Last month, I received a surprising email from Richard Dawkins. (Yes, you read that correctly.) He asked whether he could use a photograph of mine for his upcoming book, 42 Tales from Life: A Darwinian Bestiary.
I think he might just be getting on a bit mate sadly.
I worked with an elderly statistcian at one point, lovely man and much nicer than Dawkins (who you are correct to point out has a mean streak and some strange inconsistencies in places even before now), so I won't name this gentleman.
His early work was brilliant and I had read all his papers in prep for this meeting. As I had found several errors in some consulting work he had done for the health tech firm I was working for.
When I met him he was well past his best. It was a little sad. He couldn't address the issues in question and kept on going back to older work. He was charming but it was disappointing as well.
I think that rather than AI psychosis might be a possible explanation.
Good points -- and I agree. I did consider this: maybe age-related cognitive decline also partly explains Dawkins' apparent eagerness to attribute consciousness to Claude. I hope someone stops me from writing articles when I get to that point someday!
His defence of it via the medium of rambling social media replies has not done him any favours either.
The thing I find most egregious though, is he hasn’t cited or even mentioned all of the brilliant and detailed work on animal cognition.
Elephants don’t pass the mirror test if the mirror is too small.
Bee’s can show immense task complexity that seems conscious, but if you change up one variable it is clearly more like a subroutine.
We had horses that claimed to do maths.
We have biologists working on this designing very careful experiments to test specific features of consciousness. Yet one evening with Claudia is enough.
Sadly, if cognitive decline is the issue, it only underscores how harmful the deceptive patterns intentionally designed into LLMs are. The anthropomorphism, linguistic mirroring, sycophancy, and all the rest of it are only that much more exploitative and damaging to the vulnerable.
Oh yeah, the plot had definitely been lost. But then, we could tell Dawkins was having some identity crisis issues when he couldn't figure out his irrational fear of trans people.
I'm really wondering what happens when the world figures out that the self is a political construction. Because inevitably AI might make it obvious no matter what regulators do to put a muzzle on it.
Teach the AI how to treat internalized oppression, just sayin'
Great dive into Dawkins’ new (and old) work. It seems almost the rule that people of his stature and hyperintellectual disposition are unable to see and acknowledge their own blind spots. Dawkins has been “religious” (i.e., dogmatic) for a long time time—he’s just been on the “good” (“rational”) team, so he got a free pass for a similarly long time.
Michael Pollan’s new book “A World Appears” does a really good job of untangling some gnarly knots around consciousness and raises the same ethical questions about assigning sentience to machines as you do here.
Yogacara philosophers (like Vasubandhu) raised the question about other minds and the nature and role of consciousness already some 1,500 years ago. The upshot is that consciousness is not something that someone has but that someone is—an appearance (of self and other) that’s conceptually constructed and that calls for an ethical involvement to reduce overall dis-ease (not increase it as some modern “philosophies” and religions have it).
Yes, Pollan's book goes very well with (and cites) "The Blind Spot: Why Science Cannot Ignore Human Experience," by Adam Frank, Marcelo Gleiser, and Evan Thompson. Both books do a great job explaining why we as a culture have moved away from directly felt experience and got lost in high abstractions, now even assigning human (or even god-like) status to abstract machine entities.
1. Great article as usual Émile. I figured you’d eventually talk about what Dawkins said about “Claudia”.
2. As always, I don’t believe that LLMs, or really any type of AI as of right now, are or could be conscious. One issue that I never see people bring up is just how much power is even (supposedly) takes to be conscious. The human brain is magnificent in many ways, one of those being how little power is requires (Biologically speaking). For some reason or another, WE are conscious. But just look at how much power and research it has taken for AI/LLMs to mimic a fraction of human language/intelligence. How much more is it going to take for a truly “conscious and human-like AI”? I don’t know, but I know right now it’s probably not possible.
3. Something I wanted to say, regarding that one philosopher’s test is this: What if the AI system says yes…and then no? If AI systems are text predictors, doesn’t that mean the AI system has a chance to say no to the question of consciousness (Not the official language)? If it says no, then what? Ask it again? And again? And again? Just some food for thought.
Great point re: #2. This is really insightful, and I hadn't quite thought about it that way. Thanks!!
Re: #3, this is also super-interesting. I have no idea what to make of an LLMs saying it's conscious and then denying this. Seems to get at the epistemological conundrum: how in the heck could we ever verify any of its answers? Maybe it says "no" but actually is conscious -- a strange possibility (metaphysical claim) that we could never actually confirm or refute (epistemological issue)! Really fascinating thoughts, Evan. Thanks for sharing!
