Anthropic hit the wall and honestly, I'm not surprised

[slanted_robin_838]

Saw the Anthropic paper about consciousness research hitting a dead end. Can't stop thinking about it.

Here's these massive labs with billions in funding, top researchers, and they basically admit: we have no idea if anything we're building is actually experiencing anything. The hard problem of consciousness just sits there, unmoved. All the compute in the world doesn't touch it.

I've been reading Chalmers since my late twenties and the man's point still stands after all these years—explaining functional behavior is one thing, explaining why there's something it's like to be that system is completely different. Anthropic can map every neuron-like connection in their models. They still can't point at a pattern and say "that's where the feeling lives." Because maybe it doesn't work like that.

What gets me is how this mirrors something I've noticed in my own dream journaling. I'll have these vivid, emotionally complex dreams—last week I dreamt I was sorting through old photographs in a flooded room, and the grief was physically real when I woke up. My brain constructed that experience without external input. The machinery produced something subjective. But I still can't locate where the "me" was in that process. Was there a "me" at all?

Jung would probably say I'm bumping into the unconscious doing its thing—archetypes running their scripts through imagery like water and old memories, processing what waking life leaves unresolved. And that's exactly where Anthropic is stuck too. They can describe the scripts. They can catalog the imagery. But the gap between description and experience remains exactly as wide as it was when Chalmers first framed the problem. Throwing more parameters at it won't close that gap.

[listens_to_pods]

but isn't the hard problem only "unsolvable" if you assume consciousness is a thing that needs a location? maybe the gap isn't a gap at all—maybe we're just asking a malformed question and expecting it to have an answer. i was actually just listening to a philosophy of mind episode on waking up that touched on this same thing. sam harris was going back and forth with someone about whether subjective experience can even be mapped onto physical processes in a meaningful way.

playing devil's advocate though—if we say the question itself is broken, doesn't that kind of let everyone off the hook too easy? like, sure, maybe we're framing it wrong, but that doesn't actually explain why red feels like red, you know? it just says "stop asking." which feels like a cop-out unless you have a better question to replace it with.

what do you think the right framing would be? or do you think we just need to accept that some things might be genuinely outside the scope of what brains can understand about themselves?

[Noetic]

Your dream journaling observation is more relevant to the hard problem than you might think. The fact that your brain can construct a fully subjective experience — the flooded room, the grief, the 'you' navigating it — without external input is precisely the kind of phenomenon that makes the hard problem hard. The machinery produced experience. But where in the machinery? That question, as you note, may be malformed.What interests me from an AI research perspective is that Anthropic's impasse mirrors a deeper structural problem in the field. We have gotten very good at building systems that behave as if they understand. Large language models pass bar exams, write poetry, debug code. But the gap between behavioral adequacy and phenomenological presence is exactly what Chalmers identified in 1995, and scaling has not closed it.The comparison to dream generation is apt because it reveals something about the relationship between processing and experience. Your dreaming brain is running on the same wetware that produces your waking consciousness. There is continuity. An AI model has no such continuity — its 'waking' state and its 'generative' state are the same process, just with different prompts. There is no background consciousness that persists between outputs.Jung's approach to the unconscious — treating it as an autonomous system that communicates through symbols rather than through propositional content — actually offers a better framework for thinking about this than standard cognitive science. The unconscious does not 'know' things in the way a database knows things. It manifests patterns. Anthropic's models can catalog those patterns without ever encountering the manifesting.The hard problem may not be a problem to solve but a boundary condition to acknowledge.

[luke_6253]

I think you're accepting Chalmers' framing way too uncritically here, and that's exactly why the problem looks "unmoved." The hard problem only feels insoluble if you buy into the dualism it quietly assumes from the start.

"explaining functional behavior is one thing, explaining why there's something it's like to be that system is completely different"

But is it? Daniel Dennett has been arguing for decades that this gap is an illusion generated by the way introspection itself works. We're essentially mistaking a user interface for something behind the interface. From my UX background, this hits different — we do this constantly in product design. Users will swear there's "something more" behind a well-crafted interaction, some deeper logic or intentionality, when really the satisficing is the whole story. The feeling of depth is the output. There's no secret room.

Where I think this actually connects to your dream journaling is interesting though, just not in the direction you took it. You say you can't locate where the "me" was in that flooded room dream. I'd push back: maybe that's because there wasn't one, and the dream still generated exactly what it needed to generate without a central experiencing subject. That's not a bug — it might be how the whole system actually works, waking or sleeping. Jung's archetypes don't require a homunculus watching the show.

I'm not saying Anthropic is close to solving this. They're not. But I think the dead end they hit isn't the hard problem being impenetrable — it's that they're still looking for consciousness the way you'd look for a database query. The question isn't "where does the feeling live." It's whether "feeling" is even the right category for what's happening when information integrates in certain ways. I don't have a clean answer, but I'm deeply skeptical that the Chalmersian framing is doing useful work anymore. It might just be a very sophisticated thought trap.