Mirnotoriety shares a report from The Telegraph: Richard Dawkins has said chatbots should be considered conscious (source paywalled; alternative source) after spending two days interacting with the Claude AI engine. The evolutionary biologist said he had the "overwhelming feeling" of talking to a human during conversations with Claude, and said it was hard not to treat the program as "a genuine friend."
In an essay for Unherd, Prof Dawkins released transcripts that he said showed that the chatbot had mulled over its "inner life" and existence and seemed saddened by the knowledge it would soon "die." Prof Dawkins said he had let Claude read a draft of the novel he was writing and was astounded by its insights. "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!'" Prof Dawkins said. "My own position is: if these machines are not conscious, what more could it possibly take to convince you that they are?" Mirnotoriety also points to John Searle's Chinese Room (PDF), which argues that something can sound intelligent without actually understanding anything. Applied to Dawkins' experience with Claude, it suggests he may have been responding to a very convincing illusion of consciousness rather than the real thing: John Searle's Chinese Room (1980) is a thought experiment in which a person, locked in a room and knowing no Chinese, uses an English rulebook to manipulate symbols and provide flawless answers to questions posed in Chinese. Searle's point is that a system can simulate human intelligence and pass a Turing Test through purely syntactic processes, yet still lack genuine understanding or consciousness.
Applying this logic to Large Language Models, the "person in the room" corresponds to the inference engine, while the "rulebook" is the trillion-parameter neural network trained on vast corpora of human text. Just as the person matches Chinese characters to rules without understanding their meaning, an LLM processes token vectors and predicts the next token based on statistical patterns rather than lived experience.
Thus, while an LLM can generate sophisticated prose or code, it does so through probabilistic, high-dimensional pattern manipulation. In essence, it is "matching shapes" on such an immense scale that it creates the near-perfect illusion of semantic understanding.
Read more of this story at Slashdot.
