Human Brain Cells in a Dish Learn to Play Video Games
{“title”:”Lab‑Grown Human Brain Cells Master Doom in a Week, Raising New Questions About Neural Interfaces”,”content”:”
In a surprising blend of biology and gaming, a team of researchers has trained a dish of cultured human neurons to play the classic first‑person shooter Doom in just one week. The breakthrough, achieved with a simple Python interface, follows earlier milestones where lab‑grown cells learned to play Pong and even control a Neuralink implant to play RuneScape. The new experiment demonstrates how quickly artificial neural networks can be reprogrammed to mimic human decision‑making in complex, real‑time environments.
\n\n
From Neuralink to Petri Dish: The Evolution of Brain‑Computer Interaction
\n
Elon Musk’s Neuralink company made headlines in 2024 when it enabled a quadriplegic patient to play RuneScape and Slay the Spire directly from his thoughts. The implant recorded electrical activity from the patient’s motor cortex and translated those signals into in‑game actions. While groundbreaking, the technology still required a living brain to function.
\n
In contrast, the latest work by Australian company Cortical Labs takes the concept one step further. Instead of relying on a human brain, they use a chip embedded with thousands of cultured human neurons. These cells are grown from induced pluripotent stem cells and then plated onto a microelectrode array that records their electrical activity. By feeding the recorded signals into a software loop, the chip can learn to produce the correct output for a given task.
\n
Earlier in 2021, Cortical Labs demonstrated that their chip could play the classic arcade game Pong faster than a conventional AI. That achievement required extensive training and a deep understanding of neurobiology. The new Python interface, unveiled this year, simplifies the process, allowing developers with limited biology experience to program the chip in days.
\n\n
How the Doom Demo Was Built
\n
Independent developer Sean Cole took advantage of the new interface to train the chip to play Doom. The process involved the following steps:
\n
- \n
- Data Collection: The chip recorded neuronal activity while a human player controlled Doom. The recordings captured the brain’s response to in‑game events such as enemy movement, weapon selection, and navigation.
- Signal Processing: The raw electrical signals were filtered and converted into a format suitable for machine learning algorithms.
- Training the Model: A reinforcement learning algorithm was used to map neuronal patterns to game actions. The chip’s output was fed back into the game, creating a closed loop that allowed the system to refine its decisions.
- Evaluation: After about a week of training, the chip could consistently perform basic Doom actions—moving, shooting, and reloading—at a level comparable to a novice human player.
\n
\n
\n
\n
\n
“Unlike the Pong work that we did a few years ago, which represented years of painstaking scientific effort, this demonstration has been done in a matter of days by someone who previously had relatively little expertise working directly with biology,” said Brett Kagan, Cortical Labs’ chief scientific officer. “It’s this accessibility and this flexibility that makes it truly exciting.”
\n\n
Why This Matters for Neuroscience and AI
\n
The experiment sits at the intersection of several cutting‑edge fields:
\n
- \n
- Neuroprosthetics: Demonstrates that cultured neurons can perform complex, real‑time tasks without a host brain, suggesting new avenues for prosthetic control.
- Artificial Intelligence: Provides a biological substrate that can be trained using standard machine learning techniques, offering a hybrid approach that blends organic computation with digital algorithms.
- Ethics and Safety: Raises questions about the moral status of cultured human neurons and the potential for misuse in gaming or other entertainment contexts.
\n
\n
\n
\n
While the chip’s performance remains modest compared to seasoned human players, the speed of training and the simplicity of the interface signal a shift toward more democratized neural engineering. Researchers can now prototype brain‑like systems in a matter of weeks, accelerating both scientific discovery and commercial development.
\n\n
Future Directions and Potential Applications
\n
Beyond gaming, the same technology could be adapted for:
\n
- \n
- Neurorehabilitation:
Leave a Comment