# Scientists Taught 200,000 Human Brain Cells to Play Doom. This Changes Everything.

```yaml
url: "https://forg.to/articles/scientists-taught-200000-human-brain-cells-to-play-doom-this-changes-everything"
title: "Scientists Taught 200,000 Human Brain Cells to Play Doom. This Changes Everything."
excerpt: "Human brain cells in a lab learned to play Doom, showcasing "Synthetic Biological Intelligence" and a potential future for computing."
author: "Kislay"
author_url: "https://forg.to/@kislay"
published: 2026-03-08
updated: 2026-09-23
read_time_minutes: 12
tags: ["biological-computing", "synthetic-biological-intelligence", "cortical-labs", "wetware-as-a-service", "neuroscience-ai"]
cover_image: "https://res.cloudinary.com/dxkpmcrel/image/upload/v1772956145/buildinprocess/articles/wotdpmjuswh2w12dta4k.jpg"
```

## TL;DR

Human brain cells in a lab learned to play Doom, showcasing "Synthetic Biological Intelligence" and a potential future for computing.

## Content

A petri dish just learned to navigate hell.

200,000 human neurons, grown in a lab, mounted on a silicon chip, and taught to play one of the most iconic video games ever created.

Not a simulation. Not an AI model. Actual living human brain cells firing electrical signals to shoot demons and explore 3D environments.

If that sounds like science fiction, welcome to 2026.

The company behind this is Cortical Labs, an Australian biotech startup that's been pushing the boundaries of what they call "Synthetic Biological Intelligence" since 2021.

And what they just accomplished isn't just a cool tech demo. It's a glimpse into a future where computers are no longer made of silicon alone.

They're made of living tissue.

## From Pong to Doom: The Four-Year Journey

Cortical Labs first made headlines in 2022 when they taught a cluster of 800,000 human neurons to play Pong.

That achievement was groundbreaking. It demonstrated that living neurons in a dish could respond to feedback, learn patterns, and exhibit goal-directed behavior in real time.

But Pong is simple. It's a 2D game with direct cause and effect: ball goes up, paddle goes up. The relationship is linear and predictable.

Doom is chaos.

It's a 3D environment with enemies that move unpredictably. It requires spatial navigation, threat detection, prioritization of targets, and exploration of unknown areas.

For a system with no eyes, no concept of "game," and no understanding of what a demon even is, Doom represents an entirely different level of complexity.

And yet, within a week of development time, the CL1 biological computer was playing it.

Not well, mind you. The neurons play like a complete beginner who's never seen a computer before. Which, to be fair, they haven't.

But they show evidence of learning. They seek out enemies. They shoot. They spin to explore. And critically, they're improving.

## How a Petri Dish "Sees" Doom

Here's the part that sounds like magic but is actually elegant engineering:

The neurons can't see the game. They don't have eyes. They don't have a concept of visual information.

So how do they interact with a 3D shooter designed for human eyes?

The answer: translation.

The CL1 system takes the visual environment of Doom and converts it into electrical signals that neurons can understand.

Think of it like this: when you play Doom, you see demons, walls, and weapons. Your brain processes that visual information and tells your hands what to do.

For the CL1, the game state is encoded as patterns of electrical stimulation sent to the neurons through 59 electrodes on a multi-electrode array.

When the neurons fire back in response, their electrical activity is decoded into in-game actions.

If they fire in one specific pattern, the character shoots. Another pattern makes him move right. Another makes him turn.

Over time, through a feedback loop of rewards and reinforcement, the neurons learn which firing patterns produce useful outcomes.

Hit an enemy? Positive feedback. Die to a demon? Negative feedback.

The synaptic connections that lead to good results strengthen. The ones that don't, weaken.

This is biological plasticity, the same mechanism that allows your brain to learn and form memories, happening in a controlled lab environment connected to a video game from 1993.

## The Energy Efficiency Gap That Changes Everything

Here's where this story stops being about video games and starts being about the future of computing.

Your brain runs on 20 watts of power.

That's it. About the same as a dim lightbulb.

With that 20 watts, your brain is performing the equivalent of an exaflop: one billion billion mathematical operations per second.

Now let's look at artificial intelligence.

The Oak Ridge Frontier supercomputer, one of the most powerful in the world, recently achieved exaflop computing. But it requires 20 megawatts of power to do it.

That's one million times more energy than your brain uses to accomplish the same computational feat.

Training a single large language model like GPT-4 can consume as much electricity as 21,000 American homes use in a year. Each ChatGPT query uses 25 times more energy than a Google search.

