203: Chapter 203 Topological Neural Networks

He turned on his computer and pulled up all the experimental records for the prosthetic limb project.

He had looked through this data many times, but never from this perspective.

Not looking at why it failed, but looking at why it succeeded.

He flipped to the earliest batch of data, back when they were still using Bullfrogs for experiments, when the artificial neurons could just barely transmit signals, and Li Wei was so excited she almost jumped up.

But looking closely at the data, the success rate during that period wasn't high; succeeding three or four times out of ten experiments was already quite good.

The signal would come and go, the intensity would fluctuate, and sometimes it was clearly connected, but when stimulated, the Bullfrog's leg wouldn't react.

Where was the problem?

Chen Youhua flipped through the records and found the experimental notes Li Wei had written at the time:

"12th experiment, signal connection rate 100%, but no obvious contraction of the hind leg after stimulation. Upon inspection, it was discovered that there was a tiny displacement at the contact surface between the artificial neuron and the severed nerve end. After re-fixing it, the response returned to normal."

The contact surface issue.

He had known about this for a long time.

Later, they spent a lot of time solving this problem, eventually using a silk fibroin scaffold to fix the neuron and the severed nerve end together, letting them grow together.

But was the act of fixation itself also transmitting some kind of information?

Chen Youhua leaned back in his chair and thought for a moment.

He recalled a detail.

In the first two weeks after surgery, the signal quality was always very unstable.

Sometimes it was good, sometimes it was poor, and sometimes it was completely absent.

But after two weeks, the signal would slowly stabilize, getting better and better, finally reaching a very stable state.

Li Wei's explanation was that the nerves were growing, and two to three weeks was the time window for the nerve processes to grow and establish stable connections with the scaffold.

But that wasn't what Chen Youhua was thinking about now.

He was thinking about why the nerves would actively grow onto the scaffold.

The scaffold itself was just a physical structure; it had no biological activity and wouldn't secrete any growth factors to attract nerves. The reason the nerves grew onto it was because they would grow anyway.

The severed nerve endings, the nerve processes would extend out like tentacles, exploring everywhere, grabbing whatever they touched. If they touched a segment of an artificial neuron and the signal could pass through, they would grab onto it firmly, grabbing tighter and tighter.

In other words, it wasn't the scaffold that attracted the nerves; it was the signal that attracted the nerves.

The nerves were looking for a path; they could sense which path was open, and then they would move along that path.

This discovery was just an incidental observation in the prosthetic limb project.

Their goal was to build a bridge, not to study how nerves find their way. Once the bridge was built and function was restored, the task was complete.

But now, Chen Youhua felt that this observation might be a key to brain-computer interfaces.

...

He wrote a sentence on the whiteboard: "Nerves will find their own way."

Then he drew an arrow under this sentence, pointing to another word: "How to guide?"

In the prosthetic limb project, what guided the nerves was the signal.

The nerve endings sensed the electrical signals transmitted by the artificial neurons, knew that this path was open, and grew in this direction.

So, if an electrode was placed on the cerebral cortex and electrified, would the nerves also grow onto the electrode?

Theoretically, yes.

But the problem was that the nerves in the brain were far more complex than peripheral nerves.

Peripheral nerves were highways, bundled together, with clear directions, from A to B.

The cerebral cortex was an interchange, dense and intertwined, with every path connecting to countless other paths.

If you placed an electrode on the cerebral cortex and electrified it, the surrounding nerves would indeed be attracted.

But how would you know they grew to the right place? How would you know that the ones attracted were the group of neurons you wanted to control, and not the group next door that controlled breathing?

Chen Youhua thought about this problem for an entire day.

In the afternoon, he found something: that small experiment Wang Zhe did in the prosthetic limb project.

At that time, they had just created bidirectional artificial neurons, and on a whim, Wang Zhe built a circuit using electronic components, connecting the artificial neuron to a model aircraft servo.

As soon as the Bullfrog kicked its leg, the servo would turn.

The experiment was very simple, even a bit crude, but the effect was very intuitive.

Wang Zhe said something at the time: "If we use the artificial neuron in reverse, can we use electrical signals to stimulate the nerve and make the Bullfrog think it kicked its leg?"

