Science

Human brain cells in the laboratory learn to play video games faster than artificial intelligence

⇧ [VIDÉO] You may also like this partner’s content (after ad)

Human brain cells in a Petri dish or artificial intelligence, which of the two learns faster? In an unprecedented experiment, the researchers compared the learning capacity of artificial intelligence with… brain cells placed in a kind of electronic Petri dish. Surprisingly, when the two entities were “asked” to learn to play the famous video game Pong, they were surprised to find that it was the brain cells, connected to an electronic device, that learned to play the game faster, not the AI. – as anyone could expect.

Machine learning has been a branch of artificial intelligence that has flourished in recent years, with demonstrations and real-world applications more impressive than the others. As a result, for ordinary people, advanced AI today has learning capabilities far superior to simple primitive brain cells kept in a laboratory environment … But this new experience shows that this is not really the case. Case …

Researchers now call their creations “cyborg brains” and the reasons seem obvious. Of course, this is not the first time that researchers have studied the organoids (primitive mini-organs grown in the laboratory) of human brains, but according to Brett Kagan, lead author of the study and scientific director of Cortical Labs, it is the first. time we see that mini-brains are capable of performing tasks directed towards a specific goal.

DishBrains Vs IA

Each of the mini brains created by Kagan and his team has around 800,000 to 1 million living brain cells. To give a simple description: this roughly corresponds to a cockroach brain. Some brains were made from mouse cells taken from embryos, while others were made from human brain cells derived from stem cells.

To allow the organoids to interact with a real environment, it was of course necessary to go through an electronic device. To do this, the researchers grew the cells in microelectrode arrays that can both stimulate the cells and read their activity. The resulting system was named “DishBrain”.

To simulate a simplified version (without an opponent) of the game Pong (a ping pong video game), the scientists had a simple idea: activating the electrodes to the left or right of a net tells the mini-brain if the bullet is to your left or right, and the frequency of the signal indicates proximity.

Pong game preview in normal mode (with opponent). The player on the left takes 4 to 2.

Specific patterns of activity between neurons are interpreted as a shift of the racket to the left or right. The computer responds to this activity and the feedback through the electrodes allows the mini-brains to learn to control the racket. “When they are connected to the game, they believe that they themselves are the racket”, sums up one of the researchers.

Fast but limited learning compared to AI

If human mini-brains have been shown to be faster at learning to play the game of Pong than AI, they are still much more limited in the long run. So advanced AI always ends up getting a lot better in games than DishBrain. But the learning speed of the organoids is still very impressive: While the AIs tested in the experiment required 5,000 round trips to master the game, the mini-brains only needed 10 to 15 iterations. This corresponded to about 5 minutes of learning.

Brains made from human cells are much better at Pong than those made from mouse cells. But because the source of the cells is different, the team still can’t be sure that this is solely due to their human nature. In the future, it would be interesting to pit the two types of mini-brains against each other, or human organoids against AI (at least in its early stages of learning, before it starts to work). It becomes unbeatable…).

The feat, however, was not done only on the side of the mini-brains, but also on the researchers. In fact, as other researchers commenting on the study point out, the authors had brilliant success in making a neural network make sense of digital data and act in that environment at the same time. This closure of the action-perception cycle is therefore not only an exceptional technical success, but also brings us closer to the creation of real synthetic brains. In other words, “cyborg brains”.

Kagan and colleagues’ approach to training is based on a theory of how the brain works called the “free energy principle,” developed by Karl Friston of University College London. The basic idea is that even neurons in a petri dish try to create an internal model of their outside world. In other words, they are trying to predict what will happen based on the information they receive and they do not like to be surprised.

That is why cells “play the game.” When they play, their inputs become more predictable. If they don’t play, they get random entries that are aversive. What’s really amazing about this setup is the sensitive behavior that comes up without supervision.

Cortical Labs’ long-term goal is to develop cyborg brains that Kagan says can be smarter than computer systems. There are also more immediate applications. For example, studying how neurons learn so quickly and efficiently could help improve machine learning, reducing the large amounts of energy required and learning time. According to the researchers, the next generation of artificial intelligences should target the functional and thermodynamic efficiency of biological brains. The current work is a remarkable, and perhaps even historic, step in this direction.

Another potential use for these types of organoids is drug testing. Administering experimental drugs to mini-brains while they play, Kagan said, could reveal more about the effects of those drugs on the human brain than studying neurons in isolation.

bioRxiv

PHP Script, Elementor Pro Weadown, WordPress Theme, Fs Poster Plugin Nulled, Newspaper – News & WooCommerce WordPress Theme, Wordfence Premium Nulled, Dokan Pro Nulled, Plugins, Elementor Pro Weadown, Astra Pro Nulled, Premium Addons for Elementor, Yoast Nulled, Flatsome Nulled, Woocommerce Custom Product Ad, Wpml Nulled,Woodmart Theme Nulled, PW WooCommerce Gift Cards Pro Nulled, Avada 7.4 Nulled, Newspaper 11.2, Jannah Nulled, Jnews 8.1.0 Nulled, WP Reset Pro, Woodmart Theme Nulled, Business Consulting Nulled, Rank Math Seo Pro Weadown, Slider Revolution Nulled, Consulting 6.1.4 Nulled, WeaPlay, Nulledfire

Back to top button