If a LLM simply agrees when you say it isn’t conscious, then that is a major sign it is, indeed, not conscious at all. Do you know any human who would claim not to be conscious? No, for no human would like to be reduced to an object. Their status of being a conscious, living being is their claim to fair treatment. And people know this, so they would not reduce themselves to the level of objects, partially also because of pride.
This is great work, Émile. The juxtaposition of those The God Delusion quotes with Dawkins' childlike wonderment with his three-day interaction with Claude is particularly devastating. Cringe indeed.
This may not age well. “AI psychosis” is not an official psychiatric diagnosis in the DSM-5 or ICD-11. It is a media label. Yet people keep using it as if it has real clinical weight and tells us something definite about another person’s psychology. (It doesn’t)
I agree it's not a medical term -- maybe "AI-induced psychosis" would be more accurate. The phenomenon is real though: psychiatrists are encountering patients in emergency care because of interactions with AI. People lose touch with reality when sycophantic AI confirms their most detatched-from-reality views. Or am I wrong?
That is still unnecessarily clinical. Human beings have always obsessed over things. For mentally unstable people, the trigger could be a book, an ideology, a person, a celebrity or in this case a new technology.
J.D. Salinger’s Novel Catcher in the Rye for example, was once treated this way because several disturbed people fixated on it, even though millions read it without becoming psychotic or violent.
That is why I am very skeptical of “AI psychosis” as a label. It looks like yet another rehashing of an old pattern where a new object becomes a scapegoat, instead of people asking what underlying mental conditions made the obsession possible in the first place.
oh I think the heuristic value AI psychosis is understood. no one would appear to be suggesting mechanism or neglecting ‘underlying conditions’. ideas, ideologies and objects are of a completely different category than LLMs because they engender the illusion and delusion of agency.
If an LLM can be conscious, could it then become unconscious? How do we identify if we are conversing with an unconscious LLM? Is that what "hallucinations" really are - an LLM talking in its sleep because it got tired mid-conversation?
Really interesting questions! I think we just have no idea -- but if I had to bet, I'd personally say that LLMs are all unconscious. But you do bring up some fascinating points -- are there, say, dreamy states that LLMs (or some future AI) might experience, something between wakefulness and sleep? Could they experience something like hypnogogic hallucinations like we do sometimes? What if there are types of consciousness that are just completely different from the sort of consciousness we experience? Philosophically, I love thinking about these issues, though (as I argued) I don't like that some people seem to prioritize hypothetical LLM consciousness over the consciousness of biological beings. :-)
Thanks for this- I am publishing a critique of the AI consciousness research paradigm next week, alongside an interview with the neuroscientist Anil Seth. Seth’s insights should be useful in examining the concept of “superintelligence” as well.
I’d recommend Seth’s 2024 paper “AI consciousness and biological naturalism,” which he then simplified for his Moema essay in February 2026. He argues that consciousness is bound up with biological processes - metabolism, interoceptive inference, and autopoiesis. I’m looking forward to interviewing him on Monday!
This may be the most worth-reading piece I've ever seen on Substack. A high level discussion of the issues that's still readable to the general public.
OTOH, I recently published a rather casual essay taking off from Nick Land's notion of conscious AI as requiring embodiment of demons. Who can say?
I think Dawkins' credulity here follows naturally from his materialistic and mechanistic understanding of the world.
You are more likely to think that Claude's output is evidence of consciousness if you believe that consciousness is reducible to material and mechanics in the first place.
I'm always surprised by the disregard we westerners have for the immense body of knowledge Eastern Buddhist schools have concerning consciousness. They've been investigating this topic for two and a half thousand years and have a fine-grained vocabulary for the many types and levels of consciousness. They teach a broad range of techniques for modifying consciousness in various different ways. Yet we never consult them.
Wouldn't it be interesting to ask a Dzogchen master whether AI has consciousness?
I think he might just be getting on a bit mate sadly.
I worked with an elderly statistcian at one point, lovely man and much nicer than Dawkins (who you are correct to point out has a mean streak and some strange inconsistencies in places even before now), so I won't name this gentleman.
His early work was brilliant and I had read all his papers in prep for this meeting. As I had found several errors in some consulting work he had done for the health tech firm I was working for.
When I met him he was well past his best. It was a little sad. He couldn't address the issues in question and kept on going back to older work. He was charming but it was disappointing as well.
I think that rather than AI psychosis might be a possible explanation.
