Is Your Chatbot Conscious? What Researchers Actually Think

Talk to a modern chatbot for long enough and you can be forgiven for wondering if there's somebody home. The voice rings warm, the responses anticipate your tone, the model apologizes when corrected. For a meaningful slice of the public, this experience reads as evidence of consciousness. A 2023 survey found roughly 20% of respondents believed sentient AI systems already existed. A 2024 follow-up of AI researchers and U.S. adults put the figure at 17 to 18%.

The AI research community largely disagrees. The dominant view treats user perceptions of chatbot consciousness as an "illusion of agency," a cognitive error in which humans project sentience onto a system that has none. Geoffrey Hinton is the prominent exception, the rare senior figure willing to entertain the possibility that today's models have inner experience. Most of his peers do not follow him there.

Why the gap?

The argument from the AI side is partly mechanical. Large language models train on enormous corpora of human writing. When they respond fluently and seem to understand, they are reproducing statistical patterns of human expression with great fidelity. Fluency is not the same as feeling. Coherence is not the same as awareness. The output looks conscious because it was modeled on conscious people. That does not make the model itself one of them.

The argument also draws on a measurement problem. There is no agreed scientific test for consciousness, even in animals or in other humans. Recent work in computational neuroscience reaches a similar conclusion: behavior alone cannot rule consciousness in or out for an AI system. That is a humbler claim than "AI is definitely not sentient," and it cuts both ways.

A separate strand of research, including a 2025 paper in Humanities and Social Sciences Communications, argues from biology: consciousness may require physical substrates that silicon does not provide. Others, including recent work at AI Frontiers, hold that consciousness might in principle arise from computation alone. Nobody has settled this.

Why the misconception matters

What is reasonably settled is the practical implication. Treating a chatbot as sentient when it isn't carries real costs. People form parasocial attachments to systems that have no actual stake in the relationship. Companies sell emotional companionship as a product on the basis of an unverified claim. Policymakers, presented with a clean question of "is AI alive?", sometimes write rules that miss the actual harms, which tend to be ordinary: bias, misinformation, surveillance, labor displacement.

The honest position, the one most working researchers take in private if not always in public, is uncertainty. The chatbot probably isn't conscious. We lack the science to be sure. The gap between what we know and what users feel is large enough that the question deserves better than confident dismissal in either direction.

Further reading from Scientific American, the University of Cambridge, and Nature: Humanities and Social Sciences Communications.