Shulou(Shulou.com)12/24 Report--

The AI system, made up of a "mini brain" and microelectrodes constructed by real human brain cells, has been able to perform speech recognition--

The kind that accurately recognizes the voice of a particular person from hundreds of sound clips.

Recently, a cutting-edge brain-like study appeared in the Nature sub-journal.

This particular AI system even allows unsupervised learning:

The researchers just played the audio clips over and over again without providing any form of feedback to tell the system whether the answer was right or wrong.

In the end, the accuracy of the system increased directly from 51% to 78% after two days of training.

How on earth did this come true?

The main purpose of the invention of organ-like neural network is to solve the problem of high energy consumption of silicon chip.

Generally speaking, the way to solve this problem is to rely on brain-like calculation.

However, most of the "traditional" brain chips designed under this idea are directly based on the principle of digital electronics, and the ability to completely imitate brain function is really limited.

Here, the study directly uses something called "organ-like":

It refers to miniature organs that can be grown using human stem cells in the laboratory, including some of the key characteristics of the organs.

Specifically, the researchers connected brain-like organs (shaped like balls) of living brain cells to a high-density microelectrode array to create a system called Brainoware.

The function of microelectrode in Brainoware is to send electrical signals to similar organs to achieve the purpose of transmitting information to the "brain"; the other is to detect the discharge response of brain nerve cells, and then give it to external equipment for reading and analysis.

Such a system can show a function similar to that of a neural network and can carry out unsupervised learning.

Connect it to specific hardware and you can be trained in speech recognition.

In the specific task, the researchers converted 240 audio clips of eight Japanese vowels into signal sequences and then sent them to the system to recognize someone's voice.

At first, the accuracy of Brainoware was only 30% and 40%.

But after two days of training, it can identify specific speakers with 78% accuracy.

The author emphasizes here that the so-called training only repeats audio clips without giving any feedback, that is, the so-called unsupervised learning.

It is important to note, however, that currently Brainoware can only identify who is speaking, but does not understand anything.

After the experiment, the researchers tried to use a drug to block the formation of new connections between nerve cells in brain organs.

It is found that after this operation, the accuracy of the system will not be improved.

This suggests that the learning ability of Brainoware depends on neuroplasticity, the authors explain.

Will the computers of the future be made up of brains? In March this year, the team actually used the system to try to predict H é non graphs (a dynamic system that can show chaotic behavior in the field of mathematics).

Results Brainoware was also found to be more accurate than artificial neural networks without long-term and short-term memory units after four days of unsupervised learning (each day represents a training cycle).

By contrast, the latter has gone through at least 50 training cycles.

Further, an Australian research team tried to teach the "intraplate brain" to play ping-pong, only to learn it in five minutes, 17 times faster than the AI.

So in the future, will computers be made up of brains?

I don't know.

As introduced by the author of this article, their research is currently a proof of concept, and there are still many problems to be solved:

For example, the performance of Brainoware systems can still be improved, but the most important problem is that similar organs can only survive for one or two months.

Moreover, although Brainoware itself does not require much power consumption, the power consumption level of the external devices that keep it running is not low.

Such as a series of questions and so on.

In general, some scientists predict that a real general-purpose biological computing system may take decades to build.

But in any case, its research is helpful for us to further understand the learning mysteries of the human brain.

Reference link:

[1] https://www.nature.com/articles/s41928-023-01069-w

[2] https://www.newscientist.com/article/2407768-ai-made-from-living-human-brain-cells-performs-speech-recognition/

[3] https://www.genengnews.com/topics/artificial-intelligence/brainoware-organoid-neural-networks-inspire-brain-ai-hardware/

This article is from the official account of Wechat: quantum bit (ID:QbitAI), author: Fengcai

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.