Cortical Labs trains brain cells to play Doom
Multiple outlets, including AFP and The Economist, report that Australian startup Cortical Labs used a "biological computer" to teach lab-grown human neurons to play the 1990s shooter "Doom." Each CL1 device reportedly holds about 200,000 living human neurons grown from stem-cell-derived, blood-donor material and mounted on a silicon chip wired to electrodes. Researchers converted the game environment into electrical stimulation patterns, and Cortical Labs' Alon Loeffler told AFP the cultures initially kept "walking into walls" before learning to target enemies more reliably, building on earlier work training cultures to play Pong. For data-science and computing practitioners, the demo is best read as a proof-of-concept closed sensorimotor loop rather than a benchmark result: independent replication, published performance metrics, and ethics review of human-derived tissue remain the open questions.
The interesting part for AI and computing practitioners is not that neurons can "play" Doom, but that a lab has built a repeatable stimulation-and-readout loop that treats living neural tissue as a trainable, goal-directed substrate. That reframes familiar machine-learning ideas, such as reward-shaped behavior, closed-loop control, and iterative policy improvement, as properties that can, in principle, emerge in wet biology as well as silicon, even though the underlying mechanisms and constraints are very different.
What happened
Multiple news outlets, including The Economist, RTE, The Straits Times, and Scientific American, report that Australian startup Cortical Labs trained lab-grown human neurons on a silicon substrate to play the 1990s first-person shooter "Doom." Reporting states each "biological computer" contains about 200,000 living human brain cells derived from stem-cell protocols using donor blood, mounted on a device the company calls the CL1. The project follows earlier demonstrations in which Cortical Labs taught neuronal cultures to play Pong; Scientific American also reports that an independent researcher paired the neurons with a standard learning algorithm and found the hybrid system outperformed the algorithm running alone.
Technical context
The CL1 device interfaces cultures with arrays of electrodes. Researchers converted the Doom environment into patterns of electrical stimulation that activate subsets of neurons; distinct neuronal activity patterns were associated with in-game actions such as firing or turning. Alon Loeffler, identified as Cortical Labs' senior application scientist, told AFP the cultures were initially "walking into walls a lot, shooting the walls, turning around, doing funny things like that," before they "started targeting the enemies more regularly and correctly." Demonstrations that couple living neural tissue with electronics extend a longer line of work on biohybrid and neuromorphic systems; labs frequently use simple game environments such as Pong and Doom because they provide compact, well-defined state and reward signals that map cleanly to stimulation and readout channels.
For practitioners
This is a proof of concept showing that cultured human neuronal networks can perform goal-directed behavior in a closed sensorimotor loop when paired with electrode arrays and real-time stimulation. The demo is not directly comparable to conventional compute benchmarks, since latency, energy accounting, durability, and scaling constraints differ markedly between living cultures and silicon. Ethical governance and reproducibility are recurring nontechnical constraints in biohybrid research, and Cortical Labs describes the effort as exploratory rather than immediately practical.
What to watch
- •Whether peer-reviewed documentation or datasets are released that quantify performance, variability, and training protocols beyond media summaries.
- •Experimental controls and replication attempts from independent labs, since reproducibility is a known barrier in complex biological systems.
- •How researchers define metrics for energy efficiency, latency, and lifespan compared with conventional neuromorphic hardware.
- •Regulatory and ethics discussions around the sourcing, consent, and long-term handling of human-derived neuronal cultures.
Key Points
- 1The demonstration shows cultured human neurons can learn goal-directed control in a closed sensorimotor loop, useful for exploring biohybrid computing tradeoffs.
- 2The 200,000-cell cultures suit small, well-defined game environments, but scaling, noise, and cell lifespan remain open practical constraints for biohybrid computing.
- 3Independent replication, published performance metrics, and ethics oversight of human-derived tissue will determine how far this research moves toward practical use.
Scoring Rationale
This is a widely-covered, verified biohybrid-computing demonstration (Economist, Scientific American, Smithsonian, Guardian) that shows a novel closed-loop training method, but it remains an exploratory proof of concept at small scale with no peer-reviewed performance data yet, capping it below major-breakthrough territory.
Sources
Public references used for this report.
View 4 more sources
- How human neurons on a chip learned to play Doomscientificamerican.com
- A Clump of Human Brain Cells on a Computer Chip Learned to Play ...smithsonianmag.com
- A petri dish of human brain cells is currently playing Doom. Should ...theguardian.com
- Australian researchers teach brain cells to play 'Doom'banglanews24.com
Practice interview problems based on real data
1,625 SQL & Python problems across 15 industry datasets — the exact type of data you work with.
Try 250 free problems
