AGI consciousness test gaps could fuel Big Tech hype, warns Cybernews report

This piece surfaces a tension that often gets lost in capability benchmarks: even if labs hit “AGI-level performance” on tasks, we may still have no principled way to say whether a system is conscious, sentient, or anything like a moral patient. McClelland’s argument, as summarized here, is that our current theories of consciousness are too underdetermined to ground a clean test, and that this vacuum becomes fertile ground for marketing spin. ([cybernews.com](https://cybernews.com/ai-news/general-artificial-intelligence/))

For the race to AGI, that ambiguity is dangerous in two directions. On one side, companies can sell anthropomorphized narratives about conscious assistants to differentiate their models, even when underlying capabilities haven’t qualitatively changed. On the other, genuine warning signs of systems with rich internal states could be dismissed as “just stochastic parrots” because we lack shared criteria. Either way, the article is a reminder that governance and safety debates can’t only track loss curves and benchmark scores; they need a parallel track on how we talk about and regulate claims of sentience and agency. If we don’t, the first “AGI moment” may be defined as much by narrative control as by any objective technical milestone.