Grand Theft Auto 5 is a complex game that simulates a vivid city and a vast street network. You might use it to chase and race other players (to put it mildly), but others use it to teach AI algorithms about the real world.
There are lots of things you can do in GTA 5 and GTA Online, including a whole range of criminal activities. But according to MIT Technology Review, several research groups are using the game’s open world to teach and train computers how to better navigate self-driving cars, instead of playing it and spending money at the strip club.
By using a technique known as machine learning, they’re enabling computers to quickly comprehend and process the world around them, like identifying faces and recognizing speech. This approach can be done in the real world as well, but it requires large quantities of data to be collected; this can be challenging and could take thousands of hours of processing, and here’s where GTA 5 comes in.
Apparently, the game’s scenery is so realistic that it can be used to generate data that’s as good (or sometimes better) than using real-world imagery, and a team of researchers from Intel Labs and Darmstadt University in Germany done just that.
The researchers have developed a way to extract useful training data from GTA, by creating a software layer that automatically classifies different objects in the in-game captured scenes.
The software comes with its own labels, “teaching” self-driving cars the difference between a pole and a pedestrian, for example.
“With artificial environments we can effortlessly gather precisely annotated data at a larger scale with a considerable amount of variation in lighting and climate settings. We showed that this synthetic data is almost as good, or sometimes even better, than using real data for training”, said Alizera Shafaei, a PhD student at the University of British Columbia.
Let’s just hope the real-world AI won’t control a car like GTA’s NPCs do, because frankly, they’re kind of clumsy.