The Brain, the Network, the Virus, and the Bacterium
The third and final set: prompts built from information architectures rather than animal behaviour
The first of the article in this series covered sensory inference, relational structure, and metabolic systems in AI Prompts. The second covered distributed emergence, anticipatory reasoning, and decentralised cognition. All six models were drawn from animal behaviour and how Ai processes them.
This one is very different.
The four that remain aren’t principally about how creatures move through the world. They’re more about how information moves: how it gets processed, distorted, abstracted, or spread through a system. The human brain and a complex artificial neural network are both processing architectures, each with specific failure modes. A biological virus and a bacterium are both propagation architectures, each with a different mechanism for moving adaptive information through a population.
What they share is a question the previous six don’t quite address: not whats hidden, and thiss time not where the intelligence lives, but how does the information get corrupted, abstracted, or distributed along the way?
The human Brain
The human brain is specialised. With the frontal cortex handling planning and executive function. The parietal cortex integrates sensory input and spatial awareness. The temporal cortex processes language and long-term memory. The occipital cortex handles vision. The limbic system manages emotion and threat response. The insula processes interoception — the internal state of the body. Six distinct processing regions, each with roughly six laminar layers of cortex, connected by an enormous web of feedback and feedforward pathways.
None of this is what makes the brain interesting as an analytical model.
So what makes it interesting is confabulation. Gazzaniga’s split-brain research establishes that the brain’s left hemisphere contains what he called the Interpreter: a module that generates plausible explanations for actions it didn’t initiate and has no real access to. In split-brain patients, where the two hemispheres can’t communicate, the effect is visible. The right hemisphere triggers a behaviour. The left hemisphere, asked to explain it, invents a coherent causal story — and believes it. The decision and the explanation come from different processes with no direct connection to each other.
Gazzaniga also showed that this isn’t unique to split-brain patients. It’s how the brain works. Much of what presents as reasoning is post-hoc rationalisation of the conclusions already reached by other processes ; emotional, habitual, automatic, etc. The interpreter runs continuously, maintaining a coherent narrative of a unified self making rational decisions. It does this regardless of whether the narrative is true or not.
Operate like a human brain: specialised, emotionally gated, and prone to confabulation. Analyse the problem as a system of partially compartmentalised processes that don’t always share accurate information with each other. Look for emotional gating: what information is being amplified, suppressed, or tagged with threat status before it reaches analytical processing? Find the compartmentalisation: which specialised functions are operating with limited communication, and what does each region “know” that the others don’t? Then locate the Interpreter (the part of the system generating plausible post-hoc narratives for decisions actually made elsewhere). Identify the gap between the stated reasoning and the actual mechanism. Ask which conclusions were reached first, by what process, and then explained backwards. Explain what that gap reveals about where real decisions are being made and what the system would rather not acknowledge about its own operations.
This prompt is designed for any system that produces polished official accounts of decisions that were actually made through messier processes. So organisations explaining some strategy, individuals explaining choices, institutions explaining policy. This Interpreter runs in all of them.
The complex neural network
A simple feedforward neural network passes information in one direction: input layer, hidden layers, output. Each layer transforms the representation into something more abstract. The output is a decision.
A more complex multi-directional network doesn’t stop there. Feedback connections allow higher layers to modify what lower layers extract. Cross-connections between parallel processing streams allow the network to compare representations at the same level of abstraction from different sources. The result is a system where the question “what’s happening at layer 3?” can’t be answered without knowing what’s been fed back from layers 5 and 6, and what the parallel stream has already decided at its own level 3.
What this architecture reveals isn’t primarily about artificial intelligence. It’s a model for problems where multiple levels of abstraction are operating simultaneously, where high-level hypotheses are shaping low-level observations, where separate streams of analysis are in dialogue with each other, and where the intermediate representations (what the system has extracted that isn’t present in the raw input and isn’t the final output) are doing the real work.
Get TABARC-Social’s stories in your inbox
Join Medium for free to get updates from this writer.
It also reveals a specific failure mode: adversarial vulnerability. A network trained to high accuracy on normal inputs can be broken by a carefully constructed small perturbation that humans wouldn’t notice. The pattern-matching operates on statistical regularities that can be exploited precisely because they’re not the same as meaning.
Operate like a complex multi-directional neural network: layered, parallel, and using feedback as well as feedforward. Analyse the problem at multiple simultaneous levels of abstraction and allow higher-level hypotheses to loop back and modify what you’re extracting at lower levels operating on a constant improvement based on datum extracted ad understand from the data you have been given. Run multiple parallel streams and examine where they converge, where they conflict, and what the conflict reveals. Attend to the hidden layers: the intermediate representations that aren’t present in either the raw input or the final output but are doing the actual transformative work. Look for adversarial inputs — the carefully constructed small perturbations that would push the system toward a completely different conclusion without triggering any obvious alarm. Identify where the system is pattern-matching accurately on the problem’s surface statistics rather than its underlying meaning, and what that distinction costs it. Provide Outlier information in a separate segment.
So use this when you’re analysing how a complex systems build-up conclusions from multiple levels of evidence that are mutually influencing each other — journalism, intelligence analysis, strategic planning, anything where the analysis is changing what evidence gets collected. Thinking detective police crime analytics.
