Living human neurons playing
video games.

Not Artificial Intelligence. Organic Intelligence.

Right now, ~800,000 living human neurons are firing electrical signals that control Pokémon Yellow. No AI. No human controlling it. Just real biology playing a video game - and you can watch it happen live. Read the science.

How It Works

🧫
Neurons Fire
~800K human neurons on a CL1 chip produce electrical signals
→
⚡
Electrodes Read
59 electrodes detect activity and map it to game inputs
→
🎮
Game Responds
Signals are mapped to game inputs via emulated controls
~800K
Living Neurons
59
Electrodes
1
CL1 Unit
LIVE
On Twitch
Day 3
Of Streaming
Phase 1
Observation

The Science

In 2022, Cortical Labs published a landmark paper in Neuron demonstrating that ~800,000 neurons grown on a multi-electrode array could learn to play Pong in under five minutes of closed-loop feedback training, guided by the Free Energy Principle.

By early 2026, they commercialized this into the CL1 - a self-contained biological computer housing neurons on a 59-electrode MEA with integrated life support. They demonstrated it playing DOOM, capturing global attention.

DishBrainPlays takes this technology public. We're livestreaming the neurons playing Pokémon Yellow - starting with pure observation (Phase 1), then introducing feedback training (Phase 2) to see if the culture can learn to play better than random chance.

Inspired by: Cortical Labs' DishBrain Pong experiment (2022), CL1 DOOM demonstration (2026), Fish Plays Pokémon (2014), and Twitch Plays Pokémon (2014).

The Roadmap

Current
Current Phase

Phase 1 — Observation

The culture (differentiated March 5, 2026) is producing random-equivalent output. Neural signals are read from the CL1 and mapped directly to game controls. There is no feedback - the neurons fire, we listen, the game responds. Think of it like Twitch Plays Pokémon, but the chat room is 800,000 biological cells.

  • One-way signal: neurons fire, electrodes read, game responds. No feedback yet
  • 59 electrodes mapped to d-pad, A, B, Start, Select on the CL1
  • Live overlay showing neural activity, firing rate, and button presses in real time
  • Weekday streams with community engagement via Discord
Est. Early May 2026

Phase 2 — Training

The same culture from Phase 1 - now ~60 days differentiated - is introduced to closed-loop feedback for the first time. The neurons don't just play anymore. They receive electrical stimulation based on what's happening in the game. This is where we find out if they can learn.

  • Bi-directional: neurons fire AND receive feedback from the game state
  • Free Energy Principle-inspired stimulation to drive learning
  • Performance tracked against random baseline - can they beat chance?
Performance-Gated

Phase 3 — Expansion

The feedback training system has been validated. The culture has demonstrated learning. Now the project expands: new games chosen by community vote, deeper scientific experiments, and the question - does learning transfer across games, or does each one start from scratch?

  • New game chosen by community vote
  • Does learning transfer across games? The big scientific question
  • Experiments with fresh CL1 cultures (one unit active at a time)
  • Phase 2 results published as preprint or paper

Frequently Asked Questions

No. There is no artificial intelligence involved. These are real, living human neurons - biological cells grown on a chip. They produce electrical signals that are read by electrodes and mapped to game inputs. No code is making decisions. No neural network in the machine learning sense. Just biology.
No. ~800,000 neurons is orders of magnitude fewer than the ~86 billion in a human brain. These cultures do not have consciousness, awareness, or sentience. The neurons produce electrical outputs that are mapped to game inputs - that's it. Right now in Phase 1, the output is essentially random. The real question comes in Phase 2, when we introduce feedback and measure whether performance exceeds random baseline.
That's the big question. In Phase 1 (now), there's no feedback - the neurons fire, we read it, and the game responds. In Phase 2 (est. early May 2026), we introduce closed-loop feedback based on the Free Energy Principle. Cortical Labs demonstrated that neurons can learn Pong this way. Whether they can learn something as complex as Pokémon Yellow is what we're here to find out. If results are ambiguous, we'll say so.
Twitch Plays Pokémon had millions of humans typing commands into chat. DishBrainPlays has ~800,000 biological neurons firing electrical signals. No humans are controlling the game - not through chat, not through a controller, not at all. The neurons are the only player.
All neural cultures are derived from donated biological material (adipose tissue) under established consent protocols, supervised by qualified research personnel. The project operates within Cortical Labs' ethical framework for CL1 usage. No human subjects are involved - the cultures are in vitro systems maintained in a controlled lab environment.
Fat tissue is donated, stem cells are extracted, and those stem cells are differentiated into neurons that are grown on the CL1's multi-electrode array. The scientific term is autologous adipose tissue donations via mesenchymal stem cell protocols.

Stream Schedule

Next Stream
Check Twitch for schedule
Weekdays, ~2 hrs/session. Follow on Twitch for go-live alerts.
Days
Monday - Friday
Sessions
~2 hours
Platform
Twitch
Phase 2+
6+ hrs/day

Watch Live

Live

Highlights

Watch highlights on TikTok

Never Miss a Stream

The neurons don't wait. Follow on Twitch for live alerts or join the Discord for schedule updates and community chat.

Follow on Twitch Join Discord