Summary
Can artificial intelligence ever be conscious? Physicist and science commentator Sabine Hossenfelder examines a new paper from two biologists that argues today’s AI, including large language models, is unlikely to replicate the kind of processing behind consciousness in biological brains.
The video lays out the two sides of the debate: computational functionalism, which holds that consciousness comes from how information is processed regardless of the physical substrate, and biological naturalism, which says the specific biological processes of the brain are required. The paper’s authors propose a middle ground called biological computationalism and identify three key differences between brains and current AI. The brain operates across multiple scales, from molecules and cells to large-scale oscillations. Its processes span many orders of magnitude in time, blending digital and analog behavior. And it works under strict metabolic limits that force energy-efficient computation.
Hossenfelder, who personally favors computational functionalism, explains why she finds the paper a useful contribution anyway. The authors explicitly do not claim that only biological tissue can be conscious, and the video considers whether the findings mean AI is unlikely to become conscious at all, or only unlikely to do so in ways humans would recognize. It also touches on Anthropic’s reported practice of keeping backups of earlier Claude versions and on why this research direction may be economically inconvenient.
📺 Source: Sabine Hossenfelder · Published October 10, 2026
🏷️ Format: Opinion Editorial
