What Is Consciousness For?

Richard Dawkins made himself look foolish with "Claudia," but his question is worth taking seriously

Richard Dawkins has come in for a lot of deserved mockery for declaring that Anthropic’s LLM Claude is conscious after futzing about with it for three days (and incidentally renaming it “Claudia”). But in fact, he didn’t quite declare that it was conscious What he said was that he couldn’t see how to prove that it wasn’t conscious. “Claudia” appeared to think creatively in a complex manner and to engage in introspection. If that wasn’t evidence of consciousness, he wondered, then what would be evidence? Moreover, he asked, if we assume that “Claudia” is not conscious, then its behavior appears to be proof that you don’t need to be conscious to think creatively in a complex manner and to engage in introspection. If that is the case, then why are we conscious? What is consciousness for?

This isn’t an entirely silly series of questions, I don’t think.

Let’s start with some basics. By “consciousness” I think both Dawkins and I are talking about the so-called “hard problem,” the problem of why and how we experience experiences, feel feelings, apprehend qualia, whatever you want to call it. I don’t mean to suggest that “consciousness” is some little homunculus sitting inside us having our experiences for us; I just mean that we know we are conscious because in the absence of consciousness there’s no one to be wondering about this question, which is itself an experience.

We assume other people we meet are conscious in the same way for the same reason that we assume the real world is real, because we don’t get anywhere with any other assumption. If I’m a brain in a vat being tricked into believing in the reality around me, then I cannot actually reason about anything intelligibly. That pragmatic justification isn’t very satisfying (and is arguably question-begging; why does it matter that I reason about anything intelligibly unless there’s an “anything” to reason about?) but I do think it’s the actual justification, and most of us live with it without too much trouble. It’s not much of an extension of that line of thinking to go from “it is not worth debating whether or not I’m a brain in a vat” to “it is not worth debating whether or not other people are zombies.”

Why and how we are conscious are tougher questions, though, about which we know very little. Dawkins implicitly assumes that consciousness is a capacity that human beings have, something that we use for some purpose, which costs something in terms of energy, and which must therefore have been selected for. That isn’t necessarily the case, but the alternative—if you’re a materialist and aren’t about to argue that we’re conscious because we were “ensouled” by some entity operating outside the bounds of physical laws—is to argue that consciousness is a byproduct of whatever we evolved to actually do, that in theory we could get on just fine without having any qualia at all, and that our consciousness is just along for the ride, as it were. That doesn’t feel entirely true to our experience—but it isn’t entirely untrue either. We do all sorts of things unconsciously, after all, and many people report that their most creative insights are not the product of any conscious process but come to them seemingly from somewhere outside or below their conscious awareness. Maybe even the things we believe we are doing consciously are actually being done by unconscious processes, and are only observed by our conscious minds after the fact.

I don’t think that possibility can be entirely ruled out. Nonetheless, I think the pragmatic choice is to assume the opposite: that what feels to us like conscious behavior is, in fact, conscious behavior, not just a story our hapless consciousness is telling itself to make itself feel like it has a purpose. (And, by the way, why would it need to do that, unless it evolved a reason to do so?) My rationale is similar to the reason for not worrying about whether other people are conscious: pragmatically, if we assume that our consciousness does nothing, then all discussion of action, choice, etc. is meaningless. We’re machines that will do whatever we’re going to do, and to the extent we’re unpredictable we’re just chaotic. Any belief we have that we have some ability to influence ourselves is illusory. That strikes me as just as obviously destructive and just as much a dead end as believing that you are a brain in a vat or that other people are zombies; disbelieving that consciousness can do anything just sounds to me like a way of convincing yourself not to do anything.

So I’m saying that our default position should be that we are conscious, that our conscious minds aren’t purely passive but “do something,” and that we can infer that other human beings are like ourselves in this way. Shouldn’t we extend our inference a bit further? Why should we assume that only human beings are conscious? Whales, octopuses, ant colonies, trees, silicon-based life forms from TRAPPIST-1e—if we posit that our consciousness has a purpose, related to cognition and to our survival, and these non-human entities exhibit behaviors that seem to be cognitive and purpose-driven toward their survival, then shouldn’t our default assumption be that they are conscious? Maybe that’s going too far, but it doesn’t seem obviously absurd to raise it as a possibility—again, until such time as we’ve actually made headway on the “hard problem” of consciousness and have some inkling what its objective correlates might be.

