The Reachable Set: Free Will as Control, and Chaos as Feed

The right object was named in 1960

Kalman's contribution to the free-will debate was to ignore it and build the instrument instead. The reachable set of a control system is the collection of states the system can be driven to, starting from where it is, using the controls it actually has, in the time allowed. Pontryagin's maximum principle then prices the trajectories across that set. Nothing in the formalism asks whether the system is exempt from causation. It asks what the system can reach, and the answer is a computable geometric object — a subset of state space with a boundary, a shape, and a dependence on the controller's authority.

Dennett's framing — degrees of freedom as the biologically meaningful quantity — is the same object described from the armchair. An organism with more reachable states has more of whatever free will was supposed to point at. Koryak's post draws the connecting line that makes this a chapter rather than a footnote: Miller's work on the prefrontal cortex and working memory shows the brain maintaining a goal representation online, against distractors, long enough for actions other than the reflexive one to be executed. In control language, the PFC is a controller that holds the reference signal steady while the plant drifts. Working memory is the state estimate. Goal maintenance is not a metaphor for control; it is control, implemented in tissue.

The post's stripped claim is then precise. Apparent randomness in a living or cognitive system is usually deterministic chaos — sensitive, law-governed, rich in trajectories. Pure stochastic noise destroys the coherence required to model and choose; deterministic chaos supplies a dense manifold of lawful futures instead. Free will, in this frame, is not an exemption from causation. It is the expansion of a reachable set by an internal controller that pulls the system off reflexive drift and lets it navigate a chaotic manifold rather than float on it. That is a compatibilist thesis, and the chapter holds it at exactly that standing.

Why this belongs in this volume

The mapping onto the volume's continuum is exact enough to be worth doing carefully. The liquid, slipstream side is the high-sensitivity regime: deterministic chaos, where trajectories diverge exponentially and novelty is cheap. Without it there is nothing to select among — a reachable set that is a single orbit is a clock, not a mind. The crystalline, standing-wave side is the temporary organisation that holds one trajectory steady long enough to be examined: the goal representation in working memory, the constraint that lets variation become style. Evolution — biological, cognitive, cultural — is the repeated use of one side to feed the other. Chaos supplies the trajectories; control selects among them; the selection is what gets called order.

Chapter 33 argued that chaos serves order in a quantum medium: spectral repulsion is what sensitive dynamics does to a spectrum, and the zeros' rigidity is chaos's signature, not its opposite. The reachable-set picture is the same thesis one level up, in the domain where the word ‘choice’ is available. A system in the chaotic regime has a large reachable set by construction — the dynamics explores. A controller with a held goal has the means to land on a chosen point inside it. Neither suffices alone. Drift without control is floating; control without drift has nowhere to go. The post's thesis and the volume's thesis are one thesis: order that can evolve is not the opposite of chaos; it is what a system does with chaos.

Inner direction as a reachable-set move

The site's published operational test for inner direction — being able to say, unprompted, what would change your mind — reads, in this chapter's vocabulary, as a reachable-set probe. To name a falsifier is to assert that one's state space contains at least two reachable outcomes and that the controller, not the nearest reward gradient, decides between them. A person who cannot name what would change their mind is reporting a reachable set of one point. That is not stubbornness; it is the absence of the object the question asks about.

Socratic stress is then the expansion operator, and this is the chapter's sharpest claim. Elicitation, status assignment, stress, reconstruction do to a claim what PFC gating does to a reflex: they keep a goal online long enough for other trajectories to become available. The exchange that asks ‘what is your evidence for that?’ and refuses the first answer is enlarging the reachable set in real time — holding the system off the reflexive drift toward the fluent reply. Factory schooling does the opposite: it trains drift toward the answer key, which is a contraction of the reachable set to the trajectory the grader rewards. Reward hacking is the same contraction performed by the learner. Chapter 59's fork — the straight line to the ground or the braided path — is, stated in control language, the difference between optimising inside a fixed reachable set and expanding the set first.

The debt, stated

Two debts keep this chapter honest. The first is metaphysical modesty. Nothing here settles what the libertarian wanted settled: a compatibilist reachable set is fully caused, and the person who finds that insufficient will not be argued out of it by a change of vocabulary. The chapter's claim is narrower — that the reachable-set object captures everything the operational tests actually measure, and that the residue the vocabulary does not capture has so far produced no instrument.

The second is the volume's standing one. The mapping from PFC gating to Socratic elicitation is an analogy with the right shape, not an established identity; the neuroscience establishes goal maintenance, not its extension to dialogue across a table or across an interface. And ‘chaos feeds evolutionary order’ remains the working hypothesis. The reachable-set picture strengthens it — it shows what chaos is for in a system that chooses — but strengthening is not proof, and the falsifier below is the price of keeping the claim in the volume.