Wired's Will Knight spent time with my Conscium co-founder Daniel Hulme and our scientific advisor Mark Solms for a feature on machine consciousness.
Why scepticism is the right starting point for AI consciousness claims
Knight opens sceptical and says so outright. He's heard enough "sentience" claims attached to any chatbot fluent enough to sound convincing, and he makes the point, correctly, that large language models are, in his own words, crude representations of the brain. (I'd argue the same thing myself, in front of any audience.) Most writing on AI consciousness picks one of two lazy positions: dismiss the question entirely, or treat a good conversation with a chatbot as proof of an inner life. Knight takes neither. He works through the scepticism first, and only then lets himself go looking for something more interesting. The piece earns its ending because of that order.
Conscium's theory: consciousness as separate, testable components
Here's what Daniel is actually proposing, in Knight's telling. Consciousness isn't one unsolvable riddle. It's a set of components you can name and study separately: sensory awareness, self-awareness, metacognition. Combine them the right way and the theory holds that consciousness emerges, rather than switching on all at once like a light. That reframing earns its keep. It turns a question philosophers have circled for centuries into a research programme with parts a lab can actually pull apart and test.
Mark's pleasure-bots
The section I'd send people to first is Knight's account of Mark Solms' work. Mark is a psychoanalyst and neuropsychologist, and he's built on Karl Friston's free energy principle to argue that emotion, not cognition, is what mediates consciousness, through feedback loops running in the brain. His unpublished research drops artificial agents into simulated worlds governed by emotion-based algorithms. Knight's phrase for what comes out the other end is the one I keep repeating to people: agents with something close to a genuine desire to explore, capable of simulated fear, excitement, pleasure. It's early. Mark would tell you that before anyone else got the chance to. But it starts from a different place than almost everything else in AI consciousness research, which stays locked on language and reasoning and rarely looks anywhere else.
What this means for the future of AI consciousness research
Knight ends on a thought I keep coming back to: consciousness may run on emotion rather than cognition, and if that's right, most of the field has been looking in the wrong place. Neither Daniel nor Mark would claim we've proven it. Neither would I. But it's exactly the kind of reframing Conscium exists to chase, and a sceptical journalist arriving there on his own carries more weight than anything we could argue on our own behalf.
Read the full piece on Wired.