AI can predict how monkeys play Pac-Man

ANARTIFICIAL intelligence can accurately predict how a monkey plays the video game Pac-Man and mimic the animal’s eye movements. TianmingYang at the Chinese Academy of Sciences and his colleagues trained two rhesus monkeys to play Pac-Man by rewarding them with juice for collecting all the dots in a maze and evading capture by ghosts. Getting Pac-Man, the character players control, to eat the ghosts after chomping on a special dot that makes them vulnerable earned the monkeys a reward. As the monkeys played the game35times, their eye movements were tracked. Then the researchers trained a type of brain-likeAI known as a neural network to predict the monkeys’ directional choices at maze junctions. The type of neural network they used, called a transformer network, has a self-attention mechanism that allows it to prioritise the most relevant information it encounters and act on it. Wang wondered if this attention mechanism would echo what was going on in a brain. The AI’s attention mechanism seemed to mimic what the monkeys were looking at, including focusing on where to move next, several moves in the distance – suggesting some element of the AI model was pre-planning its strategy. The AI model was able to predict the monkey’s choice of next move when Pac-Man reached a junction inthemaze87.6per cent of the time (arXiv,doi.org/m57g). “We showed that, indeed, this transformer structure may do something similar to the brain,” says Yang. While it can be risky to anthropomorphise AIs, the research does indicate that they might be able to “think” or “look” in a similar way to mammals, says Noah Giansiracusa at Bentley University in Massachusetts Technology.