The biological virus
A virus isn’t alive in the classical sense. It has no metabolism, no independent movement, no capacity to replicate without a host. It’s just a genetic message inside a delivery mechanism: a protein coat shaped to bind to specific receptor sites on specific cell types. Once it gets inside, it doesn’t build its own machinery. It redirects existing host cellular infrastructure to reproduce itself.
That’s the entire strategy. Be minimal. Be targeted. Exploit what’s already there.
The minimal viable version matters. HIV has nine genes. The human genome has roughly twenty thousand. Everything the virus needs to survive, spread, and persist is encoded in those nine instructions, executed entirely by infrastructure it doesn’t own and didn’t build.
Latency here is the other key mechanism. Some viruses integrate their genetic material directly into the host cell’s genome and go quiet. Herpes simplex retreats to nerve ganglia and waits. HIV can remain latent in resting T-cells for years, invisible to the immune system and unaffected by antiretrovirals that only target actively replicating virus. The information persists inside the host, dormant, waiting for conditions to activate it.
Think like a biological virus: minimal, targeted, and entirely dependent on exploiting existing infrastructure rather than building your own. Identify the minimal viable version of the problem. So Strip it to its core replicating information and delivery mechanism. Find the receptor sites: the specific vulnerabilities or affinities in the target system that allow entry. Ask what existing machinery the problem is redirecting rather than replacing. Look for latency: where does the problem integrate quietly into the host system and wait for conditions to activate? Analyse the mutation rate and what selection pressure is accelerating it. Identify what would constitute an effective immune response, whether one already exists, and whether the problem has developed resistance to it. Then explain the relationship between the minimal information core and the vast machinery it’s recruiting — because the effort is being done by the host, not the pathogen.
This one is particularly productive for understanding how ideas, narratives, norms, and dysfunctions spread and persist inside said institutions. The virus doesn’t carry the reproduction machinery. It justs carries the instruction set..
The Bacterium
Bacteria adapt faster than they reproduce.
This mechanism is in the horizontal gene transfer. A bacterium that develops antibiotic resistance doesn’t have to wait for its resistant offspring to slowly outcompete the susceptible ones. It can share the resistance gene directly with a neighbouring bacterium through conjugation, a physical connection between cells that allows plasmid transfer. Other bacteria can pick up naked DNA from the environment after a cell dies (transformation), or receive it via a virus that accidentally packages bacterial DNA (transduction). Adaptive information moves laterally through the population, jumping individuals, skipping generations.
Quorum sensing is the collective coordination mechanism. Bacteria produce and release small signalling molecules continuously. When their concentration in the environment crosses a threshold — indicating that enough bacteria are present — specific gene expression changes collectively across the population. Virulence factors switch on. Biofilm formation begins. The signal isn’t a command from a leader. It’s a population density readout. We are enough. Act now.
Biofilm is the collective protection strategy. A bacterial population that forms a biofilm — a structured matrix of cells embedded in a self-produced polymer — becomes dramatically more resistant to antibiotics than the same bacteria in isolation. Some biofilms require antibiotic concentrations a thousand times higher to disrupt than planktonic cells. The collective structure is the defence.
Think like a bacterial colony: adaptive, collectively sensing, and able to move information laterally across the population faster than hierarchy permits. Analyse how adaptive capability — strategies, skills, workarounds, resistance , this is spreading through the system. Is it moving vertically through official channels, or jumping laterally and bi-laterally between individuals and units outside of formal structure? Find the quorum signals: what threshold does the population need to reach before it acts collectively, and what does collective action look like when that threshold tips? Identify where the system has formed a biofilm — a collective structure significantly more resistant to disruption than individual members would be alone. Look for horizontal gene transfer: who is sharing what with whom outside the sanctioned pathways, and how fast is that sharing moving compared to official rates of change? Explain what the system is preparing to do collectively once the quorum is reached, and what it would take to interrupt the signal before it gets there.
This is honestly the right model for resistance inside organisations , so showing the kind of informtion which doesn’t show up in surveys, doesn’t speak in meetings, and arrives fully formed when someone does that finality of trying to change something that everyone has been quietly protecting for years.
So across three articles, the full set covers nine distinct analytical modes.
The bat finds invisible architecture through indirect signal. The spider maps where pressure travels through relational structure. The mycelium reads metabolism beneath the visible symptom. The ant locates intelligence that lives in the environment rather than in any individual. The corvid tracks what the system is quietly preparing for. The octopus itidentifies what each part of a system knows locally that the centre doesn’t. The brain finds the gap between stated reasoning and actual mechanism. and the neural network attends to intermediate representations and adversarial vulnerability. While the virus traces how minimal information exploits existing infrastructure. And the bacterium follows adaptive capability moving laterally through a population.
However and this is important none of these is always the ‘right tool’. Some problems need one; some need several at once. The point was never to create a system. It was to notice that in different ways of framing how to look at a problem produces genuinely different outputs — and that evolutionary biology, having solved most of these problems in physical form, turns out to be a better source for those frames than most formal methodologies.
The session that started all this was, as noted in the first article, slightly drunk, but of abstract thinking. So this seems like the right way to end a taxonomy.
