Researchers proved that tiny lab-grown brains can process information and read Braille just like a real computer

Three brain organoids read Braille letters far better than one, and the living tissue held up even when researchers damaged the data.

Could a computer one day run on living brain cells instead of silicon? A small experiment has just given that question a plain, partial answer.

Specks of human brain tissue, known as organoids, were grown in a dish in a lab. They received the shapes of Braille letters as electrical pulses, and a program read the letters back from how the cells fired.

One speck got the letter right about 61 percent of the time. Three specks, with their signals combined, got it right 83 percent of the time.

Tianyi Liu, a doctoral student at the University of Bristol, led the work with Hemma Philamore and Benjamin Ward-Cherrier in the University's School of Engineering Mathematics and Technology.

Brain organoids tested in a lab

The brain organoids are clumps of human neurons and support cells grown in a lab. Each is about a fiftieth of an inch (half a millimeter) across, the width of a mechanical pencil's lead.

Nothing in this experiment learned anything, and that matters for how far the result reaches.

"We can take information from a real tactile sensor, turn it into a form that a living neural network can receive, and then recover meaningful information from the organoid response," said Liu.

"That gives us a practical way to connect artificial sensors with biological neural systems, which are a very interesting possible route towards more energy-efficient computing," she added.

The attraction is power. A human brain runs on roughly 20 watts. Silicon brain-inspired chips scaled up to the size of a cortex would need kilowatts to megawatts, the team wrote.

Other labs have already had lab-grown brain cells tell Japanese vowels apart and play a simple video game.

Turning Braille into pulses

The brain organoids never touched a Braille dot. A touch sensor did that for them. The team mounted a sensor with a soft silicone pad, a kind of electronic skin, on a small robot arm.

The arm slid the pad across each of the 26 letters on a 3D-printed Braille board about 9 inches by 5.5 inches (23 by 14 centimeters).

A camera inside the pad recorded tiny shifts in a grid of markers as the dots pressed into the gel, the way a robot fingertip gets its sense of touch. Each letter was scanned 10 times at each of five pressing depths, 1,300 recordings in all.

The team split each recording into eight regions, one for each electrode under an organoid. The touch in each region then became a train of electric pulses.

More contact in a region meant more pulses, between four and ten, and a longer contact meant longer pulses.

The organoids live on electrode arrays run by a company called FinalSpark, and the team reached them over the internet. Kept under a steady flow of nutrient liquid, an organoid can stay electrically active for up to about 100 days.

Three organoids beat one

For each letter, the team counted the cells' firing on each of the eight electrodes for half a second after the pulses arrived. A standard pattern-sorting program then learned which firing patterns went with which letter.

With a single organoid, the program named the right letter about 61 percent of the time, where guessing would get just under 4 percent right.

Three organoids on the same array, each read on its own, scored 59, 62 and 67 percent. Combined into one pattern, the three organoids' signals lifted the score to 83 percent.

So the gain didn't come from one unusually good organoid. It came from combining them.

The brain organoids differ slightly

Why that works isn't settled. The team's best guess is that the three organoids differ slightly, so each one transforms the same input a little differently. The program then gets three partly different views of one letter.

"Different organoids may have slightly different network connections, developmental states, and ongoing activity, so the same input can produce similar but not identical responses," Liu said.

"Combining those responses may give us complementary information that we would not get from a single organoid. At the same time, if every organoid behaves differently, it obviously makes it harder to build a predictable and standardized system," she added.

Noise made the difference

There was an obvious objection. The pulse patterns were designed to preserve each letter's shape, so maybe the program was reading the team's own encoding and the tissue was only passing it along.

The team tested that directly. Given the pulse patterns with no organoid in between, the same program scored 82 percent, almost exactly the three-organoid result. On a clean signal, the organoids made almost no difference.

Then the team corrupted half of the channels in four ways: two kinds of random noise, some entries set to zero, and some values tripled. Reading straight from the pulses, the program fell to between 55 and 67 percent.

Reading from the three organoids, it held at 83 and 82 percent under random noise, and fell to 72 and 68 percent when entries went missing or were tripled.

In every one of the four tests, the organoids came out ahead. That, more than the headline score, is the contribution the team claims for the living tissue. Ward-Cherrier, the senior author, summed up the two results.

"After processing by the organoids, the tactile data still contains enough information to classify Braille letters with good accuracy. Combining outputs from several organoids also showed an improvement in performance."

What remains unknown

The signal in this experiment ran one way. Pulses went in, firing came out, and nothing went back to the tissue to say whether it had been right. No organoid got feedback, so none could get better with practice.

"We showed that tactile information could be encoded, processed through the organoids, and decoded, but we did not demonstrate learning or long-term changes in the organoids," Liu said.

"We also tested one tactile task using a fixed encoding scheme and a relatively simple measure of neural activity."

Eight electrodes per organoid is also a coarse view of a clump of cells, too coarse to see fine patterns in the firing.

And the team hasn't measured how alike the three organoids' responses are, which would show whether they carry different information or three copies of it. Whether a fourth organoid would help is also untested.

Can living tissue be a computer part?

In 2024, a robot with a camera in its fingertip read Braille faster than a person can, and nothing here beats that.

The question this study asks is different: whether living tissue can hold a job inside a machine, and do it on a fraction of the power.

In the next experiments, the team plans to give the organoids textures and edges to feel, and then feedback on what they do with them.

Only then can anyone ask whether this tissue can learn.

The full study was published in the journal Frontiers in Artificial Intelligence.

-- -

Check us out on EarthSnap, a free app brought to you by Eric Ralls and Earth.com.