The Bridge to Psychology
The gradient of Part I is a claim about the structure of mind, a mapping from computational requirements to psychological phenomena, and it can be checked against human experience and not only against simulations.
What Comes for Free
The lower rungs are the background of conscious experience, the stream of feeling that is always present, rarely attended to, and not intrinsically about anything. In psychological vocabulary:
- Mood: the tonic valence gradient, approach or withdrawal as a whole-body orientation. Any viable system has it, which fits the finding that mood is always present, precedes appraisal, and colors perception before cognition begins.
- Arousal: processing intensity as a dimension of state. Present in every substrate tested.
- Habituation and sensitization: compressed representations that have learned, across evolutionary or developmental time, what matters and what can be ignored.
- Animistic perception: attributing agency to non-agents, the cheapest compression available, reuse of the agent template already running on everything else. The developmental drift from childhood animism to adult mechanism is a fall in across most targets, and I think the fall has to be taught. That is the hypothesis of Part II, and it is still a hypothesis.
- Emotional coherence: that feelings make sense, that internal states map reliably onto behavioral tendencies. In the simulations this develops under selection rather than being present at the start, which suggests something testable: coherence is an achievement of developmental history, not a given of neural architecture.
- Temporal depth: carrying the past into the present. In the simulations memory was selectable; two of three lineages kept longer retention and one discarded memory entirely and did fine. Minimal temporal integration is a viable strategy, not a deficit.
None of these need awareness. None need a self. They are the geometry of being alive, and the claim that experience has geometric structure is grounded here rather than in anything reflective.
The biological instance turned up recently where nobody was looking. Xenobots, aggregates of cells taken from frog embryos and left free in a dish, self-organize into moving collectives that navigate, gather material, and show healing-like behavior absent from any individual cell's program. Anthrobots, built from human tracheal cells, do the same. No nervous systems, no neurons, and they are doing everything on the list above: approaching gradients, withdrawing from barriers, developing coherent behavioral modes, accumulating temporal depth through interaction with their local topology. Michael Levin, who developed them, describes the result in the terms this book uses, systems solving problems in a space defined by their biology, discovering in real time what it takes to persist somewhere new. The synthetic program is not exotic. It makes visible something biology has been doing at these rungs for hundreds of millions of years, too small and too quiet to notice.
Reactivity and Understanding
Higher up, the vocabulary changes kind. Below the agency threshold an organism is described by what it does: it approaches, withdraws, habituates, anticipates. Above it, by what it considers: it imagines, plans, regrets, evaluates. Call the two modes reactivity and understanding. Reactivity responds to what is happening now, each channel driven by present-state associations, decomposable in principle. Understanding compares what would happen across available options, and that comparison couples across any partition imposed on the system, because the comparison is the representation. This is the decomposability wall of Part I in psychological terms.
What separates them is architecture, not intelligence or evolutionary seniority. Lenia patterns have affect geometry, faint world models, memory, and biological-looking integration dynamics, and they never cross, because their "actions" bias a continuous field rather than changing an environment they then observe. Protocell agents with real action-observation loops () cross from initialization. The agency threshold is a substrate threshold.
Three capacities sit above the wall and inherit its requirements in order. Counterfactual reasoning needs a repertoire of possible actions, a model of how each changes the world, and a comparison between actual and counterfactual outcomes. Self-awareness needs a self-model more accurate than an outside observer could build from the same data, privileged self-knowledge in its minimal form. Moral reasoning needs both of those plus an asymmetry in the viability gradient between cooperative and exploitative action, which makes it the most demanding because it inherits everything below it. The ordering predicts development from requirements rather than from watching children: mood and arousal from birth, animistic perception as the childhood default, emotional coherence through experience, counterfactual reasoning around three to four, self-awareness from mirror recognition to autobiographical self, and moral reasoning last.
Depression as Lost Purpose
One formulation earns its own place because the geometry alone undersells it. Decoupled modes, flattened gradients, and collapsed rank are the measurable signature. The deeper description is that depression is what a cause-effect structure looks like when it has lost its model of its own purpose and is stuck with the inertia of its physical design. The system keeps asking what action increases future worth and the answer keeps coming back none. Motivation shuts down. Time gets heavy. The self-model loses agency. In this frame suicidality is not a wish to die. It is a control action proposed by a computation that is wrong: terminate the loop to end the mismatch between the running cost of persistence and a return that reads as zero.
The error is specific, and naming it is the therapeutic content. The system in depression cannot see the stock (Part II). It sees only the instantaneous flow, and the flow reads null. A system generating zero new bits still holds the accumulated structure of everything it has already integrated, and a null rate leaves that untouched. The qualifier is that the integral is not monotonic; the rate can run negative; so what this guarantees is that pausing does not erase, not that the stock is large. Treatment, in these terms, is anything that makes the stock visible again or restarts the rate. And the mortality interrupt above is the same correction arriving through the body rather than through argument.