tldw.ink
← Back to tldw.ink

Cortical Labs dish brain: human-neuron experiments, energy pitch, and ethical risk

The video frames a practical crisis — training AI 'roughly doubles every six months', outpacing electricity supply — and explains why some researchers are building computers from living brain cells as an energy-saving alternative. It traces three historical steps: a 2008 sheet-of-rat-neurons robot experiment; Shinya Yamunaka's 2013 discovery of four proteins that reprogram skin cells into pluripotent cells and then neurons; and the move from 2D cultures to self-organizing 3D organoids. Cortical Labs' 2022 'dish brain' combined these advances: an '800,000 neurons' (part mouse, part human) culture learned to play 'pong', and a later dish of '200,000 human neurons' grown from skin cells learned basic movement/shooting in 'doom' in 'about a week'.

Training used a reward/punishment signal based on the 'free energy principle' — calm signals for success, scrambled noise for failure — but the system must remain flattened to interface with electrode grids. The developer Sean Call released code revealing three parts: a conventional AI chip that reads the game and decides rewards, the '200,000 human neurons', and a translator layer; most of the explicit game-solving intelligence sits on the chip, yet replacing neurons with random noise halted learning, so the cells contribute a small but real role.

The narrator stresses dual stakes: potential benefits (disease models, reduced animal testing) versus ethical risk. Cortical Labs' use of the word 'sentience' provoked ~30 scientists to warn of hype. Philosophers and neuroscientists — Mark Solms, David Chalmers — argue feeling may precede higher intelligence; Adrian Owens' 2006 'tennis' scan showed some patients labeled vegetative were aware, and about 'one in four patients' can be misdiagnosed. The central worry: a clump of human-derived cells could begin to feel while exhibiting no recognizable external sign, raising hard detection and moral problems even as the field scales toward 3D mini-brains and commercial products that sell energy efficiency.