Main menu


Neurons in a dish learned to play table tennis in virtual reality

featured image

Scientists taught hundreds of thousands of neurons in a dish to play ponBy using a series of strategically timed electrical zaps, the neurons not only learned the game in the virtual environment, but over time they rallied longer, made fewer mistakes, and performed better than before. showed levels of adaptation that were thought impossible.

why? It literally takes a chunk of brain tissue, digests it down to individual neurons and other brain cells, dumps them (gently) onto a plate, and responds and adapts to new tasks outside of a living host. Imagine being able to teach Uses electric zap only.

It’s not just fun and games. Biological neural networks are a growing pantheon of attempts to deconstruct, reconstruct, and one day master a kind of general “intelligence” based on the human brain, the extension of its artificial cousin DeepMind. Participate in deep learning algorithms.

The brainchild of Australian company Cortical Labs, the entire setup was dubbed dish brainis the “first real-time synthetic biological intelligence platform,” according to the authors of a paper published this month. neuronA setup that’s smaller than a dessert plate is very sleek. Chips that can record the electrical activity of cells and trigger precise zaps to alter their activity are attached to isolated neurons. Similar to brain-machine interfaces, the chip is controlled by a sophisticated computer program and does not require human input.

The chip acts as a bridge for neurons to link to the virtual world. As a translator of neural activity, it integrates biological electrical data with silicon bits to enable neurons to respond to the world of digital games.

dish brain It is set to extend to further games and tests. Neurons can sense and adapt to their environment and output the results to a computer, which could be used as part of a drug screening test. It could also help neuroscientists better decipher how the brain organizes and learns activity, inspiring new machine learning methods.

But according to Dr. Brett Kagan, Chief Scientific Officer at Cortical Labs, the ultimate goal is to harness the inherent intelligence of living neurons for superior computational power and low energy consumption. So there is no reason not to use the real thing as opposed to neuromorphic hardware that mimics neural computation.

“Theoretically, a generalized SBI [synthetic biological intelligence] The inherent efficiencies and evolutionary advantages of biological systems may predate artificial general intelligence (AGI),” the authors write in their paper.

meet dish brain

of dish brain This project started with a simple idea. Neurons are incredibly intelligent and adaptable computing machines. Recent research suggests that each neuron is a supercomputer in its own right, with branches that were once thought to be passive functioning as independent minicomputers. Just like people in a community, neurons have the inherent ability to connect to diverse neural networks, which dynamically change according to their environment.

This level of multi-parallel low-energy computation has long been the inspiration for neuromorphic chips and machine learning algorithms to mimic the brain’s inherently powerful capabilities. Both have made progress, but none have been able to replicate the complexity of biological neural networks.

“From worms to flies to humans, neurons are the starting point for generalized intelligence. rice field.

come in dish brainDespite its name, the plated neurons and other brain cells are from the actual brain with consciousness. Regarding “intelligence,” the authors refer to it as the ability to gather information, collate data, and coordinate firing activity—that is, the way neurons process data in ways that help them adapt toward their goals. defined as ability. For example, you quickly learn to put your hand on the handle of a hot pan without burning the rim.

As the name suggests, the setup starts with cooking. Each bottom is covered with a computer chip HD-MEA that can record from stimulated electrical signals. Cells isolated from the cortex of mouse embryos or derived from human cells are then placed on top. The dishes are immersed in a nutritious liquid for the neurons to grow and thrive. grow in shape.

Within two weeks, mouse neurons self-assembled into networks in tiny houses and exploded with spontaneous activity. Human-derived neurons (skin cells and other brain cells) took a little longer, establishing networks in about a month or two.

Then came training. Each chip was controlled by commercially available software and linked to a computer interface. Using it to stimulate neurons is similar to providing sensory data, such as you get from your eyes when focused on a moving ball. is how neurons (in the body) respond to dish brain The two parts were designed to integrate in real time. Similar to how humans play table tennis, neurons could theoretically learn from past mistakes and adapt their behavior to hit a virtual ‘ball’.

ready player dish brain

Here’s how the pong works. The ball bounces quickly across the screen. Players can slide a small vertical paddle (looks like a thick line) up and down. Here the “ball” is represented by an electrical zap based on its position on the screen. It essentially converts visual information into electrical data for processing by biological neural networks.

The authors then defined different regions of the chip for ‘feeling’ and ‘moving’. For example, one area gets incoming data from the motion of a virtual ball. Then one part of the “motion area” moves the virtual paddle up and another part moves it down. These assignments were arbitrary, the authors explained, meaning that neurons had to coordinate their firing to gain the upper hand in the match.

So how do they learn? If a neuron “hit” the ball, i.e. showed the corresponding type of electrical activity, the team zapped them at that location with the same frequency each time. did. This is like establishing a “habit” for a neuron. If they missed the ball, they were zapped with electrical noise that confounded the neural network.

Kagan explains that the strategy is based on a learning theory called the free energy principle. Basically, neurons are thought to hold ‘beliefs’ about their surroundings and coordinate and repeat their electrical activity to better anticipate their environment and change their ‘beliefs’ or behaviors.

The theory panned out. In just 5 minutes, both human and mouse neurons rapidly improved gameplay. For example, improved rallies, fewer ace’s where Paddle couldn’t intercept the ball without his one hit, and long gameplay with three or more hits in a row. Surprisingly, mouse neurons learned faster, but ultimately outperformed human neurons.

Stimulation was important to their learning.another experiment with dish brain Without electrical feedback, performance would be much worse.

Game Start

Kagan said the study is a proof-of-concept that neurons in a dish are sophisticated learning machines that can even show signs of sentience and intelligence. It is not conscious. Rather, it has the ability to adapt to goals when ’embodied’ in a virtual environment.

Cortical Labs isn’t the first to test the boundaries of the data processing capabilities of isolated neurons. In 2008, Dr. Steve Potter and his team at Georgia Tech found that just a few dozen electrodes could stimulate rat neurons to show signs of learning in a dish.

dish brain dominates by compressing thousands of electrodes in each setup, and the company hopes to harness its biological power to aid drug development. may serve as microbrain surrogates for testing neuropharmaceuticals and gaining insight into the neurocomputing capabilities of different species and brain regions.

But the long-term vision is for a “living” bio-silicon computer hybrid. “The integration of neurons into digital systems has the potential to deliver performance unattainable with silicon alone,” write the authors. Kagan envisions developing a “biological processing unit” that combines the best of both worlds for more efficient computation.

“This is the beginning of a new frontier in understanding intelligence,” Cagan said. “It touches not only on the fundamental aspects of being human, but also on the fundamental aspects of being alive and intelligent, processing information and having senses in an ever-changing and dynamic world. .”