Try balancing a ruler vertically on the palm of your hand while walking. It’s not easy. Your eyes constantly track its movement. Your arm and hand make tiny adjustments to prevent tilting. All the while, your brain sparks with activity with one clear goal: Keep the ruler upright.

Scientists have now trained mini brains, or brain organoids, to master the same problem, simulated in the digital realm, with electrical zaps alone.

Mini brains have grown popular with researchers since their invention over a decade ago. Commonly made from stem cells, organoids are jam-packed with neurons that form densely connected networks. Earlier versions loosely resembled the developing brains of preterm babies; now they can mimic the neural wiring of a kindergartener. As the blobs become more sophisticated, scientists are asking: Can they learn?

In the new study, researchers challenged the mini brains with a classic engineering task similar to balancing a ruler on your hand. Mastering the task takes practice, but our brains are wired to receive feedback, often in the form of a small jolt of electrical activity. Called reinforcement learning, the technique has already been adapted to train AI—and now, mini brains too.

The goal isn’t to replace silicon-based controllers with living tissue. It’s to test the organoids’ ability to listen and learn and reveal how they break down.

“We’re trying to understand the fundamentals of how neurons can be adaptively tuned to solve problems,” study author Ash Robbins at the University of California, Santa Cruz said in a press release. “If we can figure out what drives that in a dish, it gives us new ways to study how neurological disease can affect the brain’s ability to learn.”

The Mini Revolution

Attaching living brain tissue to computers sounds like science fiction. But brain organoids have already made it reality.

These blobs of brain cells often start life as skin cells that have been turned back into stem cells. After bathing in a special cocktail of nutrients, they develop into various types of brain cells that self-organize into intricate three-dimensional structures similar to parts of the brain. Neurons form networks, ripple with electrical waves, and when connected to other tissues—such as an artificial spinal cord and lab-grown muscles—can control them.

Bioengineers have taken notice, envisioning organoids as potential living processors. Our brains use far less power and are more adaptable than the most advanced neuromorphic chips and brain-inspired AI. Brain organoids linked together into computers could theoretically enable computation in a dish at a fraction of the energy cost.

There are hints this blue-sky idea could work. Scientists have taught hundreds of thousands of isolated neurons to play the video games Pong and, more recently, Doom. Separately, researchers used cultured neurons to control the simple movements of a vehicle.

But mini brains are different. Unlike isolated neurons, organoids’ 3D structures and connections are harder to decipher. Yet predictable learning is essential to realizing “organoid intelligence.” Their electrical activity needs to rapidly adapt to inputs, strengthening or weakening circuits.

Reinforcement learning from trial and error is a perfect test. When we succeed at a new task, neurons in the brain’s reward center blast dopamine and rewire their connections. Failures don’t bring about similar activity. Over time, we learn not to touch a hot pan, take care when hammering a nail, and other life lessons.

But cortical organoids, which resemble the outermost part of the brain, lack neurons that communicate using dopamine. Can they still learn through experience?

Zapping Away

The new study tackled the question with a hybrid organoid-computer system. The team grew cortical organoids from mouse stem cells. These then self-organized into neural networks and developed a layered structure within a month.

The researchers chose this type of brain organoid “due to the cortex’s well-established role in adaptive information processing and its ability to encode, decode, and modify responses to novel inputs,” they wrote.

The team embedded the brain blobs on a chip that captures their electrical pulses and interacts with a computer to “teach” the mini brains and process data. (The chip’s sensors don’t cover the entire organoid as more recent devices do.)

After recording spontaneous activity, the team figured out how best to stimulate the organoids and built a programmable system with a simple interface.

“From an engineering perspective, what makes this powerful is that we can record, stimulate, and adapt in the same system,” said study author Mircea Teodorescu.

Next, the team challenged the organoids with the cartpole problem, a classic engineering task that asks the player to balance an upright pole on a moving cart. If the pole tips over a certain angle, it’s a fail. The player has to constantly adjust the cart as its cargo wobbles.

To train the organoids, the scientists delivered electrical zaps after the pole tipped too far to either side and tracked the responses. In essence, the mini brains played a video game, with human coaches nudging them toward success. The team grouped performance—how long the system balanced the pole—into sets of five trials, each ending when the pole fell. If the most recent performance improved over the previous 20 trials, they considered it a success and delivered no zaps. If performance didn’t improve, the team gave the organoids a zap.

“You could think of it like an artificial coach that says, ‘you’re doing it wrong, tweak it a little bit in this way,’” said Robbins.

Compared to random or no zaps, the rewarding zaps boosted the success rate from 4.5 to 46.5 percent in continuous trials, suggesting the organoids learned from electrical cues alone—without dopamine. A closer look showed the cells released another chemical that strengthens neural connections, and blocking the process prevented them from learning.

“This demonstrates that biological neural networks can be systematically modified through precise electronic control,” wrote the team.

However, the learning didn’t last. After roughly 45 minutes without stimulation, the organoids’ performance reset to baseline. Their fleeting memory may reflect the lack of neural highways required for long-term memory. The team is now culturing multiple types of brain organoids together—each mimicking a different region—to potentially preserve learning and memory.

“These are incredibly minimal neural circuits. There’s no dopamine, no sensory experience, no body to sustain, no goals to pursue,” said Keith Hengen at Washington University in St. Louis, who did not participate in the study. But they could still be nudged toward solving a real control problem. “That tells us something important: The capacity for adaptive computation is intrinsic to cortical tissue itself, separate from all the scaffolding we usually assume is necessary.”