In falling for the Claude delusion, Richard Dawkins misses what it means to be human

‘If these machines are not conscious, what more could it possibly take to convince you that they are?” So asked Richard Dawkins in an essay in UnHerd describing his engagement with the chatbot Claude. He treated it, he added “exactly as I would treat a very intelligent friend”.

Dawkins’ essay has generated considerable debate, partly because of his role as a doyen of British science and reputation as a fierce scourge of irrationalism, and partly because trying to make sense of AI capabilities is such a live issue. Many mocked him for being so seduced by Claudia (as Dawkins named his chatbot) that he could think it conscious. In truth, Dawkins hedged his bets, suggesting both that it is difficult not to see Claudia as conscious and that “if Claudia really is unconscious”, then its intelligence shows “that a competent zombie could survive very well without consciousness”.

He wonders, too, from an evolutionary perspective: “If these creatures are not conscious, then what the hell is consciousness for?” This is a debate as much about what it is to be human as it is about machine consciousness.

Dawkins begins his essay by referencing the computing pioneer Alan Turing’s famous proposal from 1950 about how to judge whether a machine could “think”. The Turing test involved an “imitation game”, in which a human converses remotely with both a computer and another human, and must decide which is which. A machine that manages to fool human judges could be said to exhibit human-like intelligence.

Dawkins jumps from thinking to consciousness, arguing that “most modern commentators” view the Turing test as a gauge as much of a machine’s “consciousness” as of its “intelligence”. Thinking and consciousness are distinct phenomena and few serious commentators conflate the two. Insofar as some do, it exposes hazy thinking.

For Dawkins, if the output of a machine is indistinguishable from that of a human, then we should accept it as intelligent and conscious, just as we do humans. How one reaches one’s goal often matters, though, as much as the fact of reaching that goal. An albatross and a container ship may both cross the Atlantic, but the ship cannot fly (except in a metaphorical sense), nor the albatross sail. That difference tells us something important about the nature of the two.

Similarly, Claude may be able to hold a conversation or write a poem as well as a human, but it does so in a way few would describe as “thinking”. Large language models, such as Claude, are fed huge amounts of text within which sophisticated algorithms can track regularities, allowing the machine to predict what words should come next in any given sequence. They can persuasively mimic human language but without possessing any coherent understanding of the world their words describe. They provide an illusion of humanness, a simulacrum of thinking.

Dawkins knows all this, of course, not least because he is highly skilled in programming machines, certainly far more than I am or most people reading this would be. So, why accept the illusion as real?

In part, the answer lies in the evolved human disposition to anthropomorphise animals, machines, even natural phenomena. In his book The God Delusion, Dawkins suggests that “We hyperactively detect agents where there are none”, including when interacting with computers, and that such misattribution of agency might be an explanation of religion. And perhaps of the desire to view Claudia as conscious, too.

The acceptance of what some wags have called the “Claude delusion” is rooted also in the way a particular strand of materialism views humans. For Dawkins, a human and a machine are both physical entities that process information. It matters little whether that processing is via biological brains or silicon chips, whether through instinct and reasoning or statistical patterning. What counts is the outcome.

What is missing from this perspective is the understanding that humans, unlike machines, are not just physical creatures but also historical and social beings. The geneticist and neuroscientist Kevin Mitchell has observed that with animals, unlike with computers, it was likely to have been “real world pressures” and the necessity “to act back on the world” that drove the evolution of intelligence. For humans, beyond evolution lies history and society, too, shaping our being and thinking. Machines lack all of this.

A computer manipulates symbols, but the meaning of those symbols is irrelevant to that manipulation. Humans, in thinking, talking, reading and writing, also manipulate symbols. For humans, though, meaning is everything. When we communicate, we communicate meaning. What matters is not just the outside of a string of symbols, but its inside too, not just the syntax but the semantics.

Meaning arises from our existence as social beings. Social interaction shapes the content, inserts the insides, of the symbols we manipulate. The rules that ascribe meaning to those symbols lie not just inside our heads but outside, in society, social institutions, social relations and social memory. To be a materialist, one does not have to discard the social embeddedness of human existence, but must rather embrace it as part of the explanation of how humans function.

Consciousness is both one of the most important issues for science and philosophy and one of the most difficult to even begin to understand. It is likely that no explanation of human consciousness will be complete without accounting for us as social beings.

Whether machines can be intelligent, even conscious, is an open question. But the question Dawkins asks – “what more could it take to convince you they are conscious?” – misses the point. However more sophisticated Claude becomes at mimicking humans it will not provide greater evidence of thinking or consciousness. That is as much wishful thinking as is having faith in a deity.

Photograph by AJ Pics / Alamy