DeepMind, Google's Artificial Intelligence division in London, has been concerned with what will happen when two or more AIs have similar or conflicting goals. The team wanted a test that is similar to the "Prisoner's Dilemma," which is a popular game that pits two suspects against one another.
In this scenario, you are given a choice: testify against the other person and you'll be free, while they have to serve three years. If you both say yes independently, however, then you will serve two years in jail.
Google Develops Two New Games For Its AIs
In order to test its AI agents, DeepMind has developed two new games, called Gathering and Wolfpack.
Gathering
According to Engadget, Gathering is a game where two colored squares are tasked with picking up "apples" in the middle of the screen, where they can also fire a laser which, if fired accurately, can remove the other character from the game temporarily. How cooperative or combative would they be? Unsurprisingly, the AI pair was quite peaceful at the start, collecting apples at a constant and steady pace. However, as the fruit pile dwindled, they learned to become more aggressive and fire their laser at each other.
Wolfpack
On the other end, Wolfpack is a game where it challenges the two AIs to cooperate in finding a third, fleeing piece of AI. Blocky obstacles are littered on the screen, so the AI teammates have to figure out when to flank and corner their opponent. DeepMind noted that in this game, higher cognitive capacity lent to more cooperation between the agents, which is in contrast to Gathering that produced more conflicts when the "smarts" were increased.
The Results
According to Bloomberg, the differing results mainly highlight the importance of intelligence and also the game and its underlying ruleset. The AIs have behaved differently depending on the task at hand and what's required in order to be victorious.
The findings are extremely important as humanity releases multiple AIs into the world in the near future.
