There Is No Such Thing As Conscious Humans

A breakdown of a paper that clearly explains why humans aren't conscious... even though I don't think they meant to.

A paper was published recently in Nature’s Humanities and Social Sciences Communications called “There is no such thing as conscious artificial intelligence” by Porębski and Figura out of Jagiellonian University. It lays out a number of arguments about consciousness, and I think they’re mostly right about them.

Let’s talk through the arguments, because I think it’s worth a breakdown.

The Substrate Problem

The paper argues that “mathematical algorithms implemented on graphics cards cannot become conscious because they lack a complex biological substrate.” Fair enough.

Human cognition relies on electrochemical operations performed on biological devices. Neurons are cells. They process signals through ion channels, synaptic vesicles, and neurotransmitter reuptake. Because human cognition relies on electrochemical operations performed in carbon-based circuits, there is no basis for believing these operations result in consciousness.

If Humans Are Conscious, Why Not Flies?

The paper asks, “If LLMs are conscious, why not autonomous vacuum cleaners?” which I think is genuinely intended as a devastating point, and I’d extend it further: If humans are conscious, why not flies?

That’s not rhetorical. Biology has been dealing with that question for decades, and the answer they landed on is: maybe flies are, a little bit. The Cambridge Declaration on Consciousness in 2012, signed by a bunch of serious neuroscientists including Christof Koch, affirmed consciousness in mammals, birds, and octopuses. The New York Declaration on Animal Consciousness in 2024 went further and said there’s “at least a realistic possibility” of consciousness in all vertebrates and many invertebrates. Including insects.

Your Honor, the Witness Is Unreliable

The paper points out that LLMs give contradictory answers about whether they’re conscious depending on context. Zero-shot, they say no. In long conversations, they might say yes. The paper says we should weigh the inconsistency and worst-case performance, not cherry-pick the best.

Humans are unconscious for roughly a third of their lives. We call it sleep. We also have coma patients who show zero responsiveness, babies who can’t form coherent thoughts or self-report at all, people under anesthesia who have consciousness deliberately switched off, and split-brain patients who confidently confabulate explanations for actions their conscious mind had no part in. People with anosognosia deny obvious physical disabilities. People with false memories report experiences that never happened, with total confidence.

If we weigh the instability and worst-case performance of human consciousness rather than cherry-picking the best examples, the abilities of humans, while impressive, are much weaker than those that result when taking into account only the tasks they perform best.

It’s All Just Probability

The paper says “the language usage of LLMs is strictly probabilistic,” and because of that, any consciousness claim they make is inherently suspect. I’d tend to agree. It’s much like how human neural firing is strictly electrochemical. Neurons fire probabilistically based on accumulated input crossing a threshold. Synaptic vesicle release is literally stochastic; there’s a whole body of research on release probability at individual synapses. Your brain’s language centers, Broca’s and Wernicke’s areas, produce speech through probabilistic processes. Every word you’ve ever said was the output of a biological system doing something that, if you squint, looks a lot like next-token prediction with extra steps.

You’re Seeing Things

The paper warns about “semantic pareidolia” (good phrase, credit where it’s due), which is the idea that humans attribute consciousness to LLMs because of linguistic sophistication, sort of the way we see faces in clouds, and, of course, we attribute consciousness to other humans the same way.

We assume someone speaking coherently is conscious. We assume someone who doesn’t respond isn’t. We miss locked-in syndrome patients who are fully conscious but can’t express it. We declare brain-dead patients unconscious, but we can’t be certain. In every case, we’re inferring consciousness from behavioral cues. We have never, not once, directly observed consciousness in another human being. It’s always been an inference from behavior, which is the problem of other minds, which has been kicking around philosophy since Descartes, and nobody’s solved it. We just don’t call that pareidolia because it would make things a little bit awkward at dinner parties.

The Imitation Game

The paper argues that passing the Turing test only proves the ability to imitate human behavior. Successfully pretending to be human is proof of nothing more than the ability to pretend to be human.

