In the same week that The Atlantic asks what if Freud was right, a neuroscientist arguing for psychoanalysis as a genuine cure for mental illness, an arXiv paper arrives proposing that the oldest thought experiments in philosophy of mind, Leibniz's mill, the Turing test, Searle's Chinese Room, should be repurposed as safety tools for AI. The coincidence is either remarkable or inevitable. When the tools for measuring machine consciousness are the same tools we use to measure human consciousness, the boundary between the two questions has effectively collapsed.
The Turing Test Was Always a Mirror
Peter David Fagan's 2026 arXiv paper revisiting these classic thought experiments for AI safety makes the argument that existing frameworks measure behavior and capability, not inner states. This is precisely what the Atlantic's Freud rehabilitation is also about: the claim that neuroscience, for all its imaging and pharmacology, still has trouble accounting for the felt interior of experience. Freud's talking cure worked not by mapping neural correlates but by taking the reported inner life seriously as data. If that method is now being reconsidered for human minds, it raises an uncomfortable question about AI: should we be doing the equivalent? Interviewing the model about its inner life rather than benchmarking its outputs?
What the Safety Researchers and the Analysts Have in Common
The overlap here is not trivial. ChatGPT dominating Congress's AI spending means that the most powerful institutions on earth are deploying systems whose inner workings remain, by Fagan's account, unevaluated by any serious consciousness framework. Congress is not asking Searle's question. It is just paying the invoice. The Atlantic piece notes that the resistance to Freud in mainstream neuroscience has always been partly territorial, a discipline protecting its methods. The resistance to consciousness-based AI safety evaluation feels structurally identical. Culture Slop's conversation with Eugenia Kuyda on raising AI like a garden, shaping software instead of being shaped by it, offers a practitioner's version of the same intuition: the interior of the system matters, even if we can't measure it cleanly.