Technologists are still working to perfect A.I. and automation technologies that can accomplish one or two tasks in a very complex world. But it’s not hard to envision a future in which multiple computer agents will work together to solve even bigger problems. Alphabet Inc’s DeepMind team set out to illustrate how learning machines work together.
The team ran thousands of iterations of two simple games, allowing programs to learn the most effective strategies. In one, two A.I.s competed to gather the most resources (green ‘apples’), and could also choose whether or not to sabotage their competition.
As the team detailed in a blog post, computer agents in the apple-gathering scenario tended to coexist peacefully when there were plenty of resources. But when apples became scarce, they quickly turned to sniping one another to get an edge.