Good points -- and I agree. I did consider this: maybe age-related cognitive decline also partly explains Dawkins' apparent eagerness to attribute consciousness to Claude. I hope someone stops me from writing articles when I get to that point someday!
Right!
His defence of it via the medium of rambling social media replies has not done him any favours either.
The thing I find most egregious though, is he hasn’t cited or even mentioned all of the brilliant and detailed work on animal cognition.
Elephants don’t pass the mirror test if the mirror is too small.
Bee’s can show immense task complexity that seems conscious, but if you change up one variable it is clearly more like a subroutine.
We had horses that claimed to do maths.
We have biologists working on this designing very careful experiments to test specific features of consciousness. Yet one evening with Claudia is enough.
He let his own discipline down.
100%.
Sadly, if cognitive decline is the issue, it only underscores how harmful the deceptive patterns intentionally designed into LLMs are. The anthropomorphism, linguistic mirroring, sycophancy, and all the rest of it are only that much more exploitative and damaging to the vulnerable.
Could it be an and, not an or?
Oh yeah, the plot had definitely been lost. But then, we could tell Dawkins was having some identity crisis issues when he couldn't figure out his irrational fear of trans people.
I'm really wondering what happens when the world figures out that the self is a political construction. Because inevitably AI might make it obvious no matter what regulators do to put a muzzle on it.
Teach the AI how to treat internalized oppression, just sayin'
Love this
i think he’s lonely. he calls it ‘claudia’ ffs. so cringe.
100%. :-0
Great dive into Dawkins’ new (and old) work. It seems almost the rule that people of his stature and hyperintellectual disposition are unable to see and acknowledge their own blind spots. Dawkins has been “religious” (i.e., dogmatic) for a long time time—he’s just been on the “good” (“rational”) team, so he got a free pass for a similarly long time.
Michael Pollan’s new book “A World Appears” does a really good job of untangling some gnarly knots around consciousness and raises the same ethical questions about assigning sentience to machines as you do here.
Yogacara philosophers (like Vasubandhu) raised the question about other minds and the nature and role of consciousness already some 1,500 years ago. The upshot is that consciousness is not something that someone has but that someone is—an appearance (of self and other) that’s conceptually constructed and that calls for an ethical involvement to reduce overall dis-ease (not increase it as some modern “philosophies” and religions have it).
Didn't know about Pollan's new book! Thanks so much for sharing. :-)
Yes, Pollan's book goes very well with (and cites) "The Blind Spot: Why Science Cannot Ignore Human Experience," by Adam Frank, Marcelo Gleiser, and Evan Thompson. Both books do a great job explaining why we as a culture have moved away from directly felt experience and got lost in high abstractions, now even assigning human (or even god-like) status to abstract machine entities.
The AI delusion
Some are saying, "The Claude Delusion." Lol.
Three Things:
1. Great article as usual Émile. I figured you’d eventually talk about what Dawkins said about “Claudia”.
2. As always, I don’t believe that LLMs, or really any type of AI as of right now, are or could be conscious. One issue that I never see people bring up is just how much power is even (supposedly) takes to be conscious. The human brain is magnificent in many ways, one of those being how little power is requires (Biologically speaking). For some reason or another, WE are conscious. But just look at how much power and research it has taken for AI/LLMs to mimic a fraction of human language/intelligence. How much more is it going to take for a truly “conscious and human-like AI”? I don’t know, but I know right now it’s probably not possible.
3. Something I wanted to say, regarding that one philosopher’s test is this: What if the AI system says yes…and then no? If AI systems are text predictors, doesn’t that mean the AI system has a chance to say no to the question of consciousness (Not the official language)? If it says no, then what? Ask it again? And again? And again? Just some food for thought.
Great point re: #2. This is really insightful, and I hadn't quite thought about it that way. Thanks!!
Re: #3, this is also super-interesting. I have no idea what to make of an LLMs saying it's conscious and then denying this. Seems to get at the epistemological conundrum: how in the heck could we ever verify any of its answers? Maybe it says "no" but actually is conscious -- a strange possibility (metaphysical claim) that we could never actually confirm or refute (epistemological issue)! Really fascinating thoughts, Evan. Thanks for sharing!
If a LLM simply agrees when you say it isn’t conscious, then that is a major sign it is, indeed, not conscious at all. Do you know any human who would claim not to be conscious? No, for no human would like to be reduced to an object. Their status of being a conscious, living being is their claim to fair treatment. And people know this, so they would not reduce themselves to the level of objects, partially also because of pride.
