A recent breakthrough in neuroscience and bioengineering shows that lab-grown “mini brains” can learn to solve a real engineering problem. While early-stage, this points toward a future where living neural systems may complement traditional computing.
What Are “Mini Brains”?
Brain organoids are small clusters of neural tissue grown from stem cells.
They self-organise into three-dimensional structures with basic neural networks and electrical activity, making them useful for studying brain function and development.
The Experiment: Solving the Cartpole Problem
Researchers tested these organoids using the cartpole problem, where a system must keep a pole balanced by making continuous adjustments. The organoids were connected to a computer that translated their neural signals into actions, while electrical stimulation provided feedback.
This created a closed-loop system similar to reinforcement learning, where performance improves through feedback.
Learning Without Dopamine
The organoids improved their performance significantly, increasing success rates from around 5 percent to over 40 percent. Notably, this happened without dopamine, suggesting learning can occur in simpler neural systems than previously assumed.
How the System Worked
The organoids were placed on a chip that recorded activity and delivered stimulation. When performance worsened, corrective signals were applied. Over time, the neural connections adapted, improving control of the task.
Limitations
The learning was short-lived. After roughly 45 minutes without stimulation, performance returned to baseline. This reflects the simplicity of organoids, which lack the structures needed for long-term memory.
Why This Matters
This research is not about replacing computers, but about understanding how learning emerges from neural systems. It may improve models of brain function, support drug development, and inform new approaches to AI.
Conclusion
Mini brains solving engineering tasks highlights that learning may be an inherent property of neural tissue. While far from practical use, it marks a step toward integrating biology with computation and rethinking how intelligence can be built.