The Machine That Wasn't Ready for Its Close-Up

Dennett spent the late 1990s telling everyone to take the Turing test more seriously, and everyone spent the next quarter-century doing the opposite. That is the central, almost comic irony of rereading *Brainchildren* in 2026. When Dennett argued that Turing's original imitation game was far more rigorous than the parlor-trick versions proliferating in AI demos and philosophy seminars, he was issuing a warning dressed as a clarification. The warning was this: we would begin mistaking fluency for thought, and we would do it long before machines gave us any reason to. He was correct on the timeline and, arguably, on the mechanism. Large language models now pass debased versions of the Turing test every day — customer service bots, therapeutic chatbots, AI companions that users describe as "understanding" them. Dennett saw that the failure would not be in the machines but in the humans administering the exam, lowering the bar until passing it meant nothing. What he could not have anticipated was the speed and scale. He imagined the confusion would be philosophical. It turned out to be commercial.

The book's blind spots are the blind spots of its decade. Dennett writes as though the primary danger of AI confusion is epistemic — that we might *believe wrong things* about machine minds. He does not reckon with the possibility that millions of people might not care whether the machine thinks, so long as it performs intimacy convincingly enough. The social dynamics of 2024-2026 — grief bots, parasocial relationships with AI characters, the quiet erosion of the distinction between simulation and presence — are not on his radar. This is not a failure of imagination so much as a failure of sociology. Dennett was a philosopher of mind, not of loneliness. He assumed the question "Can machines think?" would remain the interesting one. The question that actually arrived was "Does it matter if they can't?"

What hits differently now is Dennett's insistence that the Turing test is a "conversation-stopper" — a challenge so demanding that meeting it would settle the debate. He meant this as a compliment to Turing's design. But in hindsight, the phrase has a darker resonance. Conversations about machine cognition have indeed been stopped, not by the rigor of the test but by the sheer ubiquity of systems that sidestep it entirely. No one is running five-hour interrogations of ChatGPT to determine if it possesses genuine understanding. People are just using it, and the philosophical question Dennett wanted to sharpen has been blunted by practice. The test was designed for a world where encountering a thinking machine would be an event. We live in a world where encountering a language-producing machine is a Tuesday.

Within the broader corpus of philosophy of mind and AI, *Brainchildren* sits at a hinge point. It inherits from Turing, obviously, but also from Dennett's own prior work — *Consciousness Explained*, the intentional stance, the rejection of Cartesian dualism. It gave its successors a vocabulary for skepticism about AI hype that remains useful, even if the hype has outrun the skepticism. You can draw a line from Dennett's arguments here through the "stochastic parrots" debate of the early 2020s: the insistence that producing language is not the same as understanding it. Dennett would have recognized that argument. He might have been disappointed that it needed to be made again. The book also prefigures, without resolving, the alignment problem — the question of what happens when we build systems whose behavior we can describe but whose "reasoning" we cannot access. Dennett was interested in the philosophy of that opacity. The engineers of 2026 are interested in its consequences.

One question remains, and it is not the one Dennett was asking. He wanted to know whether machines could think. The question *Brainchildren* now raises, without meaning to, is simpler and worse: if we can no longer agree on what counts as evidence of thinking — in machines or in each other — what exactly was the test ever protecting?