Notes

The core scientific problem

Ethics needs a theory of consciousness, and so does any honest answer about whether the machines we are building have minds.

Why it is the core problem

Moral truths weren't handed down from a mountain-top. They can be described as a result of cultural evolution and group selection, and that makes many people cynical about morality. I disagree. I have experience, and it matters. I infer that other beings also have experience that matters, and I include their experience in my utility function.

My dog has experience that matters. Ants, I'm less sure. Rocks almost certainly not. Every one of those judgments is a claim about consciousness. If conscious experience is what grounds ethics, then doing ethics properly requires a theory of consciousness.

What I mean by consciousness

A first-person perspective: there is something it is like to be you (Nagel). I don't believe there is something it is like to be a rock. A dog? Maybe. You? Yes. Me? Without a doubt. No evidence or argument could convince me that there is not something it is like to be me.

But couldn't that be an illusion?

Take the dress. I see it as white and gold, and I know it is "really" black and blue. There are two true facts here. The fabric has a reflectance curve that most people see as black and blue. And about half of brains wrongly infer that the dress is lit by artificial light, so they see it as white and gold. The second fact is just as real as the first: look inside my brain and you would find white-and-gold color representations.

Claiming that the contents of consciousness can be illusory is different from claiming that the container is an illusion. There is something it is like to be perceiving the dress wrongly.

Two arguments against illusionism:

• Reports are caused by the thing. A simulated experiencer experiences, and that causes our verbal reports of consciousness. Suppose instead that whether the lights are on has no causal role. Then all our reports and deepest intuitions are independent of the truth, and whether anyone is conscious could never, even in principle, be determined. The whole field that searches for neural correlates of consciousness would have nothing to learn.
• Even zombies would need the variable. Grant that philosophical zombies are possible. They report negative states as unpleasant and act to avoid them. They also have prediction errors that they don't report and don't avoid. So even for zombies, some latent variable explains their behavior better than prediction error alone, and it causes their reports of consciousness. Pragmatically, that variable is causally real. I propose we call it consciousness.

If you accept either argument, you are a functionalist: mental states are constituted by their functional role, stimulus → pain → reaction and report.

Computational functionalism, and time

The stronger claim is that the functions of consciousness are relations within a computation, independent of substrate. A common objection: this makes consciousness independent of physical time, which seems absurd. Freeze a brain, thaw it, run one step, wait a year, run another step. Was it conscious for that year?

My answer is that consciousness exists in simulation time, not physical time. Our simulation is tuned to track reality, so the two usually line up, but they come apart: drugs, the slow motion of near-death experiences, flow states. In dreams they detach completely. A moment of experience binds information within a window, and a flashed image below the threshold of awareness leaves a gap we never notice. Continuity is something the simulation reports, not something physics guarantees.

The real crux

People rarely reject computational functionalism because they've found a flaw in the arguments. They reject its implication: the clankers may be conscious too. Denying that is the default that minimizes prediction error, and there is a heavy Chesterton's fence around our moral beliefs.

So what's the plan?

Solve the easy problems first. Pin down the functions of consciousness, find the neural correlates, and take on the meta-problem: why we report having experience at all. Then test the existing theories against those functions: global workspace, integrated information, higher-order and attention-schema theories, active inference.