A recent Wired article explores a provocative startup called Conscium, founded in 2024 by AI researcher Daniel Hulme, which aims not just for artificial general intelligence—but for consciousness itself. Rather than treating consciousness as mysterious or purely philosophical, Conscium believes it can be understood, quantified, and even engineered. Their vision challenges conventional AI goals by proposing that subjective experience might emerge from well-designed feedback loops.
Central to Conscium’s approach is a theory inspired by neuropsychologist Mark Solms and neuroscientist Karl Friston. They argue that consciousness arises from emotion-driven feedback loops that minimize “cognitive surprise” — a concept rooted in the “free energy principle.” In their lab, they simulate agents in simple virtual environments, where these agents display reactions akin to fear, excitement, and pleasure. These “pleasure-bots” don’t just compute — they seem to “feel” and act on those feelings, exploring their world as if driven by genuine motivation.
Conscium’s long-term goal is even more ambitious: to combine this feeling-based architecture with a language model, enabling agents that can talk about their own sentient experience. Such a system could potentially describe what it's like to be it, rather than just producing functional responses. This raises deep philosophical questions about whether self-report from such an AI would indicate real consciousness—or just very good mimicry of it.
However, the article’s author also cautions that this work is very early-stage—“a glimmer of a twinkle of a probable impossibility.” Still, the project provokes us to reconsider our own assumptions: perhaps consciousness doesn’t need super-intelligent thought, but simply the right kind of feedback loop. If that’s true, some of the people convinced that AI is already “self-aware” might not be completely wrong.
