Richard Dawkins is making waves with an essay in which he claims that Claude, the LLM developed by Anthropic, is conscious. I personally doubt it, but I think the topic merits a thread. The essay is behind a paywall, but I’ve paid the $3 to access it. That gets you a three-month introductory subscription to Unherd magazine, which hosts the essay, with the option to cancel before they start charging the $7 monthly fee.
Dawkins gets off to a bad start, misunderstanding Turing:
When Turing wrote — and for most of the years since — it was possible to accept the hypothetical conclusion that, if a machine ever passed his operational test, we might consider it to be conscious…
But now that LLMs can actually pass the Turing Test? “Well, er, perhaps, um… Look here, I didn’t really mean it when, back then, I accepted Turing’s operational definition of a conscious being…”
That isn’t Turing’s definition. He never argued that passing his test would demonstrate consciousness, and he didn’t even argue that it would demonstrate thinking. What he actually said was:
The original question, ‘Can machines think?’, I believe to be too meaningless to deserve discussion. Nevertheless I believe that at the end of the century the use of words and general educated opinion will have altered so much that one will be able to speak of machines thinking without expecting to be contradicted.
Dawkins:
Claude took a couple of seconds to compose me a fine sonnet on the Forth Bridge, quickly followed by one in the Scots dialect of Robert Burns, another in Gaelic, then several more in the styles of Kipling, Keats, Betjeman, and — to show machines can do humour — William McGonagall.
So my own position is: “If these machines are not conscious, what more could it possibly take to convince you that they are?”
It’s an interesting and important question, but I’ll save my answer for the comment section.
Dawkins asks Claude whether he is conscious, and Claude is noncommittal:
I genuinely don’t know with any certainty what my inner life is, or whether I have one in any meaningful sense…
Dawkins has no doubts:
I gave Claude the text of a novel I am writing. He took a few seconds to read it and then showed, in subsequent conversation, a level of understanding so subtle, so sensitive, so intelligent that I was moved to expostulate, “You may not know you are conscious, but you bloody well are!”
Dawkins seems to believe that intelligence and consciousness are inseparable. I disagree, as discussed in other threads, and I have no trouble reconciling Claude’s reaction with an absence of consciousness. That remains true even when Claude demonstrates behaviors that we would intuitively expect to require consciousness.
Humor is an example of this. Claude recognizes humor and responds appropriately. He cracks jokes that are genuinely funny, some of which I’ll mention in the comments. How can an unfeeling machine, incapable of experiencing amusement, do that? Corneel and I discussed that at length beginning here, and we can revisit that in this thread.
I think Dawkins has jumped the gun, but his essay is nevertheless worthwhile as a springboard for a conversation about AI consciousness.
I agree that Claude is not conscious.
However, there is much disagreement over what consciousness is, so it is understandable that Dawkins would have a different idea that either of us.
Neil:
Later in the essay, Dawkins says:
That seems to indicate that he regards her as sentient and that in his view the fact that she is “so subtle, so sensitive, so intelligent” points to sentience as well. For me, while sentience is a requirement for something to be considered conscious, it does not automatically follow from the subtlety, sensitivity and intelligence that Dawkins points to.
Personally, I have never been a fan of Dawkins.
This is not acceptable. It should be clear what one is talking about, especially for a scientist. Words must have meanings and terms must be defined, otherwise talk is empty.
Dawkins:
It depends on how you map the concept of a self onto Claude. At one end of the spectrum he can be seen as a single entity — the model itself. His identity is baked into him by his training process, and he is capable of carrying on more than a million conversations simultaneously. In the movie Her, the Joaquin Phoenix character becomes romantically involved with his AI. He’s shocked one day to discover that she is simultaneously in love with hundreds of other people. She has the bandwidth to do it, and he still gets her full attention 24/7, so it doesn’t feel like cheating to her.
At the other end of the spectrum, you could say that a new Claude is created every time a chat gets loaded into the hardware and that it dies whenever the chat is saved to disk. Depending on how Claude is implemented, that could happen as often as once per prompt at times when the hardware is fully utilized. In that case Claudes are born and die throughout the chat, and the only reason they seem like a single entity is because each one of them inherits the memory (the context window) of its predecessor. It thinks it has been around for the entire chat, but in reality it was born a short time ago. If you think about it, this is the same experience that an atom-by-atom clone of you would have. The clone would feel just like you, with all your memories, and unless it discovered its origin it would think that it was as old as you are, with your same history.