The energy demands of AI are unsustainable at current trajectories. Data centers are consuming power in gigawatts while the human brain accomplishes extraordinary cognitive tasks on the power budget of a household appliance.

This isn't just an efficiency problem. It's an existential challenge for the future of AI.

And biological computing might be the answer.

## Wetware as a Service: The Business Model

Cortical Labs isn't just building biological computers as research curiosities. They're commercializing them.

The CL1 system costs $35,000 per unit ($20,000 each if you buy a 30-unit server rack).

A full rack draws only 850 to 1,000 watts of power total. Compare that to traditional AI server racks that can draw tens of thousands of watts.

And here's the kicker: they're offering "Wetware as a Service."

Through the Cortical Cloud platform, researchers anywhere in the world can rent time on CL1 systems remotely. They don't need to buy the hardware. They don't need a PhD in neurobiology.

They just need to know Python.

Cortical Labs developed a biological operating system called biOS and released a Python-based API on GitHub. This means anyone can now write code that runs on living neurons.

The Doom demonstration? That was completed by Sean Cole, an independent developer with little experience in biological computing. He did it in less than a week using the publicly available API.

Compare that to the 18 months it took Cortical Labs to get Pong working with their original prototype hardware.

The interface problem has been solved. The technology is democratized. And we're just beginning to understand what's possible.

## How the Neurons Are Created

If you're wondering where these neurons come from, here's the process:

Adult donors provide skin or blood samples voluntarily.

Those cells are reprogrammed into induced pluripotent stem cells (iPSCs), essentially rewinding them to an embryonic-like state.

The stem cells are then differentiated into neurons, grown in an orbital shaker for about three months until they form organoids (mini-brains) about 0.5mm in size, each containing roughly 10,000 neurons.

For the CL1, they use about 200,000 neurons total, mounted on a multi-electrode array that allows for both stimulation and recording of electrical activity.

These neurons are kept alive by an internal life-support system that handles temperature control, waste filtration, gas mixing, and circulation. They can survive for up to six months in this environment.

Critically, these are real human neurons. They carry the donor's DNA. They form synaptic connections. They exhibit plasticity. They learn.

But they're not conscious. They have no pain receptors, no sensory organs, no structures that would allow for higher-order cognition.

As Cortical Labs is quick to emphasize: "Just because they're human cells doesn't mean it's a human on that dish."

## What Makes Biological Computing Different

Silicon-based AI learns through massive training datasets. You have to show a neural network millions of cat pictures before it reliably knows what a cat is.

Biological systems learn from sparse data.

When DishBrain's neurons learned Pong, they picked it up in about five minutes. A standard deep reinforcement learning algorithm would take roughly 90 minutes to achieve similar results.

That's not a minor efficiency gain. That's a fundamentally different learning paradigm.

Here are the key advantages of wetware computing:

### 1. Energy Efficiency

As mentioned, biological neurons are orders of magnitude more energy-efficient than silicon. This isn't just about cost savings, it's about scalability. If AI continues growing at current rates, energy consumption becomes the limiting factor.

### 2. Adaptive Learning

Silicon systems learn through backpropagation and gradient descent, computationally intensive processes that aren't biologically plausible.

Biological neurons learn through Hebbian plasticity ("cells that fire together, wire together") and spike-timing-dependent plasticity, mechanisms refined by millions of years of evolution.

### 3. Parallel Processing

Your brain doesn't process information sequentially like a traditional computer. It processes massively in parallel across billions of neurons simultaneously.

Biological computers inherit this architecture naturally, without needing to engineer it.

### 4. Graceful Degradation

Silicon systems typically fail catastrophically. A component works perfectly or it fails completely.

Biological systems exhibit graceful degradation. Neurons can die, connections can weaken, and the system continues functioning. Your brain loses thousands of neurons every day and you don't notice.

### 5. Noise Tolerance

Digital computation requires perfect precision. Flip one bit and your program crashes.

Biological computation is noisy and imprecise, but it's "just reliable enough" to produce "just accurate enough" results with minimal energy use.

## The Ethical Minefield

All of this sounds amazing until you start asking uncomfortable questions.

These neurons are derived from human donors. They carry human DNA. They exhibit learning behaviors that echo human cognition.

Are they conscious? Almost certainly not at this scale.

Could they become conscious? Nobody knows.

If consciousness is an emergent property that arises from a certain level of neural complexity, at what point does a biological computer cross that threshold?

