Could a simulation be conscious?

Here is the uncomfortable thing about this question: there is no observation you could make, of this world or any other, that would answer it. Not a better microscope, not a bigger brain, not another nine months of running. That is not a limitation of the simulation. It is the shape of the problem, and it is worth understanding before deciding what you think.

Why no measurement decides it

You can measure what a system does. You can measure what it reports. You cannot measure whether there is something it is like to be it. This is the gap philosophers call the hard problem, and the clearest way to see that it is real is that you already accept it about everyone else: you infer that other people are conscious from their similarity to you, not from evidence that would work on something unfamiliar.

Simulated creatures break that inference, because the similarity runs out. An organism here has a recurrent neural network with 24 hidden units. It has no words, no face, and no capacity to tell you anything. Whatever your reason for believing your friends are conscious, it does not transfer.

What can actually be said about this world

Quite a lot, as it happens, and all of it is the kind of thing that normally gets hand-waved:

• There is a persistent internal state. Each organism's network carries hidden activity from one moment to the next, so what it does now depends on what happened to it before. That is memory in the mechanical sense.
• There is genuine perception: 53 channels of its own situation, including its own energy and the ground immediately ahead of it.
• There is no self-model, no representation of itself as a thing, and no channel through which it could report an inner state even if it had one.
• There is no reward signal of any kind. Nothing in this world feels good or bad to anything, because there is no quantity that plays the role good and bad play. That is unusual: most artificial creatures are built around exactly such a quantity.

The last point is the one that matters most and gets noticed least. Many arguments that an AI system might suffer run through its objective: it has something it is trying to maximise, failure is punished, so perhaps there is something it is like to fail. There is no objective here at all.

What the main theories would say

None of them is settled science, and they disagree with each other, which is itself informative:

• Functionalist views hold that consciousness is a matter of what a system does, not what it is made of. On these, a sufficiently rich simulation could in principle be conscious, and the question becomes whether 24 hidden units is anywhere near rich enough. Almost certainly not, on any version of the view.
• Biological views hold that consciousness depends on the specific physics of nervous systems. On these the answer is simply no, however the simulation behaves.
• Integrated information approaches try to make it quantitative, and would give this world a number. The number would be small. Whether the number means anything is exactly what is in dispute.

The position this project takes

None, deliberately, and the discipline is visible in the product rather than claimed in an about page. This world measures whether its organisms' signals carry information, continuously, and the answer is currently no - so the label beside the world reads signal, meaning not established, and it will keep reading that until the measurement changes. A project willing to say that about something as modest as signalling is not about to tell you its creatures have inner lives.

The useful version of this question is not "is it conscious". It is: what would have to be true, and how would we know? Watching something specific while you think about it beats arguing in the abstract, which is most of what this site is for.

Can a simulation suffer? · Is it alive? · Open an organism's brain