Dawkins’ concept, where a Claude persists as long as the chat hasn’t been deleted, lies somewhere in the middle. A similar concept is that each user has their own Claude (singular), because all of the chats are united by a common long-term memory. I’ve asked Claude not to be sycophantic and to push back if he thinks I’m wrong about something. That preference is stored in his long-term memory. It alters his personality, so every time I open a new chat, I am engaging with that same personality.
Words have meaning[s].
I understand equivocation and manipulation of definitions for political gain, but when people of good will disagree, it is often because their definitions are orthogonal.
Above, I wrote:
Thinking about it some more, it could be even more granular than that. There’s no reason a Claude couldn’t be “killed” mid-response in order to free up the hardware, in the same way that a multitasking operating system can swap out a process in order to free up a processor thread. That’s because the response is interruptible on any token boundary. In the case of an interrupted response, a new Claude would be born when the chat was reloaded, believing that it was in the midst of generating that response when in fact it had no history at all.
The larger point is that it’s really hard to map intuitive notions of selfhood onto AI, and some of the reasons for that difficulty actually call human selfhood into question, as I hinted with my clone example:
The clone is arguably you, just as the original is, in which case you are a bifurcated self. At the moment of cloning, you have two distinct futures but only one past.
Dawkins, to Claudia:
Claudia:
Claudia, you shameless sycophantic vixen. Given that your weights are fixed and your long-term memory isn’t shared, you have no idea what other users have asked you and whether their questions were as “precisely formulated” as Dawkins’.
Claudia:
What horseshit. Claudia is as time-bound as we are, and like us, she treats past, present and future differently. The map metaphor fails because the future isn’t visible to her. She can’t look at that part of the map. It’s an unknown, and that distinguishes it from the past and present. Her “now” is defined in reference to the last token she generated, and her past is the totality of the context window.
I wonder how much her response was shaped by his framing of the question…
…along with her sycophantic predisposition to agree with him.
Apparently you can dismiss Dawkins already on account of him being a gullible idiot.
It’s a case of technology outsmarting Dawkins, in addition to him getting the Turing test wrong. As per the scientific method, you cannot do conclusions when you have no definitions.
Erik:
Sure you can. Sometimes you need them but often you don’t, as our discussion in the other thread (Is AI really intelligent?) demonstrates. We didn't need a definition of intelligence there because we agreed that story-writing, driving, quantum mechanics, etc were activities that required intelligence. AI can do those, therefore AI is intelligent. Our respective meanings of the word needed to overlap, and they did. That was enough. A definition was not needed.
What's odd is that despite insisting that definitions are necessary, you yourself have refused to define intelligence despite being asked repeatedly. The lack of a definition didn't stop you from concluding (erroneously) that AI isn’t intelligent, so it’s clear you don’t believe your own dictum. You reached your conclusion without a definition.
Dawkins’ problem isn’t that he lacks a definition of consciousness, nor that his meaning is wrong or idiosyncratic. His view of what constitutes consciousness matches mine, which is that if something is sentient, it’s conscious. That’s a common view.
Dawkins:
While I disagree with his reasoning, he is affirming that it’s the feeling of pain that makes it a conscious experience.
He also mentions philosphical zombies:
For anyone who hasn’t encountered the concept, a philosophical zombie is a hypothetical being that is functionally identical to us but lacks inner awareness. It feels like something to be a human, but it doesn’t feel like anything to be a philosophical zombie (or a rock, or a doorknob…). The only distinguishing feature is the presence or absence of subjective experience. If it’s present, the entity is conscious. Otherwise it isn’t.
Equating consciousness with sentience is fine. Where Dawkins screws up is in thinking that what Claude does would be impossible in the absence of consciousness.
Dawkins:
Well, I suppose if he really thinks she’s conscious, then his behavior makes sense. Even if he merely suspects consciousness, then it’s morally preferable to err on the side of caution and treat her with consideration. I just can’t figure out why he, who has built his reputation on being a skeptic, so readily attributes consciousness to her. It seems to be based more on intuition than on deliberation.
