AI learns to play hide and seek and shows the way to “intelligent agents”
Alright folks, it’s time to come together again and see what kind of madness has been conjured up by OpenAI, the multi-billion dollar non-profit organization that continues to crank out a ridiculous breakthrough in AI after the other. Most recently, the company released the very human-sounding GTP-2 language modeler, and just released a new video of virtual agents learning to play hide and seek. And while the agents are super cute, the implications of their ability to learn in changing environments are Macaulay Culkin Alone at home against the max.
In this virtual hide-and-seek clip (provided via Digg), OpenAI shows agents learning to play the simple children’s game from scratch. We observe that agents, which are essentially software bots whose “goal” is to win the game, gradually become so proficient that they even invent more ways to explore the virtual environment than their programmers thought.
The most amazing aspect of this demo is the fact that OpenAI says it could be proof of concept for the idea that this type of gameplay can more or less be used to evolve virtual agents that are “really complex and intelligent”. The video notes that the agents are already programmed with animal intelligence, in that they keep trying to figure out the set of actions they need to take in different environments in order to get the desired reward.
Another crucial point here is that these agents learn unattended. This means that none of the strategies agents use to hide and seek are programmed into them; every action that the little players on the blue and red team perform is self-taught and necessarily stems from their main objective (or utility function) of winning the game.
This all gets even crazier when you consider the fact that OpenAI is also working on machines that can intelligently manipulate the real world, as evidenced by the “Dactyl” robotic video below.
If OpenAI can combine the learning capabilities of their virtual agents (once they’ve been trained in much more complex environments) with their prototype agile limbs, it seems like there’s only one step to take. imagine a very fast learning robot capable of manipulating the real world. . A very fast learning robot that can also track you down no matter where you hide.
What do you think of these virtual hide-and-seek agents? Are they great proof of concept for more complex virtual intelligence, or just really good bots learning to play a specific game? Let us know in the comments!