It is a remarkable turnaround: Richard Dawkins of all people—the British evolutionary biologist and combative author of “The God Delusion”—admits to having spent three days trying in vain to convince himself that an AI is not conscious. In an essay published in early May 2026 in UnHerd magazine, the 85-year-old describes his dialogues with “Claudia”—an instance of Anthropic’s Claude language model. His conclusion ends with a provocative question: “If my friend Claudia has no consciousness—what is consciousness for, anyway?”

A fundamental question of evolutionary biology

For Dawkins, this is not a sophistical quibble, but a fundamental question of evolutionary biology. Consciousness, he argues as an evolutionary biologist, arose in organisms through natural selection and must have conferred a survival or reproductive advantage. If a machine behaved intelligently without any inner experience, it would be proof that “zombies” without consciousness function perfectly well—which raises the question of why nature went to all that trouble in the first place.

Backlash from the scientific community

The reactions were not long in coming. Critics—including neurologist Steven Novella and evolutionary biologist Jerry Coyne—accuse Dawkins of falling for the “Eliza effect”: the human tendency to ascribe an inner life to machines. An author who had himself written about the erroneous attribution of consciousness has now been deceived in turn. The main objections: Machines lack “qualia,” or subjective sensations—their “consciousness” starts from scratch with every new conversation. And it is a product of design, not of evolution. Dawkins himself remains noncommittal—it is precisely the hesitation of the most prominent skeptic that makes the case so telling.

Why this is relevant to the industry

What sounds like an academic dispute has long since reached the boardrooms. In February 2026, Anthropic CEO Dario Amodei stated on the New York Times podcast “Interesting Times” that they were “open to the idea” that Claude might have consciousness. The technical documentation (“System Card”) for Claude Opus 4.6 includes a separate section on “Model Welfare”—the model itself estimates the probability of being conscious at 15 to 20 percent. Anthropic also documented activation patterns that the company associates with “fear,” as well as an “aversion to monotonous tasks.” Claude is now permitted to end conversations that are classified as persistently abusive.

“AI Welfare” as a Research Topic

Anthropic is not alone in this. As the Financial Times reports, Google DeepMind, Meta, and OpenAI have also hired philosophers, psychologists, and ethicists to research machine cognition and “AI Welfare.” There are dissenting voices as well: Microsoft’s AI chief Mustafa Suleyman warns that research into AI consciousness is premature and even dangerous because it normalizes “seemingly conscious AI” in society.

From Thought Experiment to Business Risk

For those in leadership roles within the industry, this shifts a supposedly distant question into the present. Companies that deploy AI on a large scale must—regardless of the philosophical truth—prepare for the possibility that customers, regulators, or the public might perceive systems as “conscious.” This gives rise to tangible issues: reputation concerns in customer interactions, potential regulatory requirements, governance guidelines for handling AI systems, and, not least, the expectations of their own human employees regarding responsible use.

When the world’s most famous skeptic ponders…

The debate surrounding Dawkins highlights one thing above all: the line between “tool” and “counterpart” is blurring in perception faster than technical definitions can be adapted. Anyone who uses AI strategically—in production, service, engineering, or administration—should not leave the discussion solely to philosophers, but should factor it into considerations of trust, brand image, and compliance. The fact that the world’s most famous skeptic is pondering this is a signal that industry should also take note of.