Scientists have taught 800,000 living brain cells in a bowl how to play the legendary arcade game Pong.
This is the work of a team of neuroscientists and programmers from Cortical Labs, Monash University, RMIT University, University College London and the Canadian Institute for Advanced Research. Their detailed findings were published in the journal earlier this month neuron.
Of course, the actual setup is more complex than just sticking a clump of neurons in a Petri dish. In this system, called DishBrain, the nerve cells are placed on a multi-electrode array, which is like a kind of CMOS chip that can read very small changes in the neuron’s electrical activity.
Nerve cells have well-known action potentials – they fire in response to a specific sequence of voltage changes across the cell membrane. This makes them behave almost like gates in a computer circuit.
The cells connect, integrate into the chip and can survive for many months. The array of electrodes allows the researchers to send and read signals from the nerve cells at specific locations on the grid at a specific rate. So electrodes on the array could fire to one side or the other to tell DishBrain where the ball is, and the frequency of the signals could indicate how far the ball was from the racquet. By lighting up a certain pre-programmed array of electrodes, DishBrain could trigger motor activities such as B. moving the paddle up and down in the game.
“We can decode information going out and code incoming information just by using these very small electrical signals, and thus represent what’s happening to the cells,” says Brett Kagan, chief scientist of biotech start-up Cortical Labs and lead author of the neuron Paper. “Video games help people understand what is going on. If we just did it as a function of random numbers, people wouldn’t appreciate or understand the meaning of the results.”
But why did you choose Pong?
“From a scientific point of view, we needed a task that was real-time, continuous, and had a really discrete loss condition (the gain condition wasn’t that important) that was pretty easy to design and code into the cells,” he says, Kagan.
It’s also a game that has been a staple for the computational neuroscience community. For example, Google’s DeepMind used “Pong” in 2013 to train its machine learning algorithms.
“If you think about it, there are really about six rules for how this environment works. This is what we call a structured information landscape,” says Hon Weng Chong, Chief Executive Officer of Cortical Labs. “The conclusion we have is that these neurons must try to create a model internally [influenced by these six rules]. Whatever it is, we don’t know for sure yet, and this is up for further study. And it’s trying to use that to optimize the parameters we set, which are don’t miss the ball, hit the ball.”
[Related: How an AI managed to confuse humans in an imitation game]
In addition to the question of how exactly perceived “intelligence” arises from DishBrain, the team also wants to put its performance to the test with an artificial neural network. They also want to see how well DishBrain plays the game while under the influence of drugs and alcohol.
“We want to show that there is a dose-response curve for their ability to play the game to confirm that these neurons can be used in actual drug testing and discovery, as well as for personalized medicine,” says Chong.
When Cortical Labs came out of stealth in March 2020, its goal was to build biological computer chips.
Brain cells, notes Kagan, are an interesting biomaterial system that can efficiently process information in real time without requiring mountains of input samples. “A fly, a very simple system, has more general intelligence in terms of navigating its environment than the best of machine learning,” he says. “This can be done with a fraction of the power consumption. Why imitate what can be used?”
But while using these types of chips for applications in computer science research is interesting, Cortical Labs has a more immediate focus on how it plans to commercialize its technology platform.
[Related: Growing Micro-Brains From Skin Cells Sheds Light On Autism]
“I think the primary commercial aspect for us is to help researchers in very difficult areas like dementia research, epilepsy and even depression to use the technology we’ve developed to search for new therapies and new drugs,” says Chong . “That’s kind of a commercialization perspective that we’re trying to monitor at the company.”
Since the nerve cells used in DishBrain can be obtained from pluripotent human stem cells, this opens up possibilities for personalized medicine. “You can take samples from donors, grow genotypically similar neurons, which we can then use to test drugs that will hopefully have the same parameters as donor cells,” says Chong. This could potentially shorten the process of trying different treatments for conditions like epilepsy. “If you had a system that would let you know instantly which drug to take for the best outcome with the fewest side effects, it would be a huge change in the lives of many people with this disease.”