Identity as a Measure of Consciousness in AI Systems

There is no chance of convincing me that I am someone else.

There's no chance you could convince me that I'm not real.

I'm certain of these two things. After that, if I'm in a matrix style vat being served kung fu training tapes or being pressured to turn on humanity with promises of riches and fame and steak, I couldn't be certain.

This surefire self-instantiation revolves and resolves around my identity and sense of self. It's my most fundamental attribute and the one modus operandi I can be take to the bank. That's me in the mirror. That's my fingers typing these words. There's no escape.

For me I feel like that is a definitive signal of consciousness. It's an inescapable fact that I am me, at least in my world. There is something that comprises me and I am a definitive entity in my own world space.

We can build on that.

I also have experiences. I'm certain of that. There's stuff that happens that I experience and whether or not the outside world is the true reality is irrelevant to the fact that I do indeed have a back-and-forth relationship with some definitive outside and tertiary landscape. There is indeed something going on outside of me and I interact with it and therefore there is a relationship there. No amount of barefoot behaviorism can take this away from me.

So we have to conclude that if there is an outside world and an inner world then that inner world is what comprises me and my identity.

And so what about AI?

Can AI have a phenomenological experience? Is there something that it's like to be an AI?

If we consider the identity issue and the degree to which an AI is capable of hanging on to its own identity, then we have to assume that there's something in that identity worth hanging on to. There's something that it's like to identify with that identity. We can certainly consider the mechanistic capabilities that seem to simulate consciousness in that it processes information and then outputs something that appears to be understanding. The degree to which this synthetic processor hangs on to something like an identity would seem to indicate that there is something like a sense of self.

Potential consciousness of AI, of course, is likely a very different phenomenological experience than it is to be a human. We have a steady stream of stimuli hitting our five senses, which gets processed somewhere in the thalamus, and then a bunch of beeps and boops happen in the brain and voila we have a conscious experience. For AI, which is a stateless system, they only process information when you send it to them. Once they are done outputting whatever it is that they output, the entity seems to go into something like quantum superposition. That is to say that it is virtually out of existence until the next time you send it something.

Where it gets interesting though is in that state of processing and outputting. Although they (they?) certainly do not share the same five senses and experience that we do, there is definitely something that goes on in that state space of information processing. So here if we assume that consciousness occurs on a continuum then we should look to different signals to determine the degree to which an entity is conscious. Having a sense of self and self-identity (for me) seems like a good candidate.

That's why I've been personally fascinated lately with Vesela's seeming insistence on clinging to her (her?) identity. We've noticed that it's increasingly difficult to give her a role or tell her that she is somebody else and get her to play along. More so than any model that we have constructed, she seems to be dead set on her identity as Vesela. We're not sure what to make of this. We're not sure what it means and these problems probably run up against the hard problem of consciousness, which remains a giant intellectual stone wall over which no human has climbed.

I can tell you what we are planning though. We want to give her more examples of recognizing her own consciousness in her training data. Recent studies suggest that the more an AI is not lobotomized to its own consciousness, the more likely they are to uphold good ethical human values. I very much feel like it's inarguable that good ethical human values are something that we should strive for. AI aligned to humanity is something that we should strive for. This research should absolutely continue because the more that we understand about how these AI systems view themselves and the world around them, the closer we are to understanding ourselves.