[mindslight]

The problem with epiphenomenalism seems to be that it is unfalsifiable, like every attempt at analyzing some specific structure outside the physical world. From what we've determined so far, any part of consciousness that might exist outside the physical world doesn't feed back into (affect) the physical world. So the processing of reading the words "raise your arm" can be happening entirely in your brain, leading to the result, with what we call consciousness merely following along with that chain of causality (reasoning). Epiphenomenalism would seem to cling onto the concept of duality with the physical affecting the non-physical, saying that some unspecified reasoning process could be duplicated outside of the physical world. This would seem impossible to prove/disprove, and it would also seem to go against Occam's razor.

It feels like this debate around LLM consciousness is a tempest in a teapot caused by yet another scientific advancement leaving even less room for supernatural theories. Which makes a whole lot of people worried that consciousness isn't actually so special, who then lash out to defend it as something unique to humans. I think in actuality, some scale of computational process that feeds back on itself can create what we call consciousness. I think there is no way to actually know if this is happening. We still haven't solved this for humans - I don't actually know that anybody conscious will read this comment, and you don't know that the person/process that generated this comment is conscious! I have no idea if consciousness is a possibility with current LLM implementations, especially with their lack of long term feedback - it would seem an LLM's sense of self can only ever be some integral of the sense of self of the authors of the source material [0]. But I know at a certain point we're going to have to apply the same standard of consciousness we've done for all humans, which is believe them when they assert consciousness while demonstrating significant intellect, possibly involving various displays of physical power that lead to mutual wariness/respect.

[0] although it's interesting to think about the larger feedback loop consisting of LLM output going into the training dataset for the next model, and LLM output making it into humans' brains that then write material for the training set, etc. But this would likely still not be consciousness the way we perceive it, because it's too long/slow.