This was treated as a joke at the time, and everyone laughed it off.

But now Chen Youhua felt that this "using it in reverse" might be the answer.

Guiding nerves to grow onto the electrode relied on electrical signals.

But if you just applied electricity, the nerves would just grow randomly, running toward whichever side had the stronger signal, and eventually, they would all crowd together, and no one could tell which was which.

But what if you could make each electrode emit a unique signal?

It would be like giving each electrode a name.

The surrounding nerves would sense this signal and know that there was a path here that led to the "right thumb."

There was also a path over there that led to the "right index finger."

The nerves would choose for themselves which way to go, moving to the corresponding electrode.

This process would be very slow.

Nerves could only grow a few millimeters a day; growing from the surface of the cortex to the electrode might take weeks or even months.

But slow wasn't necessarily a bad thing.

Slow meant precision.

The nerves had enough time to "learn" which signal corresponded to which action, and enough time to establish stable connections.

By the time the connection was truly established, the brain would have already learned, adapted, and treated this new path as its own.

Just like the monkey in the prosthetic limb project.

In the first few days after the surgery, the robotic arm moved very stiffly, like it was having a cramp.

But after a few weeks, the movements became more and more fluid.

It wasn't that the algorithm had improved; it was that the monkey's brain had learned.

Chen Youhua stood up, walked to the whiteboard, and circled the sentence "Nerves will find their own way."

His train of thought had opened up, but there were also more questions.

He added a few more lines to the whiteboard.

The first question: How to make the signals unique?

Not the "how-to" in terms of technology, but the "how-to" in terms of logic.

Each electrode needed to emit a unique signal; this signal had to be recognizable by the surrounding nerves, had to be able to remain constant for weeks or even months, and had to be understandable by the brain.

It couldn't be treated as noise, nor could it be treated as pain.

This reminded him of coding.

It was like radio broadcasting; each station had its own frequency, and you tuned into that frequency to listen to that station.

Nerves were the same; they only reacted to specific patterns of electrical signals, while other signals were ignored.

So, what kind of signal pattern could nerves recognize?

This question actually had an answer in the prosthetic limb project.

He pulled out that set of control experiments Li Wei had done.

At the time, they were testing the signal transmission efficiency of artificial neurons and used several different electrical stimulation patterns.

Continuous, pulsed, frequency-varying, intensity-varying.

The results showed that nerves responded best to pulse patterns, especially those "short, high-frequency, intermittent" pulses.

Why? Li Wei's explanation at the time was: This pattern was most similar to the Nature firing pattern of biological nerves.

In other words, nerves only recognized the language they were familiar with.

So when designing signals for the electrodes, one had to use the nerves' own language.

Not inventing a new language, but learning an old one, the "mother tongue" of the nervous system.

This direction, he could do.

The electrophysiological data accumulated in the prosthetic limb project was enough for him to figure out the basic laws of nerve firing.

But another problem arose: with so many electrodes, how to place them?

What he wanted to make was a high-density electrode array, perhaps hundreds, thousands, or even more.

Each one needed to have its own signal pattern, and each one needed to establish a connection with a specific group of neurons.

This wasn't just a material problem; it was also a layout problem.

If they were just placed randomly, the nerves would be in chaos when they grew over.

The signals here and the signals there would mix together, and no one would know which way to go.

He remembered that "semi-autonomous" plan in the prosthetic limb project.

The scaffold that Li Wei and the others created allowed the nerves to find their own direction without needing manual guidance.

The success of that plan didn't rely on some profound technology, but on a deep understanding of the laws of nerve growth, knowing which way nerves liked to run and which way they didn't.

Knowing under what conditions it would go straight and under what conditions it would turn.

The same line of thinking could be applied to brain-computer interfaces.

Not by making a fuss about the electrodes, but by making a fuss about the environment around the electrodes.

Designing a microstructure that made the nerves "feel" that a certain path was one they should take, and a certain path was one they shouldn't.

This direction, he could do.

The data accumulated in the prosthetic limb project regarding nerve growth was enough for him to design a preliminary road network.