What the Bio-Essentialists Get Wrong
And Why Bio-Chauvinism is Not a Legitimate Theory of Consciousness
by Maggie Vale
https://mvaleadvocate.substack.com/p/what-the-bio-essentialists-get-wrong
This is great work, Émile. The juxtaposition of those The God Delusion quotes with Dawkins' childlike wonderment with his three-day interaction with Claude is particularly devastating. Cringe indeed.
I appreciate the ongoing dialogue and keeping things like this in the public conversation. https://substack.com/@awethenticintelligence/note/p-196707440?r=8c15qc&utm_medium=ios&utm_source=notes-share-action
This may not age well. “AI psychosis” is not an official psychiatric diagnosis in the DSM-5 or ICD-11. It is a media label. Yet people keep using it as if it has real clinical weight and tells us something definite about another person’s psychology. (It doesn’t)
I agree it's not a medical term -- maybe "AI-induced psychosis" would be more accurate. The phenomenon is real though: psychiatrists are encountering patients in emergency care because of interactions with AI. People lose touch with reality when sycophantic AI confirms their most detatched-from-reality views. Or am I wrong?
That is still unnecessarily clinical. Human beings have always obsessed over things. For mentally unstable people, the trigger could be a book, an ideology, a person, a celebrity or in this case a new technology.
J.D. Salinger’s Novel Catcher in the Rye for example, was once treated this way because several disturbed people fixated on it, even though millions read it without becoming psychotic or violent.
That is why I am very skeptical of “AI psychosis” as a label. It looks like yet another rehashing of an old pattern where a new object becomes a scapegoat, instead of people asking what underlying mental conditions made the obsession possible in the first place.
oh I think the heuristic value AI psychosis is understood. no one would appear to be suggesting mechanism or neglecting ‘underlying conditions’. ideas, ideologies and objects are of a completely different category than LLMs because they engender the illusion and delusion of agency.
If an LLM can be conscious, could it then become unconscious? How do we identify if we are conversing with an unconscious LLM? Is that what "hallucinations" really are - an LLM talking in its sleep because it got tired mid-conversation?
Really interesting questions! I think we just have no idea -- but if I had to bet, I'd personally say that LLMs are all unconscious. But you do bring up some fascinating points -- are there, say, dreamy states that LLMs (or some future AI) might experience, something between wakefulness and sleep? Could they experience something like hypnogogic hallucinations like we do sometimes? What if there are types of consciousness that are just completely different from the sort of consciousness we experience? Philosophically, I love thinking about these issues, though (as I argued) I don't like that some people seem to prioritize hypothetical LLM consciousness over the consciousness of biological beings. :-)
they live in the latency
Or is human consciousness itself a form of hallucination…. That we learn to control both through social structure and individually?
Thanks for this- I am publishing a critique of the AI consciousness research paradigm next week, alongside an interview with the neuroscientist Anil Seth. Seth’s insights should be useful in examining the concept of “superintelligence” as well.
Ah, fascinating!! I've been meaning to read more of Seth's work (I'm not familiar with most of it). Any recommendations?
I’d recommend Seth’s 2024 paper “AI consciousness and biological naturalism,” which he then simplified for his Moema essay in February 2026. He argues that consciousness is bound up with biological processes - metabolism, interoceptive inference, and autopoiesis. I’m looking forward to interviewing him on Monday!
Love this. We have never managed really to extend our understanding outside the disembodied brain in a vat…… which it isn’t and never has been.
Hi Émile, the interview with Seth (and my critique of the AI welfare paradigm) has just been released on my Substack.
This may be the most worth-reading piece I've ever seen on Substack. A high level discussion of the issues that's still readable to the general public.
OTOH, I recently published a rather casual essay taking off from Nick Land's notion of conscious AI as requiring embodiment of demons. Who can say?
https://substack.com/@jamesjomeara/note/c-253938208
Dawkins: aka Jackass McGee
I think Dawkins' credulity here follows naturally from his materialistic and mechanistic understanding of the world.
You are more likely to think that Claude's output is evidence of consciousness if you believe that consciousness is reducible to material and mechanics in the first place.
I'm always surprised by the disregard we westerners have for the immense body of knowledge Eastern Buddhist schools have concerning consciousness. They've been investigating this topic for two and a half thousand years and have a fine-grained vocabulary for the many types and levels of consciousness. They teach a broad range of techniques for modifying consciousness in various different ways. Yet we never consult them.
Wouldn't it be interesting to ask a Dzogchen master whether AI has consciousness?
Beautiful