AI’s Next Frontier? An Algorithm for Consciousness (www.wired.com)

A London startup called Conscium, founded in 2024 by Daniel Hulme with advisers including neuropsychologist Mark Solms and theoretical neuroscientist Karl Friston, is pursuing a bold, testable route to machine consciousness. Rather than equating conversational fluency with sentience, the team decomposes consciousness into measurable components—perception-action loops, metacognition and affect—and implements them using algorithms inspired by Friston’s “free energy” or active inference framework. Solms has reportedly run unpublished lab experiments in which simple simulated agents (“pleasure‑bots”) operate with emotion‑like internal signals (fear, excitement, pleasure) that drive hypothesis‑testing, prediction‑error minimization and exploratory behavior; the claim is that this feeling‑mediated feedback loop is what gives rise to subjective experience. If validated, the work reframes the AI debate: consciousness might be an emergent property of specific architectural motifs (affect-driven active inference plus metacognitive layers) rather than a byproduct of scale or language training alone. That implies new research directions—integrating affective, homeostatic signals and active sensing into agents, and potentially combining those mechanisms with LLMs so systems can report on internal states. But the project is nascent and controversial: experiments are unpublished, theoretical bases (e.g., Fristonian accounts) remain debated, and ethical and measurement challenges loom large. Still, Conscium’s approach offers a concrete, scientifically oriented way to probe long-standing philosophical claims about machine sentience.