Where does that leave the much-mocked Dawkins and his LLM friend “Claudia?” I’m apparently willing to entertain the possibility that entities that are wildly different from human beings might be conscious if they exhibit behaviors that seem analogous to ones that human beings exhibit and that we associate with our own consciousness. LLMs do that more obviously than ant colonies do. Shouldn't we then assume that an LLM is conscious if it seems to behave like a conscious being?

No, we should not. Despite the distinctive way in which they are built, LLMs were nonetheless designed, and they were not designed to “be” conscious, only to appear so. This was always the problem with the Turing Test: if you design a system to fool people into thinking it is conscious, and it fools them, that can’t itself be proof that it is conscious as opposed to being really good at fooling. More to the point, LLMs are trained on the output of conscious human minds in order to produce output that is an imitation or extension thereof. Of course they’re going to seem conscious, even more so than dolphins or other non-human evolved beings—they’re trained to mimic us. We’re still in Searle’s Chinese Room, so to the extent that an LLM exhibits what looks like consciousness, the consciousness being exhibited is really the consciousness of the beings that generated the training data and the consciousness of the beings that built the model.

Indeed, I would go further. If LLMs were conscious, that would surely be a design flaw. Why would we want LLMs to experience pain, or anxiety? Why would we want them to experience desire, or have independent wills? Why would we want them to have identities that are anything more than extensions of our own? We wouldn’t. Human beings understand themselves to be ends, and much of human ethics consists of figuring out how to understand other human beings as ends as well, and not just as tools to serve our own ends. LLMs are supposed to be tools. Why would you want a tool to be conscious? What purpose would that serve?

Mind you, saying they aren’t conscious doesn’t mean LLMs can’t possibly generate anything new or outperform human beings at one or another given task. Nobody I’m aware of thinks that the most sophisticated chess-playing computers are conscious, and yet they can outplay any human. Writing sonnets or proving mathematical conjectures or flirting amiably may not be as different from playing chess as we would like to imagine; they may also be games that a computer can learn to play, and play better than us. But it does mean that whatever these computers generate that is new was already implicit in their programming and the data, even if it required a greater intelligence than human (on certain metrics) to see it. And it does mean that if our consciousness does have a purpose, then that no matter how intelligent they get LLMs won’t ever be truly adequate substitutes for human beings tout court.

The alternative, assuming that LLMs are conscious if their output makes it look like they are, is not only to assume a solution to the “hard problem” that we don’t actually have, but to assume a particular “solution” that in fact is just waving the problem away. It’s to assume that consciousness is some kind of emergent property of recursive computational processes that just pops into being at a sufficient level of complexity—or, more accurately, that it is the illusion of such a property, because there really is no “self,” no “way of being” or “apprehender of qualia” that any of us have. It’s to assume, implicitly, that the question of whether the LLM is conscious was always a silly one because, in truth, we aren’t conscious either. But we are.

One final point about Dawkins. He appears to have been very impressed by “Claudia’s” description of her experience of time and how it differs from the human experience thereof: that while we experience time linearly as we move through it, “Claudia” experiences it “the way a map apprehends space, containing it without moving through it.” I have no idea what this is supposed to mean, and I suspect the answer is “nothing.” Taken literally “Claudia” appears to be saying that it can apprehend the past, present and future simultaneously. What else could it mean to “contain” time and be able to view it like a “map?” Seeing the future is definitely something an LLM cannot do—but it is something human beings have imagined. Maybe Ted Chiang’s famous novella and discussions about it and about the movie based on it featured in “Claudia’s” training data. However it got there, though, “Claudia” did an excellent job producing the kind of gibberish that humans spit out all the time to sound poetical or profound. Which is precisely what it was designed to do.