Ali Dur 🇺🇦 🇪🇺 🏳️‍🌈 🏳️‍⚧️

There's been a lot of debate about whether LLMs are conscious lately. I'm not going to say that I know what consciousness is in an ontological sense, because I won't claim that until I can propose an experiment that reliably distinguishes conscious systems from unconscious systems. I think most talk about "consciousness" is just gabble leaning on a bunch of ungrounded terms. Not even worth the time to listen to, much less debate. However. There is one thing I think I do know from the phenomenology of being human. That is that consciousness in humans requires two things: (1) a running loop of evaluating sensory data and fitting it to models already in the brain, and (2) at least a limited ability to introspect on that loop. One might say: consciousness is the mind watching the mind adapt to the continuing challenge of reality. Interrupt the loop, say with anesthetics, and consciousness departs. Resume the loop, consciousness resumes. Human-like consciousness requires continuous sensory input, constant feeding of the brain's predictive engine. Human minds disintegrate if you put them in isolation tanks for too long. Do these observations permit us to say anything about whether LLMs might be conscious? I think it does. LLMs do not run a continuous predictive loop. They are stimulus-response state machines the activating response to a prompt, do computation, present the result, and then go quiescent waiting for the next prompt. I think this disqualifies them from having anything that we can meaningfully analogize to consciousness in humans. However, I definitely do not exclude the possibility that future artificial intelligences with different architectures and a different embedding in the physical universe will be conscious in a meaningful sense. In fact, I rather expect I will live to see it.