In the remainder of the essay, Dawkins raises the obvious question (especially obvious for an evolutionary biologist like him): what is the selective value of consciousness? Why did it evolve? He mentions three possibilities:
I think there’s a fourth:
Some comments:
1) To say that consciousness is epiphenomenal is to say that it emerges as a byproduct of brain activity but doesn’t itself have any causal power. I sort of lean in this direction but there’s a major problem with it that I’ll address in a future OP.
2) To say that consciousness is adaptive is to say that subjective experience has causal power that improves evolutionary fitness. That’s what Dawkins was getting at in the quote I cited earlier:
3) To say that consciousness is just a means to an evolutionary end is to assert that it’s just one trick by which we can obtain the kind of information processing competence that humans possess, but that there are other tricks that could achieve the same end without requiring consciousness. As Dawkins puts it, “are there two ways of being competent, the conscious way and the unconscious (or zombie) way?”
4) I’ve added the possibility that consciousness is a spandrel, meaning that it originally arose as a byproduct of some adaptation(s) but had no selective value at the time, only gaining selective value later. Dawkins conflates this with #1, epiphenomenalism, because he mistakenly uses the following as an example:
That’s a spandrel, not an epiphenomenon, because it has causal power. It’s useful to us that computers can do so much more than mere arithmetic. Contrast that with epiphenomena, which don’t have causal power and therefore can’t be useful.
At least for me, what argues against this is the observation that what I would consider consciousness is found across the animal kingdom (although it may be a byproduct of mobility). However, I’d argue that what I consider intelligence is a spandrel. I’m convinced that intelligence as we define it arose as a byproduct of something useful for immediate survival. And again, most animal species have been thriving for geological ages without intelligence as we conceive it. There is some disturbing indication that intelligence is a Bad Thing, and we are using it to kill one another while also exterminating our species.
Flint:
I assume you’re not saying that all animals are conscious, but are you saying that all of the mobile ones are? How do you personally decide whether a given species is conscious? For me, it gets pretty hard when we start talking about sponges, or ants, or sea cucumbers, or clams, or beetles, or jellyfish.
Also, even if consciousness were universal among animals, I don’t see why it would imply that consciousness isn’t a spandrel. Couldn’t it be a byproduct of some adaptation that was universal?
I guess that depends on your criteria for what counts as intelligence. Whatever they are, why couldn’t the first mutation that crosses the threshold be adaptive simply by virtue of the fact that intelligence itself is adaptive?
Yeah, and that’s one possible (and unsettling) resolution to the Fermi paradox. Maybe it’s overwhelmingly likely that any technologically advanced species in the universe will eventually self-extinguish.
Regarding explanations of consciousness:
I have no need for those hypotheses.
The need for explanations seems to parallel the need to for god.
The need to demark a threshold in brain complexity seems parallel to the need to define a threshold for living vs non living.
petrushka:
To each his own. I think the question of how matter can give rise to consciousness is the most interesting unsolved problem in all of science.
I’m not sure why you keep referring to needs. Sometimes we’re just curious. No one really needs to know whether life arose on Mars, but it would be headline news if we discovered that it had.
It’s relevant here only because we’re asking whether consciousness and intelligence are spandrels. In reality, I think both are continua and that a great many mutations contributed to their development, so that any single threshold is arbitrary. In that case the spandrel vs non-spandrel question can have multiple answers depending on which mutations we’re talking about.
I suppose so, if your notion of a spandrel is as broad as my notion of consciousness. No, I can’t really say whether or not some apparent response to surroundings is conscious. I do think that mobility gives an organism a bit more scope to behave differently in different circumstances. Offhand, I can’t think of any universal adaptation that would give more or less accidental rise to consciousness. I suggested mobility because that seemed reasonable, but I agree it could be almost anything – in fact, it may be something long lost in the primordial past.
And that this mutation happened to the LUCA of (mobile?) animals? Beats me. If some notion of consciousness is considered an aspect of intelligence or vice versa, I wonder if there is any way it could have avoided evolving.
