(CN) — Researchers in Australia show that gaming is for everyone, even a layer of brain cells hooked up to a computer chip. In a paper published Wednesday in the journal Neuron, researchers present the first synthetic biological intelligence able to adapt its behavior on a real time basis. Dubbed BrainDish, the system of brain cells on a multielectrode array was able to play one of the world’s earliest video games: Pong.
“We often think about neurons in a medical sense, or physiological sense; the exciting thing now is can we consider them for what they are, which are tiny information processors that are hugely power efficient. So, can we use them as a biomaterial to create something that we weren’t able to create with our existing technologies?” said the paper’s lead author, Brett Kagan.
Helmed by Cortical Labs, a biotechnology startup based in Melbourne, the study tests the DishBrain system’s ability of its neurons to organize and respond to electrical stimulus in real time to successfully play Pong, learning eventually how to correctly move the game paddle to hit the virtual ball.
It is the first demonstration of goal-directed learning from neurons and provides a basis for future research into how drugs and diseases affect neurons, as well as future development of machines integrated with biomaterial that can do work that may be dangerous or hard to access for humans.
The study is based on the concept of the free energy principle, a theory developed by British neuroscientist Karl Friston, who is cited as a senior author of the study. The free energy principle attempts to describe the ability of the brain to function and adapt.
“What it suggests is that neurons want to be able to build a model of the external world, and they want to be able to predict it,” explained Kagan, who is Cortical Labs’ chief scientific officer. “You don’t really see the world, you see a representation of the world; when we try to act upon the world, we’re not really acting on the world. Our brain creates a model of the world based on the information we receive and creates a model that we act upon that.”
A panel of electrodes recorded spikes in activity as neurons from mouse and human cells learned how to play the game based on stimulation feedback from either successful hits or misses. Failure to block the ball would result in a critique from the Cortical Labs’ SpikeStream program, from which the neurons would adjust its playstyle to minimize that stimulus. Simulations with no stimulation to the neurons at all did not demonstrate any learning at all.
Testing showed that not only were the cells able to learn in real time how to better play the game, but it was also able to self-organize in specific conditions, indicating, according to the study, the true adaptability of neurons to respond to varied stimuli.
The use of Pong was, according to Kagan, partially inspired by a nostalgia for the game.
“It’s in real time, it’s continuous, there’s a very clear success and failure condition, it was simple and it worked with the constraints we had; amongst those, it was also a classic game that people benchmark a lot of machine learning with,” he said.
Although the study was conducted with mouse brain cells, researchers also utilized an ethical method to generate stem cells from human skin or blood cells, known as induced pluripotent stem cell (iPSC). The neurons generated from this method were utilized the same way as animal neurons. Researchers even noticed that the human brain cells were able to play the game slightly better than the mouse cells.
The study is also broadly applicable to the field of neuroscience as a whole.
“This is going to answer questions that are fundamental to how we think about ourselves as humans, because yes, you can use neurons as a material, but also how we became the intelligent thinking creatures that we are,” Kagan said. “As we uncover how these neurons work for their biomaterial properties, we also learn what it means for us in the human sense.”
Read the Top 8
Sign up for the Top 8, a roundup of the day's top stories delivered directly to your inbox Monday through Friday.