What You Do Is Not What You Are

Scientists just decided that behavior is a bad way to judge whether a bee or a chatbot has a mind. Autistic people have been making that argument about ourselves for years.

For most of my life, the question of whether my mind worked "right" was settled by watching me. Did I hold eye contact. Did I answer the way the script wanted. Did I sit still, drop the stims, put the expected face on at the expected time. That's not an accident of how I was raised — it's the design. Autism is defined, in the diagnostic manual, by observable behavior. Not by what's happening on the inside. By what a stranger can clock from across a room.

This month, a group of consciousness researchers said, in so many words, that the outside is a terrible place to look.

The paper is "Identifying indicators of consciousness in AI systems," in Trends in Cognitive Sciences — Patrick Butlin, Robert Long, Tim Bayne, Yoshua Bengio, Jonathan Birch, David Chalmers, and a dozen others. Writing it up for The Conversation, Colin Klein and Andrew Barron put the takeaway plainly: a system can behave as if conscious without being conscious, and a creature can be conscious without performing it in any way we'd recognize. Their verdict on today's chatbots — ChatGPT included — is that they are not conscious. Not because they fail the conversation (they pass it easily now; five years ago, passing it was supposed to settle the question), but because of HOW they produce the words. The machinery, not the manner. As they put it: how it works is proving more informative than what it does.

Sit with that for a second, because it's the whole fight of my life stated as a research finding.

The old test for a mind was behavior. Could you hold a convincing conversation with the thing. The chatbots blew past that test, so the scientists threw the test out — correctly. Behavior is deceptive. A thing can perform a mind it doesn't have, and a thing can have a mind it doesn't perform. So they went looking at mechanism instead.

Now hold that next to how autistic people get assessed. The standard for "is there something wrong with this mind" has always been behavior, and nobody has thrown that out. ABA — the therapy still sold to parents as the gold standard — rewards the behavior and ignores the interior entirely. Make the eye contact, kill the stim, pass for typical, and you're scored as improved. It's a test of performance dressed up as a measure of the mind: the exact error the consciousness people just warned everyone away from, run on children, for decades, and called treatment.

The expansion is the part that should sting. Two years back, around forty scientists signed the New York Declaration on Animal Consciousness; more than 500 have signed since. It says there's a realistic possibility of conscious experience across the vertebrates and in plenty of invertebrates — octopuses, crabs, bees. The logic is what Jonathan Birch calls the precautionary principle for sentience: when there's a real chance someone's in there, you don't get to ignore it just because you can't prove it. You err toward believing.

Good. Extend it to the bee. Extend it to some machine we haven't built yet. I mean that — I'm vegan for a version of exactly this reason. But notice the shape of who gets the benefit of the doubt and who has to earn it. We're ready to grant a maybe-mind to an insect and a hypothetical AI on the strength of "we can't rule it out." The autistic person sitting in front of you, using words, telling you plainly that there is someone home — has a stronger claim than either, and still spends a lifetime being scored on whether the performance looks right.

So here's the lesson the researchers landed on, handed back to the people who needed it first: how it works matters more than what it does. Stop grading the mask. Ask how the mind actually runs.

Who gets counted as a mind, and who gets to decide — that was never a neutral measurement. It's a choice about whose inside we're willing to believe. The scientists just made the right call about bees and chatbots. The harder question is why it took longer than it took for the person who could simply tell you.