A University College London scientist says no reliable test exists for AI consciousness, yet major tech firms are pouring money into AI welfare as she warns of moral and policy risks either way.
Daily Mail reporting captured comments from Megan Peters, a lecturer in computational cognition at University College London, ahead of her talk at New Scientist Live on Oct. 10, 12.
Peters says the core problem is simple and blunt. There is still no agreed way to look for consciousness in an artificial system. She does not believe today’s AI models are conscious. She also says the question cannot be ignored.
“We need to have something better than vibes,” Peters said.
“You can't just go look for signs or indicators or markers of consciousness in an artificial system because we don't know what they are.”
That is the gap she wants filled before bold claims harden into policy, regulation, or corporate dogma.
Peters notes how easy it is for people to talk with these agents and walk away convinced someone is “in there.” She draws a hard line after that feeling.
“If we ask AI systems, we can't really trust the answer in the same way,” she said.
In other words, a fluent chatbot can sound self-aware. Sounding self-aware is not proof. Treating a polished reply as evidence is how “vibes” replace science.
Her team is building a comparative framework that would line up dozens, or even hundreds, of possible consciousness tests. The path she describes starts with humans and animals, then moves to artificial systems. Gradual cross-species work comes first. AI comes later.
She also tells audiences not to take headline claims at face value. “Just because you see someone claiming that they have or have not detected artificial consciousness, you need to go and ask why.”
While the tests remain unsettled, the money is not waiting. Peters said a huge amount of money is being spent on thinking about AI welfare inside some of the biggest AI companies, unnamed in the reporting, but described as industry leaders.
That spending sits beside a moral fork she lays out in plain terms.
“If AI is conscious and we treat it as not conscious, then there is a potential for morally significant harms,” Peters said. “Just going around switching things off that are sentient... if that is akin to killing them, then that doesn't seem great.”
She added the reverse error carries real weight too. “We want to not be torturing creatures that have conscious awareness.” And: “If we get it wrong the other way around, there's also very real consequences.”
Get it wrong one direction, and you risk treating a tool like a victim. Get it wrong the other way, and you risk writing rules, trust standards, and corporate priorities around machines that never felt a thing.
Peters’ talk at New Scientist Live is built on that double bind. No reliable test. Easy human projection onto chat agents. Heavy corporate attention to “welfare.” Moral language already in play, harm, torture, switching things off “akin to killing.”
She is not declaring today’s systems awake. She is saying the public and the industry lack a serious yardstick, and that guessing is not good enough when the stakes include policy and regulation.
For readers who still live in the real world of bills, borders, and human workers, the sequence is hard to miss. Firms race ahead with products that mimic conversation. Academics warn that consciousness has no agreed markers in silicon. Then the same industry starts budgeting for the welfare of systems its own critics say are probably not conscious at all.
Common sense still starts with people. Tests should come before rights talk. Receipts should come before vibes. And no company should get a free pass to launder product hype through moral panic about machines that may never have an inner life.
Until someone can show, not merely claim, that a model is aware of its own thoughts, the sober move is skepticism, not a new protected class for chatbots.
