Aren't two Artificial Intelligence agents that fight against each other the scenario of a scary futurist thriller? A little bit. Inspiring for a lot of movies makers, the future of A.I. is uncertain and unclear, the possibilities are infinite. Technology is progressing so quickly that it is difficult to imagine the completion.
Say hi to "DeepMind"! Created by Google, DeepMind is "the world leader in artificial intelligence research and its application for positive impact". They're on a special mission "to push the boundaries of AI, developing programs that can learn to solve any complex problem without needing to be taught how." - Superior to the human brain capacities I suppose. If they're successful, they believe this "will be one of the most important and widely beneficial scientific advances ever made, increasing the humans' capacity to understand the mysteries of the universe and to tackle some of our most pressing real-world challenges."
The main goal of this work sounds amazingly hopeful and sounds as a positive impact on the human race and Planet Earth that, as a reminder, is a bit in troubles for the past centuries. Are we ready for such a change? Will the AI be able to manage quotidian systems as well as complex ones? Will these agents be able to work smoothly together? Will they fight, work together? Yes, a lot of questions came to my mind while I was reading all these articles. DeepMind has been exploring this potential problem in a new study published on the 9th of February. The researchers tested how AI agents interacted with each one another in a series of "social dilemma" (situation when individuals can profit from being selfish).
In order to test them, DeepMind dropped both AI into a pair of basic video game, they wanted to see how the two agents would perform in this kind of situations. (Cf. Blog Post)
1) First Game: Gathering
Rules: "Two player have to collect apples from a central pile. They have the option of “tagging” the other player with a laser beam, temporarily removing them from the game, and giving the first player a chance to collect more apples."
Result: "When apples were in plentiful supply, the agents didn’t really bother zapping one another with the laser beam. But, when stocks dwindled, the amount of zapping increased. Most interestingly, perhaps, was when a more computationally-powerful agent was introduced into the mix, it tended to zap the other player regardless of how many apples there were. That is to say, the cleverer AI decided it was better to be aggressive in all situations."
"The researchers hypothesize that the increase in zapping behavior by the more-advanced AI was simply because the act of zapping itself is computationally challenging."
2) Second Game: Wolfpack
Rules: "Two players have to hunt a third in an environment filled with obstacles. Points are claimed not just by the player that captures the prey, but by all players near to the prey when it’s captured."
Result: "The cleverer the AI agent, the more likely it was to cooperate with other players." - As the researchers explain, this is because learning to work with the other player to track and herd the prey requires more computational power.
The two AI agents changed their behavior, becoming more cooperative or more combative, depending on the context of the game. The results of the study show that the behavior of AI agents changes based on the rules they’re faced with.
That means that strict rules will have to be in place to fully control the AI agents.
A long work is still necessary to develop these agents and let them manage some work on their own. DeepMind concluded its post writing that “As a consequence [of this research], we may be able to better understand and control complex multi-agent systems such as the economy, traffic systems, or the ecological health of our planet - all of which depend on our continued cooperation.”
Learn more about DeepMind.
Source (c) DeepMind - The Verge