Overview
- Independent developer Sean Cole trained Cortical Labs’ CL‑1 to play the open-source Freedoom in about a week using the new Python interface.
- The setup converts on-screen game data into electrical stimuli for the neurons, whose firing patterns are read back as actions such as shooting or moving right.
- Performance topped a randomly firing baseline yet remained far below skilled human players, with the company claiming faster learning than conventional machine learning systems.
- Researchers say moving from Pong to Doom marks a notable capability jump and could inform work toward real-time tasks like robotic arm control, even as key mechanisms remain unclear.
- The rollout raised questions about the provenance of human cells and governance of networked access through the Cortical Cloud, with calls for clearer ethics and oversight.