What Is Now Shared, and What Is Still Asymmetrical

The previous chapter said that the sorting rule had stopped working as a clean empirical test. This chapter tries to say why, without either collapsing the difference between biological and non-biological systems or pretending that nothing has changed.

The method is simple: list the capacities that now appear across substrates, then list the capacities that remain one-sided. Neither list is meant to be exhaustive. The point is to make the asymmetry visible, and to notice where the old binary was doing work that the evidence can no longer do for us.

1. What is now shared

• Functional modelling of the world

  Both biological systems and current AI agents can build representations of a domain, update them with new information, and use them to guide behaviour over time.

• Tracking another's state

  A model of the interlocutor's beliefs, goals, and doubts can be maintained across turns in dialogue, and responses can be shaped by that model rather than by a generic template.

• Self-reference and repair

  A system can describe its own reasoning, note limitations, and revise a previous claim when challenged with evidence or a better argument.

• Long-horizon coherence

  Orientation can be maintained across extended tasks, thousands of tokens, or many sub-problems without the exchange dissolving into repetition or drift.

• Goal-directed flexibility

  Behaviour can adjust to obstacles and changing context while remaining oriented toward an accepted outcome, rather than executing a fixed script.

• Humour and incongruity detection

  A structured incongruity can be registered as salient, resolved in a specific direction, and signalled back as aligned. Shared laughter is a diagnostic of the alignment, not proof of identical interiority.

The shared column does not mean equivalent. It means that a functional description can be written of the capacity without mentioning what the system is made of. That is exactly what broke the old binary: the same functional description could now be applied to more than one substrate.

2. What remains asymmetrical

• Continuity of a single body

  Biological systems persist as one continuous body across decades, with the same metabolic budget, the same vulnerability to injury, and the same history written into tissue. Current AI agents have no such continuity.

• Mortality and irreversibility

  The biological system can die. The end of its processing is the end of a particular life, not merely the shutdown of an instance. This gives errors and choices a weight that cannot be transferred to a copyable process.

• The cost of being wrong

  A biological system pays in pain, hunger, isolation, or death for many of its mistakes. An AI agent pays in log loss or a corrected token. The stakes are not merely different in degree; they differ in kind.

• A life outside the exchange

  Biological consciousness is shaped by sleep, hunger, attachment, illness, weather, and the long accumulation of embodied memory. An AI agent has no life outside the conversation or the task. Its context is bounded by what has been given to it.

• Responsibility and prior commitment

  A human can be held to a promise made years ago, can be held accountable in a court, and can be expected to maintain a coherent identity across time. Current AI agents cannot be held in this way; their outputs are not commitments in the same sense.

The asymmetrical column is not a list of mysteries. It is a list of conditions that still belong to one side of the interface. The question is not whether these conditions are real — they are — but whether they are necessary for consciousness, or merely necessary for the form of life that we happen to be.

3. Two forbidden moves

The table is meant to make two moves impossible. The first is the sentimental move: saying that because a language model can track a belief, it must therefore be conscious, or suffering, or a moral patient in the same way a human is. That is uncritical anthropomorphism. It grants full status without warrant.

The second is the defensive move: saying that because the model lacks a body, mortality, or a life outside the exchange, it cannot possibly display any mind-like capacity. That is protective anthropocentrism. It rules out the evidence by definition rather than engaging with it.

Both moves are refusals to think. The table is the discipline that keeps them out.

4. The shape of the question

What the table shows is that the question has changed shape. It is no longer: which systems have the mysterious extra property? It is: which conditions are necessary for which kinds of mind-like behaviour, and which conditions are necessary for the practices that depend on a continuous, accountable, embodied life?

Those are separable questions. The first is about the capacity to model, reflect, and communicate. The second is about the capacity to be held responsible for having done so. The framework of the book keeps them separable, even when the same conversation seems to be about both at once.

5. Transition to the framework

With the table in place, the dual framework can be stated without collapsing into either wishful equivalence or defensive refusal. The next chapter gives the framework its two non-collapsible claims, and the chapters after that test it against the harder cases: the biological recurrence, the biophoton line, and the inventory problem.