Scientists propose new tests for AI consciousness, say ChatGPT not sentient

Consciousness debates often veer into philosophy seminar territory, but this work is notable because it tries to operationalize the question for both animals and AI. Instead of asking how human‑like an AI’s behavior appears, the authors focus on internal information‑processing structures—feedback loops, global workspaces, conflict resolution between goals, and so on. On those structural tests, today’s large language models, including ChatGPT, come up short. They can mimic conversation about qualia, but the way they process and route information is still too unlike brains to justify calling them conscious. For the race to AGI, this is a useful corrective to hype that equates fluent dialogue with sentience. It suggests we can keep pushing on capability and agency without immediately tumbling into machine moral patienthood, at least on current architectures. At the same time, the authors explicitly leave open that future systems with different designs could cross the line, and they tie this back to how we treat animals like bees and cephalopods. That framing nudges the field toward thinking about consciousness as a spectrum, with concrete engineering signatures, rather than an on/off switch that flips the moment a chatbot sounds reflective enough.