When speeding through a French town in the Gran Turismo video game you may see a Corvette behind you trying to catch your slipstream.
The technique of using the slipstream of an opponent’s racing car to speed up and overtake them is a favorite of hardcore gamers of the realistic PlayStation racing game.
But who pilots the Corvette is not a human being: It is GT Sophy, a powerful Artificial Intelligence (AI) agent created by Sony, the manufacturer of PlayStation.
Gran Turismo players have been racing against computer-generated race cars since the franchise launched in the 1990s, but the new AI driver that launched last week in Gran Turismo 7 is smarter and faster because it has been trained with the most current AI methods.
“Gran Turismo had built-in AI that has been around since the game’s creation, but it has a very narrow performance band and it’s not very good,” explains Michael Spranger, Sony AI COO. “It is very predictable. Once you get past a certain level, it stops appealing to you.”
But now, he adds, “this AI is going to put up a fight.”
Visiting an AI lab at universities and companies like Sony, Google, Meta, Microsoft, and ChatGPT maker OpenAI, it’s not unusual to find AI operatives like Sophy racing cars, throwing angry birds at pigs, holding battles. epic interstellar battles or helping human players build new Minecraft worlds – all part of the job description for computer systems trying to learn how to be smarter at games.
But in some cases, they’re also trying to learn how to be smarter in the real world. In a paper published in January, a Cambridge University researcher who created an AI agent to control Pokémon characters argued that it could “inspire all sorts of uses that require managing teams under conditions of extreme uncertainty, including managing a team of doctors, robots, or employees in a constantly changing environment, such as a region hit by a pandemic or a war zone.”
And while that may sound like a child advocating playing three more hours of Pokémon Violet, video game study has been used to advance AI research — and train computers to solve complex problems — since the middle of the century. xx.
At first, the AI was used in games like checkers and chess to test the ability to win strategy games. Now a new branch of research focuses more on performing open-ended tasks in complex worlds and interacting with humans, not just for the purpose of defeating them.
“Reality is like a super-complicated game,” says Nicholas Sarantinos, author of the Pokémon paper and who recently turned down a PhD offer at Oxford University to start an AI company to help corporate workplaces Build more collaborative teams.
In the web-based battle simulator Pokémon Showdown, Sarantinos developed an algorithm to analyze a team of six Pokémon, predicting how they would perform based on all the possible battle scenarios that await them and their comparative strengths and weaknesses.
Microsoft, which owns the popular Minecraft game franchise and the Xbox gaming system, has assigned artificial intelligence agents a variety of activities: from walking away from lava to cutting down trees and building furnaces. The researchers hope that some of their learnings may eventually play a role in real-world technology, such as how to make a home robot do certain tasks without having to be programmed to do them.
While it “goes without saying” that real humans behave quite differently from fictional creatures in video games, “the core ideas can be used,” Sarantinos says. “If you use psychology tests, you can take that information to conclude how well you can work together.”
Amy Hoover, an assistant professor of computer science at the New Jersey Institute of Technology who created algorithms for the digital card game Hearthstone, says that “there really is a reason to study games,” but it’s not always easy to explain.
“People don’t always understand that the reason is the optimization method and not the game,” he adds.
Video games also provide a useful test bed for AI — even for some real-world applications in robotics or healthcare — that are safer to test in a virtual world, explains Vanessa Volz, an AI researcher at Danish startup Modl.ai. , which creates AI systems for game development. However, she adds, “it can be overrated.”
“Probably, it won’t be a breakthrough and everything will be transferred to the real world,” adds Volz.
Japanese electronics giant Sony launched its own AI research division in 2020 with entertainment in mind, but it has drawn broader academic attention. His research paper that he presented to Sophy last year made the cover of the prestigious scientific journal Nature, which claimed it could have effects on other applications, such as drones and autonomous vehicles.
The technology behind Sophy is based on an algorithmic method known as reinforcement learning, which empowers the system by rewarding it when it does something right while doing virtual races thousands of times.
“The reward will tell you ‘You’re making progress. This is good’ or ‘You went off track. That’s not good,’” Spranger details.
The best Gran Turismo players in the world continue to beat Sophy in tournaments, but average players will find it difficult to win and can adjust the difficulty settings depending on how much they want to be challenged.
PlayStation players will only be able to attempt to compete against Sophy until March 31 on a limited number of tracks, so she can get some feedback and get back into testing. Peter Wurman, director of Sony AI America and leader of the GT Sophy project, says that it takes about two weeks for AI agents to train on 20 PlayStation consoles.
“For it to spread throughout the entire game, it takes more progress and some additional time before we’re ready for that,” he says.
And to take it to real streets or to the tracks of Formula One? That could take much longer.
Self-driving car companies adopt similar machine learning techniques, but “they don’t relinquish full control of the car the way we can,” Wurman says. “In a simulated world, no one faces the risk of losing their life. You know exactly the kind of things you’re going to see in the environment. There are no people crossing the street or anything like that.”