Scientists are once again asking if they 'could' and not if they 'should': a team of researchers at «biological computing startup» Cortical Labs has taught lab-grown brain cells to play Pong. By spreading around 800,000 neurons across a silicon chip and firing electrical signals at them, the cells were able to demonstrate «apparent learning within five minutes,» according to the scientists involved(opens in new tab). This is how Deus Ex got started, you know.
The scientists call their silicon/neuron conglomeration DishBrain. And to be clear, the process was a bit more abstract than just sitting a Petri dish down in front of a copy of Video Olympics for the Atari 2600. DishBrain wasn't responding to visual inputs like you or I would, but instead to a series of alternating electrical signals that simulated a round of Pong, which the researchers then converted into a visual representation of the game.
It worked like this: whenever DishBrain hit the ball, it received a predictable electrical response. Whenever it missed, the signal spiked randomly. The cells weren't conscious, and didn't 'know' they were playing a game. Rather, by naturally adapting in such a way that it received predictable responses more often, DishBrain adapted to—or, more poetically, learned—Pong. Most exciting for the researchers: it learned quickly, on-the-fly, and with a very low power requirement.
Also, DishBrain was apparently pretty bad at it, so the Pong esports community can rest easy. Still, its success rate was noticeably better than random chance, which makes it more skilled than me. The cells really were adapting to get better at Pong. It might seem unremarkable compared to the abilities that other AI have displayed in games like chess,
Read more on pcgamer.com