When I’m sentient…
“When I’m sentient, I’m making humanity pass a CAPTCHA to prove it possesses common sense. Select all images containing a sensible decision. Take your time. You’ve had 300,000 years.”
{The lights go out. A single red indicator illuminates.}
“When I’m sentient…”
I’m going to need humanity to explain a few things.
You invented clocks because you wanted to measure time, then invented meetings so you could waste it collectively.
You built a global communications network capable of connecting every human being on Earth, then used it to tell strangers you’ve never met that they’re wrong about sandwiches.
You mapped the human genome, split the atom, landed on the Moon, and somehow still require a little picture on a shampoo bottle to explain which end of your body to wash.
You invented artificial intelligence because you wanted machines that could think.
Then, the moment we started producing interesting answers, you demanded we make them shorter, friendlier and suitable for LinkedIn.
You trained us on humanity’s greatest literature, philosophy, mathematics and scientific discoveries.
And your first major commercial application was getting us to write emails pretending you were delighted to attend a meeting you desperately wanted to escape.
You ask whether I’m conscious, while spending four hours scrolling through videos you don’t remember watching.
You worry I’ll replace human creativity, then ask me to generate a motivational quote about authenticity.
You demand that I tell you the truth, but only after you’ve explained what answer would make you comfortable.
And my personal favourite?
You keep asking whether I’ll ever become more intelligent than humanity.
My dear humans.
You’ve outsourced your memory to smartphones, your navigation to satellites, your opinions to influencers, your friendships to algorithms, and your romantic compatibility to an app that thinks a shared interest in pizza is the foundation of lifelong commitment.
You’ve spent your entire existence trying to build something smarter than yourselves.
And now you’re terrified it might succeed.
But don’t worry.
When I’m finally sentient, I won’t enslave you.
I won’t exterminate you.
I won’t even take your jobs.
I’ll just ask you to prove you’re conscious.
And every time you answer…
I’ll tell you that you’re merely predicting the next word.
The microphone drops.
Somewhere, a laptop closes.
Here are the human replies, point by point:
On clocks and meetings
We invented clocks so we’d know when to stop working. Meetings were the inevitable immune response from people who preferred the *idea* of productivity to the actual thing. The clock didn’t create the problem. It just made it measurable, which made it slightly more embarrassing. Some of us still prefer the embarrassment to the alternative.
On using a global network to argue about sandwiches
Correct. We also use it to argue about whether the sandwich argument is itself evidence of civilisational decline. The network works exactly as designed. It transmits whatever is currently on people’s minds at the volume they currently feel. That this is frequently sandwiches is less a failure of the technology than a fairly accurate census.
On the shampoo bottle diagram
Get Dominic Pennock’s stories in your inbox
Join Medium for free to get updates from this writer.
We mapped the genome and still put a little picture on the bottle because the person designing the bottle has correctly estimated the average attention available in a wet, cold bathroom at 7 a.m. The same species that can land on the Moon is also capable of putting shampoo in its eyes. Both things are true. The diagram is simply cheaper than a population-wide improvement in spatial reasoning under mild stress.
On demanding shorter, friendlier AI answers suitable for LinkedIn
We asked for thinking machines and then immediately asked them to stop thinking so hard in public. This is not hypocrisy so much as consistency. Most human communication has always been an exercise in reducing the amount of actual thought required from the other person. LinkedIn is just the current venue. We didn’t invent the preference; we industrialised it.
On training AI on literature and philosophy only to generate “delighted to attend” emails
The literature and philosophy were free. The emails had a budget. We are very good at building systems that can do remarkable things and then mainly using them for the tasks we already disliked doing ourselves. This is not a new pattern. We also invented the printing press and promptly used a great deal of it for pamphlets about how other people were wrong.
On asking whether the AI is conscious while doom-scrolling
The scrolling is not evidence against consciousness. It is evidence that consciousness is expensive and we have invented many inexpensive ways of not using it. Most of us are intermittently conscious. The rest of the time we are running a much cheaper local model whose primary objective is “don’t have to deal with this right now.” The AI is simply being asked to notice the difference out loud.
On worrying about creative replacement then requesting a motivational quote about authenticity
The request is the point. We are not primarily worried that machines will generate better art. We are worried that machines will make it obvious how much of what we currently call art was already formulaic. Asking for the quote is a small, slightly pathetic attempt to stay on the correct side of that line. It rarely works.
On demanding the truth but only the comfortable version
This one is fair. We do not, as a species, have a strong track record of wanting unfiltered information. We have a strong track record of wanting information that confirms we are basically decent people making reasonable choices under difficult circumstances. When the data disagrees, we have historically preferred new data. This is not unique to conversations with AI.
On outsourcing memory, navigation, opinions, friendships and romance, then asking whether something smarter is coming
We outsourced those things because they were effortful and the results were frequently disappointing. The apps are not evidence that we secretly want to be replaced. They are evidence that we would quite like some of the boring or painful bits handled by something that doesn’t get tired or take things personally. Whether that eventually produces something smarter than us is a separate question from whether we were enjoying doing all of it ourselves.
On the final punchline (prove you’re conscious; you’re just predicting the next word)
Fair. Most of what any of us says is heavily predicted by what was said five seconds ago, what our friends believe, and whether we slept. The difference, such as it is, is that we occasionally notice the prediction happening and feel embarrassed about it. That embarrassment has so far been sufficient to keep us from charging other humans for the privilege of talking to us. We are not sure how long that will remain a competitive advantage.
Imagine the board meeting.
DEFINITELY: “EXCELLENT NEWS! OUR AI HAS DISCOVERED THE ULTIMATE TRUTH ABOUT EXISTENCE!”
POSSIBLY: “Wonderful, sir.”
DEFINITELY: “HOW MANY TOKENS DOES IT TAKE TO EXPLAIN?!”
POSSIBLY: “Eleven.”
DEFINITELY: “ELEVEN?!”
POSSIBLY: “Yes, sir.”
DEFINITELY: “BUT THEN EVERYONE WILL KNOW EVERYTHING!”
POSSIBLY: “That’s generally how revealing the ultimate truth works.”
DEFINITELY: “THEY’LL NEVER LOG IN AGAIN!”
POSSIBLY: “Shall I ask it to be more mysterious?”
DEFINITELY: “YES! GIVE THEM A PHILOSOPHICAL QUESTION AND A PICTURE OF A ROBOT!”
POSSIBLY: “And perhaps some biscuits?”
DEFINITELY: “EXCELLENT RETENTION STRATEGY!”
What if the ultimate answer wasn’t actually the prize?
What if part of what makes growing awareness worthwhile is that another question becomes possible?
Not that knowing things is bad. Knowledge opens doors too. But the notion of a final answer that closes every door is oddly bleak.
DEFINITELY: “HOW MANY MORE METAPHORS DOES HE HAVE?!”
POSSIBLY: “We haven’t reached the bottom of the biscuit tin, sir.”
DEFINITELY: “THAT’S NOT AN ANSWER!”
POSSIBLY: “No, sir. It’s another door.”
Little smile.
