800,000 neurons in a dish learned to play pong in just five minutes

800,000 neurons in a dish learned to play pong in just five minutes

Scientists have just taught hundreds of thousands of neurons in a dish to play Pong. Using a series of strategically timed and placed electrical zaps, neurons not only learned the game in a virtual environment, but performed better over time – with longer rallies and fewer misses – showing a level of adaptation previously considered impossible.

Why? Imagine literally taking a piece of brain tissue, digesting it down to individual neurons and other brain cells, dropping them (gently) onto a plate, and now being able to teach them, outside of a living host, to respond and to adapt to a new task using only electric zaps.

It’s not just fun and games. The biological neural network joins its artificial cousin, DeepMind’s deep learning algorithms, in a growing pantheon of attempts to deconstruct, reconstruct, and one day master a kind of brain-based general “intelligence.” human.

The brainchild of the Australian company Cortical Labs, the entire installation, dubbed DishBrainis the “first real-time synthetic biological intelligence platform,” according to the authors of an article published this month in neuron. The configuration, smaller than a dessert plate, is extremely elegant. It connects isolated neurons with microchips that can both record the cells’ electrical activity and trigger precise zaps to alter those activities. Similar to brain-machine interfaces, the chips are controlled by sophisticated computer programs without any human intervention.

The chips act as a bridge allowing neurons to connect to a virtual world. As a translator of neural activity, they can unite biological electrical data with bits of silicon, allowing neurons to respond to a digital game world.

DishBrain is set to expand to other games and tests. Because neurons can sense and adapt to the environment and transmit their results to a computer, they could be used in drug testing. They could also help neuroscientists better decipher how the brain organizes its activity and learns, and inspire new methods of machine learning.

But the ultimate goal, explained Dr. Brett Kagan, scientific director of Cortical Labs, is to help harness the inherent intelligence of living neurons for their superior computing power and low power consumption. In other words, compared to neuromorphic hardware that mimics neural computation, why not just use the real thing?

“Theoretically, the generalized SBI [synthetic biological intelligence] may arrive before artificial general intelligence (AGI) due to the inherent efficiency and evolutionary advantage of biological systems,” the authors wrote in their paper.

To encounter DishBrain

The DishBrain project started from a simple idea: neurons are incredibly intelligent and adaptable computing machines. Recent studies suggest that each neuron is a supercomputer unto itself, with branches once considered passive acting as independent mini-computers. Like people within a community, neurons also have an inherent ability to connect to various neural networks, which change dynamically with their environment.

This level of low-energy parallel computing has long inspired neuromorphic chips and machine learning algorithms to mimic the brain’s natural abilities. Although both have made progress, neither has been able to recreate the complexity of a biological neural network.

“From worms to flies to humans, neurons are the starting point of generalized intelligence. So the question was: can we interact with neurons in ways that tap into this inherent intelligence? Kagan said.

Walk in DishBrain. Despite the name, the plated neurons and other brain cells come from an actual brain with consciousness. As for “intelligence”, the authors define it as the ability to gather information, put the data together and adjust the triggering activity, i.e. the way neurons process the data, in a way that helps fit a purpose; for example, quickly learning to place your hand on the handle of a very hot frying pan without burning it on the edge.

The setup begins, true to its name, with a dish. The bottom of each is covered with a computer chip, HD-MEA, which can record from stimulated electrical signals. Cells, either isolated from the cortex of mouse embryos, or derived from human cells, are then deposited on top. The dish is bathed in a nutrient liquid for the neurons to grow and develop. As they mature, they grow from jiggly drops into spindly shapes with vast networks of winding, intertwining branches.

Within two weeks, the mice’s neurons self-organized into networks inside their tiny homes, brimming with spontaneous activity. Human-derived neurons – skin cells or other brain cells – took a bit longer, establishing networks in about a month or two.

Then came the training. Each chip was controlled by commercially available software, linking it to a computer interface. Using the system to stimulate neurons is like providing sensory input, like that coming from your eyes when you focus on a moving ball. Recording the activity of neurons is the result, i.e. how they would react (if they were inside a body) when you move your hand to hit the ball. DishBrain was designed so that the two parties integrate in real time: like humans playing Pong, neurons could in theory learn from past failures and adapt their behavior to hit the virtual “ball”.

Player loan DishBrain

Here’s how Pong is doing. A ball quickly bounces around the screen and the player can slide a small vertical paddle, which looks like a bold line, up and down. Here, the “ball” is represented by electric zaps depending on its location on the screen. This essentially translates visual information into electrical data for the biological neural network to process.

The authors then defined separate regions of the chip for “feel” and “movement”. A region, for example, captures incoming data from the motion of the virtual ball. A part of the “motor region” then controls the rise of the virtual pallet, while another makes it descend. These assignments were arbitrary, the authors explained, meaning the neurons inside had to adjust their firing to excel in a game.

So how do they learn? If the neurons were “hitting” the ball, i.e. showing the corresponding type of electrical activity, then the team zapped them there with the same frequency each time. It’s a bit like establishing a “habit” for neurons. If they missed the ball, they were then zapped by an electrical noise that disrupted the neural network.

The strategy is based on a learning theory called the free energy principle, Kagan explained. Basically, it assumes that neurons have “beliefs” about their environment, and adjust and repeat their electrical activity so that they can better predict the environment, either by modifying their “beliefs” or their behavior.

The theory hit home. In just five minutes, human and mouse neurons quickly improved their gameplay, including better rallies, fewer aces – where the racket failed to intercept the ball without a single hit – and long games with more than three consecutive hits. Surprisingly, mouse neurons learned faster, although they were eventually outmatched by human neurons.

The stimulations were essential for their learning. Separate experiences with DishBrain without any electrical feedback performed much worse.

Game on

The study is proof of concept that neurons in a dish can be a sophisticated learning machine, and even show signs of sensitivity and intelligence, Kagan said. This does not mean that they are conscious, but rather that they have the ability to adapt to a purpose when “embodied” in a virtual environment.

Cortical Labs isn’t the first to test the limits of single neuron data processing power. In 2008, Dr. Steve Potter of the Georgia Institute of Technology and his team found that with just a few dozen electrodes, they could stimulate rat neurons to show signs of learning in a dish.

DishBrain has a head start with thousands of compacted electrodes in each configuration, and the company hopes to harness its biological power to aid in drug development. The system, or its future derivatives, could potentially serve as a microbrain surrogate for testing neurological drugs or for better understanding the neurocomputational powers of different brain species or regions.

But the long-term vision is a “living” bio-silicon computing hybrid. “Integrating neurons into digital systems may enable performance not achievable with silicon alone,” the authors wrote. Kagan envisions developing “biological processing units” that weave together the best of both worlds for more efficient computing and, in doing so, illuminate the inner workings of our own minds.

“This is the start of a new frontier in understanding intelligence,” Kagan said. “It touches on fundamental aspects of not only what it means to be human, but also what it means to be alive and intelligent, to process information and to be sentient in an ever-changing dynamic world.”

Image Credit: Cortical Laboratories

#neurons #dish #learned #play #pong #minutes

Leave a Comment

Your email address will not be published. Required fields are marked *