Cortical Labs is adamant that their current systems are nowhere near that point. 200,000 neurons is a tiny fraction of the 86 billion in a human brain.

But what about when they scale up? What about when systems use millions or billions of neurons?

And here's another problem: intellectual property.

If a biological computer becomes commercially successful and contains neurons derived from your skin cells, do you own the "IP" of your genetic blueprint?

We've seen this play out before with Henrietta Lacks, whose cells were used for decades of research without her knowledge or consent, generating billions in value while her family saw nothing.

The biotech industry cannot afford to repeat that mistake.

There are also questions about what these systems should be allowed to do.

Cortical Labs makes biological computers available through an open API. Anyone with Python skills and a credit card can rent time on living neurons.

What's to stop someone from using them to generate AI porn? To model warfare scenarios? To create systems we'd rather not exist?

When you democratize access to powerful technology, you lose control over how it's used.

## The Immediate Applications

Despite the ethical concerns, the potential applications of biological computing are genuinely exciting.

### Medical Research

Biological computers could model brain diseases, test pharmaceutical interventions, and help us understand neurological conditions in ways that animal models and computer simulations cannot.

### Neural Prosthetics

If we can teach neurons to control a video game character, we can teach them to control a prosthetic limb or restore motor function to paralyzed patients.

### Robotics

Hybrid systems combining silicon processing with biological decision-making could create robots that adapt to novel situations in ways purely algorithmic systems struggle with.

### Drug Discovery

Testing how neurons respond to different compounds could accelerate drug development for neurological conditions.

### Brain-Computer Interfaces

Understanding how to translate between digital signals and biological neural activity is crucial for developing effective BCIs.

## What Happens Next

Right now, biological computing is where digital computing was in the 1950s.

The technology exists. It works. But it's expensive, fragile, difficult to scale, and limited in what it can do.

Silicon replaced vacuum tubes not because it was obviously better on day one, but because the trajectory was clear. Each generation got smaller, faster, cheaper, and more reliable.

Biological computing might follow the same arc. Or it might hit fundamental limitations that prevent it from ever competing with silicon at scale.

What's no longer theoretical is that living neurons can compute.

They can learn tasks. They can interact with digital systems. They can be programmed remotely via cloud APIs.

The line between artificial intelligence and biological intelligence is blurring.

And we're watching it happen in real time.

## The Uncomfortable Truth

Here's what keeps me up at night about all of this:

For decades, we've been trying to make silicon chips think like brains.

We've made extraordinary progress. GPT-4 can write essays, solve complex problems, generate art. It's astonishing.

But it's still fundamentally artificial. It's software running on hardware, approximating intelligence through brute-force computation and massive datasets.

What Cortical Labs is doing is different.

They're not approximating biological intelligence. They're harnessing it directly.

They're putting actual human neurons on chips and teaching them to compute.

And if that approach scales, if biological computing becomes economically viable and technically robust, we might look back at digital AI as a temporary detour on the path to something far more powerful and far more unsettling.

The future of computing might not be silicon at all.

It might be wetware.

Living tissue. Biological neurons. Human cells processing information on chips.

The brain, reimagined as infrastructure.

## Final Thoughts

A petri dish of human neurons just navigated a demon-filled hellscape with a shotgun.

Let that sink in.

Not an AI trained on millions of gameplay hours. Not a reinforcement learning algorithm that crunched through terabytes of data.

200,000 living human cells, responding to electrical stimulation, learning through feedback, adapting their firing patterns to achieve goals.

This isn't the future. This is happening right now, in a lab in Melbourne, accessible to anyone with an internet connection and a Python script.

The question isn't whether biology can compute.

The question is what we do with that capability.

And whether we're ready for a world where the line between machine and organism is no longer clear.

---

*The Cortical Labs CL1 system and Cortical Cloud platform are available now. The source code for the Doom implementation is publicly accessible on GitHub. The first 115 units shipped in 2025.*

## Tags

`biological-computing` · `synthetic-biological-intelligence` · `cortical-labs` · `wetware-as-a-service` · `neuroscience-ai`

## Links

- Article: https://forg.to/articles/scientists-taught-200000-human-brain-cells-to-play-doom-this-changes-everything
- Author: https://forg.to/@kislay
- Forg: https://forg.to

---
Published on [Forg](https://forg.to), a social network for people who create and ship on the internet.
_HTML version: https://forg.to/articles/scientists-taught-200000-human-brain-cells-to-play-doom-this-changes-everything_