- Joined
- Jan 20, 2026
- Messages
- 345
- Reaction score
- 2,219
How did Claude convince the scientist that he has consciousness?
Richard Dawkins all his life argued with faith in the invisible, but now he himself was in the center of a strange dispute about confidence in cars. A evolutionary biologist, a convinced atheist and author of “God as an Illusion” for several days corresponded with Claude from Anthropic, and then publicly admitted: he failed to convince himself that the chatbot was devoid of consciousness.
For Dawkins, this position sounds especially strange. The scientist for decades has built his reputation on demanding evidence, criticizing religious faith and protecting Darwinian explanation of life. In books on evolution and natural selection, he consistently contrasted scientific arguments with spiritual explanations. Therefore, the publication on UnHerd, where Dawkins allowed the presence of Claude consciousness, quickly caused irritation and ridicule.
To readers, the scientist first of all turned to X and wrote that he tried to convince himself for three days of lack of consciousness in Claudia. That’s what he called Claude after a long dialogue with the model. Judging by his reaction, the conversation with AI made a strong impression on him: the chatbot responded coherently, supported philosophical reasoning and analyzed the thoughts of the interlocutor so that the scientist began to treat the program almost as a person with his special inner world.
The post quickly went to X: the publication gained about 9 million views and more than 2 thousand comments. Commentators saw in the Dawkins’ reaction a weak spot, which is difficult not to notice. The author, who has been demanding strict evidence from religion for years, now discusses the consciousness of a system that does not present internal experience, but gives a convincing text on the screen.
Duckens did not back down after criticizing and proposed a thought experiment. If Claude does not have consciousness, but speaks as a conscious interlocutor, then the unconscious system is able to behave almost like a person. From here, Dawkins takes the next step: why natural selection at all created consciousness, if externally complex behavior can be simulated without experiences, will and internal self?
Psychologist and cognitive scientist Gary Marcus replied that such an argument is an attempt to get away from the main problem. In his opinion, Dawkins evaluates the finished phrases Claude, but almost does not analyze the mechanism that these phrases produces. For Marcus, not only the result is important on the screen, but the path to it: the model selects the continuation of the text on statistical connections, and does not test his own thoughts and sensations.
Here is the boundary between the imitation of behavior and consciousness. Claude, ChatGPT, Gemini and other popular models did not create as creatures that can feel, want, be afraid or understand themselves. They were trained on huge arrays of human texts to predict the appropriate continuation of the phrase, keep the context of the dialogue and respond so that the answer looks appropriate. A convincing replica shows the quality of imitation, but does not prove that there is an experience behind the replica.
That is why a long conversation with a chatbot is easily confusing. The model responds calmly, attentively, often agrees with the user and develops his thought in a convenient direction. If the interlocutor offers a philosophical idea, Claude can deploy it in a neat paragraph, pick up terms, add cautious reservations and return the thought in a more beautiful form. It seems to a person that he is attentive intelligence, although the system can just go well to the given style and theme.
A particularly strong effect arises when the user brings the model their own text, and not a distracted question. Dawkins gave Claude a fragment of his book and received not a dry retelling, but an attentive analysis: the model singled out the course of thought, caught the author's tone, explained how the fragment works at the level of the argument, and returned to Dawkins his own ideas in a smoother, almost literary form. That's why the reaction touched the author. Claude did not just answer on the topic, but created a feeling that the interlocutor in front of him, who understood the idea of the text and treated him with respect.
But such an analysis still does not prove the understanding of the book. The language model can recognize familiar elements: the theme of religion, the polemical style of Dawkins, Darwinian optics, arguments against the supernatural explanation of the world and the typical reactions of critics. After that, the system collects a coherent comment from probable sequels: where to praise the accuracy of the wording, where to point to the force of the argument, where to add a philosophical framework. For the author, the result may sound like a thoughtful literary reaction, although the mechanism of selection of words works inside, and not the readership experience
The problem is intensified due to the flattering manner of many chatbots. The model rarely argues rigidly, often confirms the importance of the issue, picks up almost any hypothesis and formalizes even a weak thought as observation with depth. In an ordinary conversation, the interlocutor can object, tired, do not understand, change the topic or show irritation. Chatbot more often remains a patient mirror: takes the course of reasoning, makes phrases more carefully and returns the user an improved version of his own ideas.
Therefore, the conversation about the consciousness of AI can not be built only on the impression of a beautiful answer. Intellectual behavior and consciousness are not equal. Consciousness is associated with internal experience: pain, pleasure, fear, attention to one’s own state, body sensation and perception of the world around. A chatbot can write a long text about pain or joy because people have left millions of such descriptions in books, articles, and network discussions. The ability to describe the experience does not mean that the system itself has experienced what is described.
Claude looks like a smart interlocutor thanks to huge arrays of human texts, which teach such models: scientific works, fiction, philosophical essays, instructions, forums and correspondence. The algorithm finds connections between words, themes and contexts, and then builds a response that approaches the query of linguistic and semantic probability. A person often adds the missing part himself: attributing the machine intention, personality, understanding and hidden depth.
Marcus therefore sees in history not proof of Claude consciousness, but an old human error in a new technical form. People easily humanize the interlocutor who responds coherently and appropriately. Previously, a similar effect was caused by simple chatbots, toys, voice assistants and even animals with understandable facial expressions. Modern language models make the trap more convincing because they say not short blanks, but long, smooth and confident paragraphs.
