This "brain" has 800,000 cells, and you can learn to play table tennis in 5 minutes and beat the AI!

How many brain cells are needed to play video games?

When you hear this sentence, your first reaction must be: this is a brain teaser.

No, there is a real answer to this question. This is all thanks to a neural network system called DishBrain.

If you are playing table tennis, the number of brain cells required is about 800,000.

No, it took 800,000 human brain cells 5 minutes to learn to "play table tennis".

Recently, a research team from Australia put 800,000 living human and mouse brain cells into a culture dish, connected them to electrodes, and played classic arcade games. Pong.

Scientists call it the first sentient "brain on a disk" (DishBrain).

The purpose of this research is to create synthetic biological intelligence (SBI) to provide better methods for future research on neurological diseases.

The latest research was published in the journal Neuron on Wednesday.

Paper address:

https://www.cell.com/neuron/pdfExtended/S0896-6273(22)00806-6

「Brain on a plate」Learn to play games in 5 minutes

We know that the human brain has 860 billion neurons. Synapses transmit electrical signals from one neuron to the next, but they are not thought of as information processors.

But neurons are a magical system that can process information in real time with extremely low energy consumption.

DishBrain consists of a single layer of human neurons grown on top of a microelectrode array that stimulates these brain cells.

So, where do these "human brain cells" come from?

Whether the process of obtaining brain cells complies with research standards may be everyone’s first concern.

DishBrain neural cell array at work

In fact, it is not as you think, extracting neurons and brain cells directly from the human brain. It's so unethical.

Scientists have provided a solution: using human induction.

Let pluripotent stem cells (hiPSC) differentiate into cortical neuronal cells and then culture them. At the same time, the researchers also used mouse cells for cultivation.

The picture below shows the difference between mouse and human cortical cells in a culture dish (50μm).

Among them, blue DAPI represents all stained cells, green NeuN represents neurons, tubulin (BIII) labels axons, and MAP labels dendrites.

It can be seen that mouse cortical cells (A) can grow and maintain for several months in nutrient-rich medium, and form complex morphologies with a large number of dendrites and axonal connections.

After human induced pluripotent stem cells (hiPSCs) are differentiated into a single layer of active heterogeneous cortical neurons, these neurons can also display mature functional properties and can Forms dense connections with glial cells that serve as support.

So, once the cells are cultured, how to make them play table tennis?

Obviously, we need a sophisticated device.

#The circular groove in the center of the device is where brain cells and electrodes are placed.

Cortical Labs researchers used the MaxOne multi-electrode array provided by Maxwell Biosystems, a Swiss company, for experiments.

MaxOne is a high-resolution electrophysiology platform with 26,000 platinum electrodes arranged in an area of ​​8mm*8mm, with a maximum resolution of 220*120.

The system is based on complementary oxide semiconductor (CMOS) technology and can record up to 1024 channel numbers and stimulation from up to 32 units.

How do neurons actively reason to complete the game?

To teach DishBrain to play table tennis, the research team asked this neuron to play a single-player game of table tennis.

Researchers use electrical signals to stimulate neurons on the electrode array and record their activity.

Among them, electrical signals are sent to different array areas to represent the position of the table tennis ball. Microelectrodes on both sides of the plate will indicate whether the ball is on the left or right side of the racket, and the signal frequency reflects the distance of the ball.

The neurons in the upper half of the electrode array are responsible for sensing the position of the table tennis ball, and the neurons in the lower half are divided into left and right blocks and are responsible for outputting the distance the table tennis racket moves up and down.

Then, DishBrain can generate electrical signals to move the racket to catch the ball.

But at first, their performance was very poor.

In order to play the game well, neurons need feedback. So the team developed feedback software that uses electrodes to criticize DishBrain when it misses the ball.

In order to optimize the error, the Cortical Labs team mainly uses the predictive coding formula that minimizes variational free energy, also known as the Kalman filter.

This improved the system when playing table tennis. In just five minutes, DishBrian learned to move the racket back and forth based on the position of the ball.

Hey, it seems that DeepMind’s AI has also played this game? That’s right, in 2013, DeepMind first demonstrated the performance of its artificial intelligence reinforcement learning algorithm through an Atari game.

At present, DishBrain’s performance in playing games is still not as good as DeepMind’s own reinforcement learning algorithm that has been developed for so many years. But it took the AI ​​90 minutes to learn how to play this game, while this layer of brain cells only took 5 minutes to play the game well.

In this way, the use of the computing power of living brain neurons to create synthetic biological intelligence (SBI) is completed.

Interestingly, the researchers said they would also test the effects of alcohol and drugs on DishBrain’s ability to play table tennis in the future.

