news

한어Русский языкEnglishFrançaisIndonesianSanskrit日本語DeutschPortuguêsΕλληνικάespañolItalianoSuomalainenLatina

Is The Matrix really going to become a reality? The brain in a vat may no longer be a plot in a science fiction novel.

Just recently, Tianjin University and Southern University of Science and Technology jointly developed an interactive system.

Simply implanting a brain into the robot (in the biological sense), without the need for the big AI models that we have been talking about in the past two years, it can perform simple operations such as dodging, tracking, and grasping.

Wow... Brain? Robot? You may still be confused after reading this.

In fact, this is nothing new. In the industry, this kind of research is calledOrganoid Intelligence（ OI ）。

In a corner that few people care about, this track has quietly developed to a new stage. To put it simply, organoid intelligence is to use artificial brains to control various machines to help humans complete some tasks (similar to the goal of AI).

When it comes to brain controlling machines, everyone may immediately think of brain-computer interface, but it is completely different from the organoid intelligence we are going to talk about today.

Brain-computer interface involves implanting electronic devices into the existing human brain, while organoid intelligence is about starting from scratch and cultivating stem cells into "brain-like cells" to act as "brain PUs".

If we define it in terms of software and hardware, then this newly developed human brain should be called “Wetware”(Wetware). After all, it is a living thing and can only "work" properly in a petri dish.

The requirements for the working environment are also stricter than those for ordinary AI chips. It must be well fed, and the humidity and temperature must be guaranteed to keep bacteria and viruses out.

Anyway, it is as troublesome as it can be, and even if you take good care of this "brain", its lifespan is far shorter than that of an ordinary graphics card.

So far, artificial brains can generally only work for about 100 days, and the longest record is only 12 months.

Like everyone else, Shichao was a little puzzled at first about the usefulness of this kind of organ intelligence, which is similar to the goal of AI. However, after so much effort, the results were still not good enough.Far behind the current progress of AI。。。

But when I studied it more deeply, I found the highlights of this kind of organ intelligence. It is entirely possibleSolve the presentAI Development bottleneck。

It's still the same old story.Energy consumption issuesWe have already spent two articles to explain to you how much energy the large model consumes.

If we give a large model a simple question, it may need to use a lot of computing power to solve it, which is not only time-consuming but also consumes a lot of energy. If we give it to a human brain, it may be able to answer it without much effort.

The human brain can do this easily, and it all comes down to the inherent advantages of biological neurons. . .

The nerves in the brain areplasticityof,Not only can it grow new nerve tissue, it can also expand existing connection channelsIn this way, there is no need to worry about the quality of the data. In any case, the nerves in the brain will adapt themselves and learn from the experience quickly and well, while also ensuring that the energy consumption is minimal.

Of course, AI scientists are more aware of the natural advantages of the human brain than we are. In recent years, many people have imitated the human neural structure to develop AI models.

But there is also a problem. Scientists have not yet thoroughly studied the specific operating mechanisms of the brain. It is impossible to imitate the brain in AI architecture.

Organoid intelligence is different. It is also related to the "brain", but its "life creed" isNot seeking to understand, there is no need to fully understand the mechanisms inside the human brain.

In addition, the problem of AI energy consumption has become increasingly serious in the past two years, so the results of organoid intelligence have become more and more exciting in the past two years.

For example, the following computer game Pong is similar to table tennis.Can you believe that this is the effect achieved by artificial brain cells after learning for 5 minutes? . .

The results may not be as good as AI performance, but at that time, if you want AI to learn to play this game, it will take at least 90 minutes of training.

This is the world's first sentient artificial brain, cultivated by an Australian startup company Cortical Labs, with a total of 800,000 brain cells.

Don’t start imagining all kinds of things when you hear 800,000 brain cells.Extracting neurons from the human brainIf that were the case, it would have caused an ethical scandal long ago.

The actual process is to use induction to differentiate pluripotent stem cells into cortical neuron cells, and then further cultivate them.

After the brain cells are ready, the remaining steps are similar to the brain-computer interface. Use electrodes to connect the "artificial brain" to the machine, and the brain can work directly.

However, the artificial brain playing table tennis was only developed two years ago. After two years of evolution, the "artificial brain" hasAnd toAI Stealing jobs...

Just over a month ago, another startup FinalSpark released aBioprocessor, all working inside are biological neurons of human brain organoids.

In order to make the cultivated brain cells more like the human brain, the researchers also introduced dopamine. Usually, they would put it in a "cage", and if the brain cells performed well, they would open the cage to "reward" the brain cells.

Under such guidance, according to official statements, this new biological processor consumes only one millionth of the energy of traditional digital processors.

Not only that, at the end of last year, Indiana University also developed a machine - Brainoware, a brain-like hybrid computing system.Learning capabilities are comparable to AI levels.

Take its voice recognition capability, for example. At first, Brainoware's accuracy in distinguishing eight people's voices was only 51%. But two days later, the accuracy rate rose to 78%.

Such a huge improvement in such a short period of time is all thanks to the brain. The entire system will constantly change and reorganize during the electrical stimulation response process, so as to continuously improve.

However, judging from the current progress, there is still a lot of room for improvement in artificial brains to completely reduce AI energy consumption.a lot ofThey are all still the results of the laboratory。

But what Shichao wants to say is that we can't underestimate its potential. After all, the brains of ordinary people areDeveloped 10%If we fully cultivate a human brain and develop all its potential, the capabilities it would possess would be beyond our imagination.

Interestingly, some time ago, Shichao discovered that Japan has developed a new technology that can fix human skin on the face of a robot.

With such skin and an artificial brain, wouldn't it be possible for us to achieve immortality through cloning in the future, just like Yun Tianming in The Three-Body Problem?

Written by:squirrel

edit: Jiang Jiang, Noodles

Art editor: Huan Yan

Image, source：

NEW ATLAS，'Brain-in-a-jar' biocomputers can now learn to control robots

NEW ATLAS，New technique gives robotic faces living human skin

Synced, the world's first biological computer opens service: 16 human brain organoids, saving a million times in energy consumption

Quantum bits, living brain cells made into 16-core chips, can be programmed with Python, 9 institutions conduct experiments and 36 universities line up