Limitless: An AI Podcast Human Brain Cells in a Petri Dish Just Played DOOM (This Is Real)
5 snips
Mar 10, 2026 They cover human nerve cells trained to play Doom and how researchers wired tens of thousands of neurons to a microarray. They discuss a full fruit fly brain recreated in silico and how copied wiring produced natural behaviors. They debate energy-efficient biological computing, scaling implications for AI infrastructure, and the ethical questions around simulated minds.
AI Snips
Chapters
Transcript
Episode notes
Human Brain Cells Played Doom On A Chip
- Human-derived neural cultures learned to play Doom using 200,000 neurons cultured on a microarray chip and wired into the game over a week.
- The system ran on a single ~$35,000 machine, used minimal energy, and leveraged biological learning rather than gradient-descent training.
Biology Beats Silicon On Energy Efficiency
- Biological computing may be orders of magnitude more energy-efficient than silicon, with small neural setups running on ~20 watts versus huge GPU farms.
- Josh Kale and David Hoffmann argue evolution has already produced optimized learning machinery that could beat brute-force silicon training for some tasks.
Don't Bet On Immediate Biological Disruption
- Investors and engineers should not assume biological computing will immediately disrupt GPU-driven AI; markets currently price low probability for a near-term 'bubble pop'.
- David Hoffmann cites Polymarket probabilities showing little market concern through 2026 despite recent developments.