Particle.news
Download on the App Store

Lab-Grown Human Neurons Trained to Play Doom on Cortical Labs’ CL1

A new Python-accessible interface let an outside coder build the Doom bridge in about a week.

Overview

  • Inside the CL1 device, roughly 200,000 human neurons on a multi-electrode array were stimulated and read in real time to drive in-game actions.
  • Engineers translated Doom’s visuals into spatial patterns of electrical stimulation and decoded neural spikes into movement and shooting controls.
  • The neurons received feedback-based rewards for aiming and kills, while AI refined the stimulus encoding to guide learning.
  • Researchers report adaptive, goal-directed behavior but emphasize the dish plays at a beginner level and does not exhibit awareness or higher cognition.
  • The platform’s Python API enabled independent developer Sean Cole to implement the interface in days, contrasting with the earlier 18‑month Pong project.