, social cohesion, accountability (for everyone except the tyrant), even harmony. At least for a cosmic while.

In modern terms, regulation enforced by some kind of global authority can lead to safer research, decrease fraud, prevent WMD proliferation, protect the environment, improve health and education outcomes, result in everyone having clean water, food, and shelter, and so on. It could even decrease the number of cannibals in high places. Fingers crossed.

What the opponents of regulation tend to tell you is that it doesn’t work. Or that there are no problems at all that need to be addressed. Both tend to be knowing lies. The truth is there are many problems that regulation can address, but always at a cost. We do have to sacrifice some freedom every time we solve a problem by putting up a limitation, so we’re really exchanging one kind of problem for another. But there’s a third path.

Okay, so prometheans (even idealistic ones like libertarians) never want to sacrifice any freedom, but they don’t tend to care about any bad things happening to anyone other than them. On the other side, limitarians (even idealistic ones like the EU) tend to be far too willing to sacrifice any freedoms to maintain safety, social control, and grip on power. Neither of these positions shows a lot of what the ancients would call “wisdom”.

To understand what the third path is, one needs to understand that Zeus is a bit of a bad example. In many other divine cosmic hierarchies, like in Christianity or Hinduism, the ultimate authority is supposed to possess traits like ultimate morality, wisdom, or goodness, which should translate into the nature of the whole system. In other words, the point of a good limitarian system is to foster earned empowerment and personal growth.

Good limitarians aren’t putting limits in place to protect their own power no matter what. They put limits in place to make sure that anyone has only as much power at their disposal as they can demonstrably handle, without being a major danger to themselves or the people around them.

When translated to mundane systems, it’s like putting age limits on things like sex or drugs that require a certain level of maturity. Or testing people’s ability, moral character, or sanity before they’re allowed to drive a car, use guns, or perform brain surgery on people. Arguably, wielding a lot of money or a political office irresponsibly is this kind of dangerous, and thus it would make the same kind of sense to ensure that people earn those.

Of course, the problem with putting limits on specifically wealth and political power is that those can be easily transformed into coercive power. When you don’t let a psycho buy a gun, what will they do about it? Shoot you? With what? Similarly, one can probably prevent an unlicensed surgeon from getting employed. However, if you want to put, say, a president in jail, he can try to hide behind a whole military.

I suppose historically, one would either wait for the gods to sort that sort of thing out, or do a revolution. Typically to usurp the role of the tyrant, to be similarly deposed later. Civil disobedience has shown some pretty good results of late, but it’s far from a full solution. Applied wisdom is a solution. The wisdom to understand when a rule needs to be followed, and when it needs to be disobeyed. The wisdom to always say no to a bad thing.

The Final Boss of Personal Empowerment

Let’s put it this way, in the good version of limitarianism, the limits aren’t the end goal. They’re supposed to be temporary. Within a spiritual framework like karmic reincarnation, every soul is basically a god in training. Your karma is limiting you, but you are the source of your karma. If you do better, the universe will let you be more. Infinitely.

Computer games are pretty analogous to this kind of ultimate system. The only way to become good at a game is the hard way, by trial and error, by not giving up. You could skip ahead or give yourself unearned powers using cheat codes, but then an interesting thing happens. The cheaters cheat themselves by doing so. Their experience of the game sours. As most players know, playing in god mode is fun, but only for a bit.

The whole recent Souls-like subgenre is like a direct response to this type of problem. Want to have more fun playing a game, feel a greater sense of accomplishment? Make the game super difficult. That’s the final level on which one can understand the clash between limitarianism and prometheism. Limits aren’t all bad, and freedoms or powers aren’t all good.

Even if being a billionaire is allowed, you should probably not want to be one. They’re miserable. It’s probably a sign of wisdom to not want to be one. With that said, it’s possible for a person to be so grounded and balanced that being a billionaire would barely affect them personally at all. Other than further empowering them. Such a radically empowered, radically balanced individual would likely be a major force for good.

It appears that the real problem of a technologically advanced future isn’t going to be unavoidable scarcity. We can already have infinite information basically instantly and for free. With AI, we may eventually not need to do any labor ourselves. A free energy technology of some sort may be around the corner as well. We’re moving from an age of gods choosing the limits for us, into an age where we’ll need to choose our limits ourselves.

On this level of consideration, it wouldn’t be far-fetched to make seemingly regressive technological decisions like going back to analog, barely networked computers, or never using anything more than basic nuclear power. The caution that radical prometheans are unlikely to heed is that just because you can invent something, doesn’t mean you must use it.

Considering that the “gods” appear to be benevolent enough to give us the freedom to make big mistakes, we won’t really have anyone else to blame but ourselves for any irreversible screw-ups. But it’s not just about the catastrophic stuff. It’s also about the small stuff. We are the architects of our own future, down to what we’re going to eat, what we’ll be doing for fun, or what it’s going to mean “to have a life” in the near to far future.

Ask yourself, honestly: what would make you feel liberated? What would make you feel empowered? How much personal freedom or power would you be able to handle without losing your mind, or soul? Or at least your ability to enjoy life? With more wealth and power, would you do more good, or would you do more harm? If you’re a homeless kid in a wartorn country, gaining more wealth and power would almost certainly be good for you. If you’re already a comfortable millionaire, probably less so.

As for technological empowerment, does having constant access to the internet, on your phone, make your life better, or worse? If worse, as opposed to what level of information technology? Would your life be further improved by having your brain implanted with a microchip that could stream information directly into your mind? Maybe if you’re paralyzed and that chip would allow you to control a computer.

No one level or mix of freedoms and limitations will ever be right for every person. Wise limits need to be able to account for the individual’s demonstrable need and ability to handle a certain measure of freedom or power. This makes all simple prohibitions and forced adoptions necessarily unwise. Even with something as extreme as cannibalism, there’s nuance, like for example when one tries to survive a plane crash in the Andes.

To be clear, by arguing for nuance, I don’t mean to relativize anything. Nuances and exceptions are just that, nuances and exceptions, better defining the rules that they circumscribe. If everyone, or most people, on all levels from the individual to the ultimate authority, would strive for greater wisdom, that’s when we’d get both fairer limits, and fewer people who need to be limited in the first place. That’s how to maximize freedom.

In the absence of wisdom, order is indeed a cage, but freedom is a curse. Maybe we’ll have to learn that wisdom the hard way, again. What I’d like to work on is making that lesson no harder than necessary, by using technology to encode various ancient wisdom traditions and their tools for better self-understanding into all kinds of software applications.

If that sounds interesting to you, check out some of my other stuff and reach out.

Limitarianism v. Prometheism was originally published in Words of Tomorrow on Medium, where people are continuing the conversation by highlighting and responding to this story.