To see the divide between the most suitable artificial intelligence and the mental abilities of a seven-year-old child, look no further than the famous video game Minecraft. A young human can learn how to find a rare diamond in the entertainment after watching a 10-minute presentation on YouTube. Artificial intelligence (AI) is nowhere close. But in a different computing competition ending this month, researchers hope to narrow the gap between machine and child — and in doing so, assist in reducing the computing power needed to teach AIs.
Competitors may take up to four days and use no more than eight million steps to train their AIs to find a diamond. That’s still a lot longer than it would take a child to learn, but much faster than typical AI models nowadays.
The contest is designed to encourage advances in an approach called imitation learning. This compares with a popular technique known as reinforcement training, in which programs try thousands or millions of random movements in a trial-and-error fashion to home in on the best process. Support learning has helped generate suggestions for Netflix users, created ways to train robotic arms in factories, and even surpassed humans in gaming. But it can require a lot of experience and computing power. Efforts to use reinforcement learning to create algorithms that can safely drive a car or win complicated games such as ‘Go’ have involved hundreds or thousands of processors working in parallel to collectively operate hundreds of years’ worth of simulations — something simply the most deep-pocketed governments and corporations can produce.
Imitation learning can develop the efficiency of the learning process by impersonating how humans or even other AI algorithms tackle the task. And the coding event, recognized as the MineRL (pronounced ‘mineral’) Competition, inspires contestants to use this technique to train AI to play the game.