Cortical Labs Dr. Brett Kagan said,

We are trying to create a dose-response curve with ethanol - basically making these neuronal cells "drunk" to see if they It's even worse to play like people do when they drink.

Can computers imitate the human brain?

For now, DishBrain’s strategy for playing table tennis is slow and one-sided, and winning eSports championships sounds far away, but these studies reflect the potential of merging living tissue with silicon technology. .

This is the first synthetic biological intelligence experiment to demonstrate that neurons adjust their activity to accomplish specific tasks. And they can learn to perform tasks better if they are given feedback.

This research has great potential in disease modeling, drug discovery, understanding how the brain works, how intelligence is generated, and studying how drugs affect brain activity. "We have shown that we can interact with living biological neurons so that they change their activity to produce Something like intelligence."

"This is a new direction in understanding intelligence," Kagan said. "It not only tells us what it means to be human, but also allows us to understand what it means to be 'alive', what it means to be 'smart', what it means to 'process information' and 'to be sentient' in today's ever-changing world. "

Karl Friston, a theoretical neuroscientist at University College London, said: "This result is groundbreaking in that it equips neurons with the ability to sense—feedback—to take action on the world."

A few years ago, Friston proposed a theory called the free energy principle, which proposes that all biological systems behave in ways that narrow the gap between expectations and reality—in other words, the world can change Be more predictable.

Free energy theory

According to Friston’s theory, by adjusting behavior, the world will become more predictable, and DishBrain is biologically proven Got this.

Kagan said, "The DishBrain experiment is essentially creating a more predictable environment."

The DishBrain experiment has brought some exciting possibilities to mankind, especially It's in artificial intelligence and computing.

You must know that the human brain contains approximately 80 to 100 billion neurons, which is much more powerful than any computer. It is difficult for the best computers to replicate the human brain. The closest thing we have so far is a chip with artificial synapses designed by MIT engineers, which allows us to replicate 1% of human brain activity using 82,944 processors, 1 petabyte of main memory, and 40 minutes. seconds.

MIT Artificial Synapse Chip

If this architecture were more like a real brain—perhaps even a synthetic biological system like DishBrain— —Perhaps the goal of computers replicating the human brain will not be so far-fetched.

DishBrain also allows us to understand the impact of various drugs on the brain from the cellular level. One day, it could even be possible to create customized drugs for specific patients, using neurons cultured reversely from a patient's skin stem cells.

“The potential of this is so exciting: it means we no longer have to create ‘digital twins’ to test the effects of treatments,” Fristo said.

Digital twin for customized drugs

"In principle, we now have the ultimate bionic "sandbox" in which drugs can be tested and the effects of genetic variation, this sandbox is made up of the exact same computational (neuronal) elements found in your brain and mine."

Coincidentally, in order to advance neuroscience research, the same today A study in Nature perfectly combined human and mouse brains to create brain-like organs.

In the study, researchers from Stanford University transplanted human brain induced pluripotent stem cells into the developing brains of rats.

As shown in the picture, the bright green part is the brain organoid.

It was found that brain organoids can develop and mature together with the rat brain. At the same time, these organoids will gradually develop blood vessels to provide nutrients for their own development.

Finally, it becomes part of the brain by being partially integrated with the neural circuitry of the brain.

With brain organoids, scientists can manipulate neurons in a petri dish and find the mechanisms behind potential neurological diseases.

Netizen God Comments

"Does this mean that there is some form of consciousness even if there is no "existence"."

Discussion Suddenly it rose to a philosophical level...

"I want to be the first to welcome our new brain cell overlord."

"We need one A bigger petri dish."

"I want a machine body like Ghost in the Shell."

"Neuromancer. The perfect combination of technology and magic."

"Incredible, it reminds me of David Eagleman's TED talk. He thinks The human brain is a primitive I/O device. As a baby, it is learning to process input data, and at any time we can add additional input and the brain will begin to interpret the new data."

"Reminds me of the plot of "Biscuit" in Black Mirror...it's creepy."

"But does Neuron like this game? ?"

"I have read hundreds of comments and you are the first to ask this important question!"

"These cells are no longer If you continue to improve, you will become a Trump supporter within a few days."

"They have exceeded the IQ of ordinary Trump supporters."

"In my opinion, this is slavery. Think about where this technology will be used."

Reference:

https://www. cnet.com/science/live-brain-cells-in-dish-quickly-learn-to-play-classic-game-pong/

https://www.nature.com/articles/d41586- 022-03229-y

https://www.engadget.com/brain-cells-pong-rats-182835843.html

The above is the detailed content of This "brain" has 800,000 cells, and you can learn to play table tennis in 5 minutes and beat the AI!. For more information, please follow other related articles on the PHP Chinese website!