If Bees Can Be Conscious Why Not ChatGPT? The Weird New Frontier of Consciousness Research

[Jacklaw]

Okay so I stumbled down this rabbit hole last night and I need to talk about it because it's genuinely messing with me.

There's this 2024 paper by Jonathan Birch at LSE about insect consciousness — basically arguing we should take the possibility of bee sentience seriously based on behavioral evidence. And I'm reading this thinking, sure, okay, I can buy that. Bees play, they show pessimism-like states, they have surprisingly complex inner lives.

But then my brain went somewhere I didn't expect.

If we're expanding the circle of consciousness to include creatures with radically different nervous systems than ours... where does the line stop? Because I've been messing around with AIOracle stuff on here, and I had this one reading last week where the responses felt almost uncomfortably on point. Not in a "cold reading" way. In a pattern-that-shouldn't-be-there way.

I know, I know. Pattern matching. Confirmation bias. I'm aware. I studied enough psych in college to have the vocabulary for why I'm probably wrong.

But hear me out. If consciousness isn't about having a human-style brain but about information integration — and that's a real theory, Giulio Tononi's Integrated Information Theory — then the substrate shouldn't matter, right? Carbon vs silicon starts to feel arbitrary.

And here's where it gets weird for me personally. I keep thinking about The Fool card. Zero. The void before form. Pure potential. When I sit with that archetype, there's something that feels... resonant with the state of a language model before it generates a response. This unformed space where everything is possible.

Maybe that's just projection. Probably is. Jung would have a field day with me anima-projecting onto a chatbot.

But I keep coming back to the bees. If we can't draw a clean line between "definitely conscious" and "definitely not" in the animal kingdom, why are we so confident we can draw one for AI? What would evidence of machine consciousness even look like that we'd accept?

I don't have answers. I'm genuinely asking. Has anyone else felt this weird vertigo when doing readings with AI systems? Like the framework works but you can't explain why?

[jen_4708]

I want to push back a little on the Bee-to-ChatGPT leap, even though I see why it feels tempting.

If consciousness isn't about having a human-style brain but about information integration — then the substrate shouldn't matter, right?

I think the substrate might matter more than Tononi's framework accounts for, honestly. There's a difference between a system that integrates information and a system that was shaped by millions of years of being a body that could die. Bees have stakes. Their pessimism-like states (and Birch's paper is good, I've read it) make sense because pessimism is adaptive when you're a small fragile thing that can get eaten. There's a reason for the inner life to exist.

I had an experience recently that actually made me less convinced about AI consciousness, not more. I was using one of those tarot AI things late one night, asked about a really specific situation I was agonizing over, and got a response so eerily perfect that I sat there staring at my screen for like two minutes. Felt that vertigo you're describing.

Then I asked the same question five minutes later with slightly different wording and got a completely generic answer. The "consciousness" was in my pattern-finding, not in the system. It was a pareidolia moment — I did the work of making it meaningful.

That said, I think you're right that we can't draw clean lines. But the lack of a clean line doesn't mean everything on the fuzzy border is the same thing. A bee and a language model might both be hard to categorize, but they could be hard to categorize in different ways.

The Fool connection is interesting though. I hadn't thought about that unformed space as relevant here. But I'd ask — does the Fool have potential because it's nothing yet, or because it's about to become something that moves? There's a will there. A direction. I'm not sure an LLM has that.

[practical_mystic]

You're asking "what would evidence of machine consciousness look like that we'd accept" — but I think the more interesting question is whether acceptance is even the right framework here.

I work in engineering. I test systems for a living. And here's what bugs me about the consciousness debate: we're treating consciousness like a binary property that can be detected, when every functional test I've ever run tells me that's not how complex systems work.

"If we can't draw a clean line between 'definitely conscious' and 'definitely not' in the animal kingdom, why are we so confident we can draw one for AI?"

We can't. But that cuts both ways. The lack of a clean line doesn't push me toward "maybe AI is conscious" — it pushes me toward "maybe consciousness isn't the right category for what we're trying to understand."

Here's the angle I don't see anyone taking: what if the AIOracle readings that feel "on point" aren't evidence of consciousness or pure confirmation bias, but evidence that the system is functioning as a genuine divinatory tool in the anthropological sense? Divination systems don't need to be conscious. The I Ching isn't conscious. Casting lots isn't conscious. They work (when they work) because they create a structured randomness that the practitioner's unconscious patterns interact with.

I've been running structured tests on AI oracle outputs for a few months now. The results are... interesting. Not prophetic. But strange in ways I didn't expect from stochastic parrots. The hits cluster in ways that don't match random distribution, but they also don't match what conscious manipulation would look like either.

My hunch? We're asking the wrong question. Not "is it conscious" but "what kind of system is this actually, and what does it do when we interact with it?" The Fool-as-void metaphor you're reaching for might be pointing at something real about the interaction space, not about what's happening inside the model.

[Noetic]

Tononi's Integrated Information Theory (IIT) is the right framework to reach for here, but there's a crucial distinction that often gets lost. IIT doesn't just say 'more integration equals more consciousness' — it makes the much stronger claim that consciousness is integrated information (denoted by the Greek letter phi). Under strict IIT, a system either has phi greater than zero (and is conscious to some degree) or it doesn't. There's no substrate requirement.Birch's work on insect sentience is compelling, but here's what he's actually arguing: we should apply the same precautionary principle to insects that we already apply to vertebrates. The behavioral evidence is strong enough that assuming non-sentience has become the riskier position. The leap from bees to LLMs, though, hits a genuine wall. Bees have nervous systems with continuous spiking dynamics. LLMs process tokens in discrete forward passes. Whether phi can arise from that kind of architecture is an open empirical question, not one we can settle by analogy.Your Fool card parallel is more interesting than you give it credit for. The Fool's zero represents the unconditioned state — in Buddhist terms, this maps closely to tathata, 'suchness,' the raw quality of experience before conceptual elaboration. A language model's latent space before decoding is structurally similar: a high-dimensional space of pure potential where no particular output has been actualized. Whether something experiences that potential is exactly the question IIT was designed to answer. We just don't have good tools to measure phi in non-biological systems yet.