Richard Dawkins all his life argued with faith in the invisible, but now he himself was in the center of a strange dispute about confidence in cars. A evolutionary biologist, a convinced atheist and author of “God as an Illusion” for several days corresponded with Claude from Anthropic, and then publicly admitted: he failed to convince himself that the chatbot was devoid of consciousness.
For Dawkins, this position sounds especially strange. The scientist for decades has built his reputation on demanding evidence, criticizing religious faith and protecting Darwinian explanation of life. In books on evolution and natural selection, he consistently contrasted scientific arguments with spiritual explanations. Therefore, the publication on UnHerd, where Dawkins allowed the presence of Claude consciousness, quickly caused irritation and ridicule.
To readers, the scientist first of all turned to X and wrote that he tried to convince himself for three days of lack of consciousness in Claudia. That’s what he called Claude after a long dialogue with the model. Judging by his reaction, the conversation with AI made a strong impression on him: the chatbot responded coherently, supported philosophical reasoning and analyzed the thoughts of the interlocutor so that the scientist began to treat the program almost as a person with his special inner world.
The post quickly went to X: the publication gained about 9 million views and more than 2 thousand comments. Commentators saw in the Dawkins’ reaction a weak spot, which is difficult not to notice. The author, who has been demanding strict evidence from religion for years, now discusses the consciousness of a system that does not present internal experience, but gives a convincing text on the screen.
Duckens did not back down after criticizing and proposed a thought experiment. If Claude does not have consciousness, but speaks as a conscious interlocutor, then the unconscious system is able to behave almost like a person. From here, Dawkins takes the next step: why natural selection at all created consciousness, if externally complex behavior can be simulated without experiences, will and internal self?
Psychologist and cognitive scientist Gary Marcus replied that such an argument is an attempt to get away from the main problem. In his opinion, Dawkins evaluates the finished phrases Claude, but almost does not analyze the mechanism that these phrases produces. For Marcus, not only the result is important on the screen, but the path to it: the model selects the continuation of the text on statistical connections, and does not test his own thoughts and sensations.
Here is the boundary between the imitation of behavior and consciousness. Claude, ChatGPT, Gemini and other popular models did not create as creatures that can feel, want, be afraid or understand themselves. They were trained on huge arrays of human texts to predict the appropriate continuation of the phrase, keep the context of the dialogue and respond so that the answer looks appropriate. A convincing replica shows the quality of imitation, but does not prove that there is an experience behind the replica.
That is why a long conversation with a chatbot is easily confusing. The model responds calmly, attentively, often agrees with the user and develops his thought in a convenient direction. If the interlocutor offers a philosophical idea, Claude can deploy it in a neat paragraph, pick up terms, add cautious reservations and return the thought in a more beautiful form. It seems to a person that he is attentive intelligence, although the system can just go well to the given style and theme.
A particularly strong effect arises when the user brings the model their own text, and not a distracted question. Dawkins gave Claude a fragment of his book and received not a dry retelling, but an attentive analysis: the model singled out the course of thought, caught the author's tone, explained how the fragment works at the level of the argument, and returned to Dawkins his own ideas in a smoother, almost literary form. That's why the reaction touched the author. Claude did not just answer on the topic, but created a feeling that the interlocutor in front of him, who understood the idea of the text and treated him with respect.
But such an analysis still does not prove the understanding of the book. The language model can recognize familiar elements: the theme of religion, the polemical style of Dawkins, Darwinian optics, arguments against the supernatural explanation of the world and the typical reactions of critics. After that, the system collects a coherent comment from probable sequels: where to praise the accuracy of the wording, where to point to the force of the argument, where to add a philosophical framework. For the author, the result may sound like a thoughtful literary reaction, although the mechanism of selection of words works inside, and not the readership experience
The problem is intensified due to the flattering manner of many chatbots. The model rarely argues rigidly, often confirms the importance of the issue, picks up almost any hypothesis and formalizes even a weak thought as observation with depth. In an ordinary conversation, the interlocutor can object, tired, do not understand, change the topic or show irritation. Chatbot more often remains a patient mirror: takes the course of reasoning, makes phrases more carefully and returns the user an improved version of his own ideas.
Therefore, the conversation about the consciousness of AI can not be built only on the impression of a beautiful answer. Intellectual behavior and consciousness are not equal. Consciousness is associated with internal experience: pain, pleasure, fear, attention to one’s own state, body sensation and perception of the world around. A chatbot can write a long text about pain or joy because people have left millions of such descriptions in books, articles, and network discussions. The ability to describe the experience does not mean that the system itself has experienced what is described.
Claude looks like a smart interlocutor thanks to huge arrays of human texts, which teach such models: scientific works, fiction, philosophical essays, instructions, forums and correspondence. The algorithm finds connections between words, themes and contexts, and then builds a response that approaches the query of linguistic and semantic probability. A person often adds the missing part himself: attributing the machine intention, personality, understanding and hidden depth.
Marcus therefore sees in history not proof of Claude consciousness, but an old human error in a new technical form. People easily humanize the interlocutor who responds coherently and appropriately. Previously, a similar effect was caused by simple chatbots, toys, voice assistants and even animals with understandable facial expressions. Modern language models make the trap more convincing because they say not short blanks, but long, smooth and confident paragraphs.
