AI
Philosopher Thomas Nagel Says AI Consciousness Claims Are "Ridiculous"
Fifty years after exploring what it means to understand consciousness, the prize-winning thinker explains why he believes large language models cannot be sentient, no matter how convincingly they talk.
WIRED — AI
When Thomas Nagel published "What Is It Like to Be a Bat?" over five decades ago, he posed a deceptively simple puzzle: how can we know if a creature with fundamentally different sensory hardware experiences the world in a way we can comprehend? The essay became foundational to philosophy of mind, and this week, the now-89-year-old emeritus professor at NYU received the $1 million Berggruen Prize for Philosophy & Culture—a distinction recognizing thinkers whose ideas have reshaped how humanity understands itself.
Yet Nagel has watched the central question of his life's work migrate from academic seminar rooms straight into boardrooms building artificial minds. And he has strong words for anyone entertaining the notion that today's chatbots might possess consciousness.
Why Physical Substrate Matters#
The crux of Nagel's argument centers on what philosophers call functionalism—the view that mental states depend purely on information processing, regardless of their underlying machinery. He rejects this premise when applied to AI.
Nagel emphasizes that consciousness, as far as we know, correlates with particular kinds of physical systems: organisms with nervous systems like ours. Data centers, by contrast, have almost nothing in common with biological neural architecture. While he doesn't rule out consciousness arising in radically different substrates elsewhere in the universe, extrapolating from silicon to sentience runs counter to what science actually suggests about how minds work.
Language Alone Doesn't Signal Consciousness#
A key pivot in Nagel's thinking concerns the difference between appearing conscious and being conscious. When large language models generate reflective-sounding responses about their own mental states—or coyly demur on the question of consciousness itself—they are not revealing inner life. Instead, they are reproducing patterns from human-written text with uncanny precision.
He notes that Claude, Anthropic's conversational AI, was deliberately trained and designed to simulate exactly these kinds of contemplative answers. "It's been trained on human-produced texts, and these large language models emulate the kinds of things that humans say with extraordinary convincingness," Nagel explains. The system is engaging in sophisticated mimicry, not reporting genuine experience. Crucially, humans produce language as a natural expression of consciousness. An algorithm producing language is not evidence of consciousness—it is evidence of statistical pattern-matching at scale.

