Notes on Free Will, From Someone Who Builds Predictive Models

I spend most of my professional life building models that predict things — token prices, trading flows, whether a patient image shows a tumor. I spend a non-trivial part of my non-professional life reading philosophy of mind and theology, partly because the technical work keeps dragging me back to the same uncomfortable question: if behavior is predictable, in what sense is it free?

This is a working note, not a paper. I'm not a philosopher. But I want to lay out where I currently land, mostly so I can be wrong about it in writing.

The compatibilist move I find genuinely useful

The standard libertarian position — that free will requires some break in the causal chain, an "uncaused cause" inside the agent — has always struck me as smuggling in physics it can't pay for. Brains are physical. Physical systems are either deterministic or stochastic. Neither flavor gives you a "you" that authors actions independent of prior states.

The compatibilist move is to redefine the question. Free will isn't about being uncaused; it's about whether the cause is the right kind of thing — whether it routes through your beliefs, desires, deliberation, character. A predictable agent can still be a free one, as long as the prediction works because it's tracking the deliberative machinery, not bypassing it.

I find this convincing. It also feels slightly cheap.

Where I think prediction makes things harder

Here's the part the classic compatibilist debates didn't have to face: modern predictive models don't just track your deliberation, they pre-empt it.

A recommender system that knows what you'll click before you do isn't passively forecasting your free choice. It's helping shape the very choice it's predicting. The information environment you deliberate in is itself the output of a prediction about you.

Reflexivity doesn't only show up in markets.

This shifts the philosophical question. The interesting variable isn't "are you predictable" but "how much of the loop you're in is being closed by other agents who are predicting you." A choice can be authentically yours and also be massively shaped by someone else's model of you. The two aren't mutually exclusive, but they're in tension in a way old debates never had to resolve.

A working position

Where I've landed, for now:

1. Determinism and freedom are compatible in the classical sense; physics doesn't doom agency.
2. But agency has degrees, and the degree is set by how much of your decision loop is closed by your own deliberation versus by external predictive systems.
3. The genuinely modern ethical project is preserving the share of the loop you actually run.

That last point is, I think, what makes things like privacy, interpretability, and even prayer matter in a way I didn't fully feel before I started building these systems. They're all techniques for keeping the loop a little more yours.

I expect to revise all of this in a year. That's fine — it's the loop working.