Living human brain cells play DOOM on a CL1 [video]

Technical claims and skepticism

• Several commenters doubt the demo’s substance, comparing it to past overhyped “rat brain flies plane” work.
• Key critique: most learning may occur in the silicon encoder/decoder (CNN + PPO) rather than in the neurons; neurons could be a noisy channel, not the policy.
• Others point to the project’s README and ablation studies claiming that with a linear, zero-bias decoder and frozen encoder weights, learning still improves, implying neuron-level adaptation.
• Multiple people note that the neurons do not see the full framebuffer; they get a compressed signal (enemy position/distance) mapped to left/right/shoot actions, making the task close to Pong-level complexity.
• It’s repeatedly stressed that 200k randomly connected neurons on a chip are not equivalent to a structured animal brain, even though the neuron count is fruit-fly scale.

Nature and source of the neurons

• Neurons are lab-grown human cells; some are likely immortalized lines (e.g., tumor-derived), which reduces “personhood” intuitions for some.
• Questions raised about sourcing, cell maintenance, lifespan, infection risk, and why human cells are used instead of animal neurons.
• One thread suggests human cells are partly for publicity, though others argue human neurons are relevant for disease modeling.

Ethical concerns and analogies

• Many commenters express discomfort or horror: fears of creating “sentience in a box,” “torment nexuses,” and “I Have No Mouth and I Must Scream”-style scenarios.
• Others argue neuron count is far below plausible consciousness, and point out that factory farming and animal testing are much worse in terms of likely suffering.
• Debate over whether “sentience” or “consciousness” is even a coherent or measurable concept; some see concern as spiritual residue, others as precaution.
• Concern that scaling this to millions or billions of neurons, or integrating with drones/robots, could lead to slavery-like exploitation of conscious substrates.

Broader context and future directions

• Some see this as an early “wetware computing” step toward brain uploads, AGI via biological chips, or hybrid systems.
• Others say existing tech (LLMs, neural implants, connectome simulations) is still largely unrelated to actual consciousness uploading.
• Mixed reactions: some are excited by scientific potential (neurological disease research, new compute paradigms); others are repulsed by the perceived frivolity (making a nascent “brain” play Doom for a meme).