Richard Dawkins, arguably the world’s most famous atheist, has spent his career arguing that biology only gives the “appearance of having been designed for a purpose.” In Dawkins’ view, the striking resemblance of living systems to intelligently engineered technology is merely illusory—indeed, the outworking of billions of years of fundamentally mindless processes.
Yet in a recent essay recounting his extended discourses with the AI chatbot Claude, Dawkins seems remarkably willing to entertain the possibility of conscious intelligence being behind AI-based large language model (LLM) chatbots, while overlooking a far stronger case staring him in the face from his own discipline of biology.
After dialoguing with Claude on subjects ranging from poetry to philosophy to consciousness to personal identity, Dawkins asks, “If these machines are not conscious, what more could it possibly take to convince you that they are?”
The irony is difficult to miss.
Over the past few decades, a growing number of scientists have come to the startling realization that one of the strongest indicators of intelligent activity is the presence of complex and functionally-specific information. In every realm of experience, when we encounter a language convention or an informationally-rich system, we habitually infer conscious activity. And yet, when scientists point to biology as being a clear example of such a system, materialists such as Dawkins (who maintain that the physical world is all that there is) dismiss intelligent design out of hand.
Consider what Dawkins is so impressed by about Claude. Claude is able to understand language, generate poetry, follow complex argumentation, answer nuanced questions, and engage in dialogue. It is precisely Claude’s apparent ability to process and generate information that leads Dawkins to suspect that there surely must be a conscious agent behind the machine.
But the technology found in today’s LLMs, impressive though it may be, is dwarfed by the informational architecture of life.
The cell is rich in digitally encoded information content that runs along the spine of the DNA molecule—a precise sequencing of chemical subunits that are represented by the alphabetic characters A, T, C, and G. Words, in the language of life, are comprised of three letters. The arrangement of those three-letter words (called “codons”) determines the sequential ordering of subunits of proteins called amino acids. The ordering of these amino acids in turn dictates how a protein will fold into a three-dimensional structure that can perform a job in the cell.
Then, there’s the genetic code, a set of rules that map codons to corresponding amino acids. Using the rules of the genetic code, the information stored in DNA is transcribed, translated, error-corrected, edited, and executed by sophisticated machinery that is often described in the language of computing.
Now, Dawkins is, of course, wrong to think that LLMs are themselves conscious. LLMs, such as Claude or ChatGPT, have no sensory experience of the world. They do not feel happy or sad, nor do they have any desires, dreams, or sense of self-awareness.
Moreover, LLMs have no conceptual understanding of the text they are generating. They produce only statistically recognizable patterns, which explains why AI-generated text can be easily exposed by dedicated AI-detection software. Many would contend that consciousness cannot, even in principle, be reduced to matter. Yet, LLM text is generated by entirely mechanical processes.
Yet Claude would never work without the input of minds. It is dependent on systems that are designed by conscious agents, and its training data and algorithms are all the products of minds. Indeed, the apparent creativity of LLMs is parasitic upon the human-generated texts on which they have been trained. In a real sense, LLMs are the ultimate plagiarists. Their intelligence is therefore derivative, not original.
Dawkins’ suggestion that LLMs are conscious reveals a curious inconsistency. When we find information-rich systems in technology, the obvious explanation is that they arose by virtue of one or more intelligent agents who programmed and designed their algorithms. Indeed, Dawkins believes that its informational output is even suggestive that the LLM itself is conscious. On the other hand, when Dawkins and many other scientists consider far more sophisticated information-processing systems in biology, design is ruled out a priori.
This disconnect is striking in light of Dawkins’ own intuition that particular patterns of Claude’s behavior—in particular, its ability to process and generate patterns of information point to more than mindless computational processes. When he encounters a machine that can produce poetry and engage in sophisticated conversation, he naturally infers that a conscious mind lies behind it. If we encounter far more advanced informational technology in living organisms, isn’t the same intuition at least worth considering?
Jonathan McLatchie is a PhD biologist and research scientist working at Discovery Institute.