Nature Podcast Briefing chat: ‘Can it run Doom?’ — why scientists got brain cells and a satellite to play the classic game
15 snips
Mar 13, 2026 Rachel Fieldhouse, science journalist who explains quirky research stories. She talks about why researchers reuse Doom across AI and biology. They cover neurons on a chip playing Doom and how an AI translates between game and cells. The conversation includes running Doom on odd devices, a satellite test, and why playful experiments matter for science.
AI Snips
Chapters
Transcript
Episode notes
Human Neurons Learned To Play Doom
- Cortical Labs taught ~200,000 human neurons on a chip to play Doom via an AI intermediary.
- The AI converted screen events into electrical inputs and translated neuronal output into game actions, reaching average-player performance in about a week.
Pong Was The Training Ground For Doom
- Cortical Labs previously taught their neurons to play Pong, which took about 18 months to achieve.
- After that milestone they increased complexity to Doom and succeeded much faster, achieving playable performance in roughly a week.
Brain Cells Offer Energy Efficient AI Potential
- Biological neural networks could offer energy-efficient alternatives to conventional AI for adaptive tasks.
- Cortical Labs highlights lower energy use and potential for drug-testing models that behave more like human brains than silicon-only systems.