Forget about an asteroid plowing into our planet or a plague wiping out the human species. Nope, none of that will be our demise. Nature doesn't need to intervene, as it did with dinosaurs, because it seems we're intent on passing the torch to machines, making ourselves obsolete in the process. Oh, it starts innocently enough—putting automated contraptions on assembly lines and in warehouses, and creating artificially intelligent applications. But do we have to bully our increasingly capable creations, and then teach them to play games like Doom? Apparently so.
Let's talk about the latter for a moment. Guillaume Lample and Devendra Singh Chaplot, a couple of programmers at Carnegie Melon, used what they dubbed "deep reinforcement learning" to create an AI capable of navigating the 3D environments in Doom while blasting the bad guys. The AI plays just as a human would—it tries to inflict damage while simultaneously trying to avoid being hit.
The AI gets points for moving around, killing enemies, and picking up items, and is docked for being hit or dying. This is where the reinforcement comes into play and it's what makes the AI play Doom like a human, not like one of the programmed bots in the game. Have a look:
"Typically, deep reinforcement learning methods only utilize visual input for training. We present a method to augment these models to exploit game feature information such as the presence of enemies or items, during the training phase. Our model is trained to simultaneously learn these features along with minimizing a Q-learning objective, which is shown to dramatically improve the training speed and performance of our agent," the programmers explain.
In their tests, not only did their AI outperform the programmed bots in the game, it did better than human players. If you're paranoid about a Skynet scenario one day playing out, the takeaway here is that we've taught a machine to hunt humans in a 3D environment better than humans are able to do so.
This is an undeniably impressive achievement (check out the full paper in PDF form here), let's just hope that when the inevitable happens, our robot overlords will decide to keep us around.