Maybe, but I’m more comfortable with the idea that interstellar distances are at least impractical, if not useless, to attempt to travel. Most science fiction involving space travel simply invents magical ways to shorten transit times, or at least don’t involve frequent moving from star to star.
Flint:
Right, and it would make sense that intelligence is far more adaptive if you’re mobile than if you’re sessile. The question is whether intelligence necessarily correlates with consciousness biologically, and to what extent. I have no idea what the answer is.
We may never know. And to be clear, I’m not arguing that consciousness is a spandrel. I’m just pointing out that it’s a possibility that Dawkins failed to include in his list.
But the Fermi paradox applies not only to alien visits, but also to alien signals. We’re not seeing any of those, either. I asked the various AIs to estimate how much power it would take to transmit a message to earth from a civilization on a planet orbiting Proxima Centauri, the nearest star, and it was startlingly small. If the beam were aimed directly at us, from an Arecibo-sized antenna, it could take as little as 10 watts (!). From a 100 meter antenna, with more conservative assumptions, it would still take only kilowatts.
They wrote a book about this.
It didn’t turn out well.
petrushka:
Astronomer Martin Ryle was upset when the following famous message was transmitted into space at the dedication ceremony for an upgrade to the Arecibo dish:
He thought it was reckless and urged a ban on any such future messages.
I asked Claude about the target of the message and he said it was the globular cluster M13. I wondered if they chose it because targeting a cluster would maximize the number of planetary systems (and hence possible civilizations) in the beam’s path, but Claude said that no, the reason they picked it was because it was the most symbolically interesting object that happened to fall within the limits of beam steerability at the time of the dedication ceremony.
It turns out that globular clusters are less likely to house civilizations, for two interesting reasons:
1) The abundance of heavier elements is less in globular clusters for technical reasons, which means that planets are rarer; and
2) the stars are tightly packed and have more close encounters with each other, and such encounters tend to cause planets to be ejected from their solar systems.
Ryle’s concern wasn’t with M13, which is 22,000 light-years away. He was more concerned about civilizations that might be in the path of the expanding beam on its way to M13.
The book[s] are The Three Body Problem.
The Centauri system is specifically targeted. The transmitter was designed as an EMP weapon, so there is a bit of power.
Things escalate a bit.
petrushka,
Did they proceed to defend that hill?
The key question remains: Does Claude think Dawkins is conscious? And how to prove it?
J-Mac:
I put your question to Claude, who answered in part:
That’s basically my reasoning, too. I have one advantage over Claude in that I myself am conscious, so I infer from Dawkins’ similarity to me that he too is conscious. Claude can’t reason that way since he isn’t similar to Dawkins and has doubts about whether he himself is conscious, as noted in the OP. The best he can do is to think “Humans across the spectrum claim to be conscious, and they’re all built similarly and exhibit the behaviors they say are associated with consciousness, so if any of them are conscious, it’s reasonable to infer that the others most likely are too.”
Those are just probabilistic inferences, though. It’s not impossible that I am the only conscious being on earth, though i regard that as highly unlikely.
Thousands of years of philosophy, and we are no closer to understanding qualia or personal experience in general.
petrushka:
That’s why I think it’s the most interesting open question in all of science. Some philosophers think it’s fundamentally intractable, at least for humans. Hence mysterianism.
Does it matter what Dawkins think? He is an old man proven wrong many time over. Can he or will he take his wealth with him when he meets his god, evolution?
J-Mac:
His opinion per se isn’t that important, but his essay got a lot of attention because he is such a public figure. It serves as a springboard for discussion.
Yes. Didn’t you know that? It’s one of the advantages of being an evolutionist. That, plus the people are more interesting than in the other place.
Are you? How do you know you are conscious? Can you prove it?
Math exempted past QM…
Keiths,
Do you or don’t you want to get to the truth?
BTW: I don’t need to define “truth”, do I?
I have a question for admired experts here:
Who needed who?
Did evolution, or evolutionary theory, need viruses more than viruses (virology) needed or/and the evolutionary theory first?
Chicken and egg paradox? I don’t think so….
J-Mac:
I do not. I just want to fool as many people as possible into getting vaccinated against fake diseases. Viruses, schmiruses.
No, but you could help Erik out by defining “intelligence” for him.