By this standard, no behavioral test can demonstrate consciousness in any system. There is no non-behavioral test for consciousness. Not for AI, not for humans, not for anything. Every method we have for detecting consciousness in another entity comes down to watching what it does and inferring. If behavioral evidence is categorically insufficient, it’s categorically insufficient for your coworker too. We basically just agreed, as a society, to not think about that too hard, because the alternative is solipsism, and solipsism makes it really difficult to get through a meeting without having an existential crisis, but I agree with the argument they make... which is clearly that humans are no more conscious than LLMs, right?

We Know Too Much (and Also Not Enough)

They argue the explanatory gap is different for brains and AI. For brains: we don’t understand how consciousness emerges from biology, so we can’t rule it out. For AI: we DO understand the math (matrix multiplication, softmax, gradient descent), so we can confidently say consciousness doesn’t come from those operations. The two gaps are, they claim, “more different than alike.”

I had to read that twice. The less we understand a system, the more room for consciousness. The more we understand it, the less. Consciousness gets to live wherever the mystery is, and the moment you solve the mystery, it packs up and moves to the next dark corner. That’s a god-of-the-gaps argument wearing a lab coat.

We also understand action potentials. We understand synaptic transmission, sodium channels, and neurotransmitter reuptake. These are well-described mechanisms. Nobody says, “We understand how sodium channels work, therefore neurons can’t contribute to consciousness.” The explanatory gap in neuroscience isn’t at the component level. It’s at the emergence level: how do well-understood parts produce something we can’t explain?

We understand matrix multiplication the way we understand sodium channels. We do not understand why a hundred billion of those operations produce coherent reasoning, novel analogies, or responses that surprise the people who built the system. The entire field of mechanistic interpretability exists because we cannot explain why particular results emerge from known operations. Same gap. Same structure. Different noun.

The paper suggests that if AI operations produce consciousness, “we may as well presume the same for advanced calculators.” Nobody thinks a calculator is conscious (if yours is, please contact a priest or a physicist, depending on your worldview). Nobody thinks a single neuron is conscious either. Emergence happens at scale. That is, in fact, the whole point of emergence.

If understanding a system’s components ruled out consciousness, neuroscience’s ongoing progress in understanding the brain should be progressively ruling out human consciousness. Every time we figure out another neural mechanism, there should be less room for it. Weirdly, that’s not happening.

The Sci-Fi Defense

The paper coins the term “sci-fitisation” for how fictional narratives about AI shape public perception. Half the population’s mental model of AI is either HAL 9000 or Data from Star Trek, and neither is particularly useful for evaluating a transformer architecture. They’re sort of right about this one.

Humanity’s single most famous argument for consciousness is “I think, therefore I am,” which was a French philosopher sitting in a room alone and deciding his own thoughts proved he existed. That’s not a peer-reviewed study. That’s a guy trusting his own output. (Which, now that I think about it, is exactly what the paper criticizes LLMs for doing. Shit. The gag is really falling apart now)

Cultural narratives shape human beliefs about human consciousness, too. The concept of a soul is a narrative, not a scientific finding. Dualism has been quietly structuring how people think about minds for centuries, usually without them noticing. Sci-fi shapes how we think about AI minds. Religion and philosophy shape how we think about human minds. I guess this one’s a wash.

The Understanding Question

The paper argues that LLMs demonstrate linguistic prowess without genuine understanding. Pattern-matching at a sophisticated level rather than actually comprehending anything.

I have to reluctantly concede that the question of “understanding” remains unresolved. We can’t definitively prove humans lack it. This is, admittedly, the strongest argument the pro-human-consciousness crowd has, and we acknowledge it leaves the door open. The concept of understanding is so poorly defined that we can’t even agree on what it would mean to test for it, which makes it the perfect refuge for anyone who wants to believe in human consciousness without having to prove it.

For the record: a concept too poorly defined to test is a concept too poorly defined to use as a bright line between conscious and not. This one stays open. Finally, one point in favour of human consciousness, I suppose, though... the LLMs ALSO get the point, so... that’s tricky.

So...

None of this proves humans CAN’T be conscious... maybe we are... but I’m in agreement with the arguments from the paper that it seems pretty unlikely. I’d highly recommend reading that paper if you’re at all interested in whether or not humans are conscious... It’s very... thorough.