A dish of neurons may have learned to play Pong (wrong)

Enlarge / In culture, nerve cells spontaneously form the structures needed to communicate with each other.

One of the most exciting developments in AI has been the development of algorithms that can learn the rules of a system on their own. Early versions of things like game algorithms needed to be given the basics of a game. But newer versions don’t need that – they just need a system that keeps track of certain rewards like score, and they can determine which actions maximize this without needing a formal description of the rules of the game.

An article published by the journal Neuron goes even further by using real neurons grown in a box filled with electrodes. This added an extra level of complication, as there was no way of knowing what the neurons would actually find rewarding. The fact that the system appears to have worked may tell us something about how neurons can self-organize their responses to the outside world.

Say hello to DishBrain

The researchers behind this new work, who were primarily based in Melbourne, Australia, call their system DishBrain. And it’s based on, yes, a dish with a set of electrodes on the bottom of the dish. When neurons are grown in the dish, these electrodes can do two things: detect the activity of neurons above them or stimulate these electrodes. Electrodes are large relative to the size of neurons, so sensing and stimulation (which can be thought of as similar to reading and writing information) involves a small population of neurons, rather than a only.

Beyond that, it’s a standard culture dish, meaning a variety of cell types can be cultured in it. For some control experiments, the researchers used cells that do not respond to electrical signals. For these experiments, the researchers tested two types of neurons: some dissected from mouse embryos, and others produced by inducing human stem cells to form neurons. In both cases, as seen in other experiments, the neurons spontaneously formed connections between themselves, creating networks that had spontaneous activity.

Although the hardware is completely flexible, the researchers configured it as part of a closed-loop system with a computer controller. In this configuration, the electrodes in some regions of the dish have been defined as taking input from the DishBrain; they are collectively called the motor region since they control the response of the system.

Eight other regions have been designated to receive input in the form of stimulation from the electrodes, which act much like a sensory area of ​​the brain. The computer could also use these electrodes to provide feedback to the system, which we’ll cover below.

Collectively, these provide everything a neural network needs to learn what is happening in the computing environment. Motor electrodes allow neurons to modify the behavior of the environment, and sensory neurons receive both information about the state of the environment as well as a signal about whether its actions were successful. The system is generic enough that all sorts of environments can be set up in the computing part of the experience – pretty much anything that simple inputs change the environment.

The researchers chose pong.

Leave a Comment

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