What is free will?

Free will is the everyday sense that you can steer: you can notice an impulse, choose against it, and act for a reason. The opposite experience is familiar too: “I couldn’t help it.” When those experiences are treated as metaphysical magic (uncaused choice, outside causality), the debate gets stuck.

This post uses a deliberately deflationary framing: free will is a control capacity.

A working definition

We will use free will to mean: the agent’s capacity to act according to what it endorses as right (its higher-order model, values, and commitments) rather than being dragged by a local loop it disavows. talk: The Ghost in the Machine

In this sense, free will is not opposed to determinism. It is opposed to compulsion: acting despite knowing better. talk: The Ghost in the Machine

What this dissolves (and what it doesn’t)

Determinism vs randomness is a substrate-level argument. Control is an architectural property: a causal system can implement steering, deliberation, and self-correction precisely because it is causal.

So the interesting question is not “is the universe deterministic?”, but: what kind of controller is this mind, and how reliably can it keep the policies it endorses in charge?

This also means “free will” comes in degrees. Some people have more of it (more stable governance across time scales); some people have less of it (more frequent capture by short-horizon loops).

This framing does not, by itself, settle ethics. It changes the question from metaphysics to design: what control capacities exist, what commitments can be maintained, what training signals shape behavior, and what governance loops constrain action.

Free will as governance (not willpower)

A mind is not a single policy. It is a collection of policies operating at different time scales with different training histories. Self-control is the internal governance problem of allocating authority across them. talk: Self Models of Loving Grace

Phenomenologically, conflict feels like divided will because multiple policies become simultaneously active. When governance succeeds, action feels coherent (“I meant to do this, and I did it”). When governance fails, action feels alien (“I watched myself do it again”). talk: Self Models of Loving Grace

On this view, compulsion is not moral weakness or lack of information. It is a governance failure: the endorsed policy did not control the actuators.

Practical implications (if it’s a capacity)

If free will is a control capacity, you strengthen it the way you strengthen other capacities: by changing the conditions under which internal policies compete.

Three levers show up repeatedly:

• Environment: change the input distribution (remove triggers, add friction, add scaffolding).
• Attention: notice the trigger early enough to reallocate processing resources.
• Commitment/identity: install constraints that persist across contexts (rules, promises, self-concept).

This is not a self-help slogan; it is a statement about mechanisms. A mind does not choose actions from nowhere. It selects actions through competing policies activated by cues and modulated by value. Change the cues, and the competition changes. Change attention, and arbitration changes. Change commitments, and the admissible policy set changes.

For the longer treatment (habits, commitments, and failure modes), see Reader: “Free will as control capacity”.

References

• talk: The Ghost in the Machine @ 00:36:40
• talk: Self Models of Loving Grace @ 00:32:16