A research team at Microsoft has used an artificial intelligence (AI) system to master Ms. Pac-Man for the Atari 2600. What makes the feat so impressive is that the conglomeration of AI agents working in unison managed to achieve the highest score possible, 999,990. That's something that even no human has ever achieved, the team says.
Video games, it turns out, are an excellent testing ground for teaching AI to exercise human-like intelligence. Ms. Pac-Man itself is particularly well suited to this task because it was designed to be less predictable than Pac-Man.
To take on the task, the researchers concocted a system of more than 150 AI agents working together to obtain the best outcome from any given move. From Microsoft:
The method, which the Maluuba team calls Hybrid Reward Architecture, used more than 150 agents, each of which worked in parallel with the other agents to master Ms. Pac-Man. For example, some agents got rewarded for successfully finding one specific pellet, while others were tasked with staying out of the way of ghosts.Then, the researchers created a top agent – sort of like a senior manager at a company – who took suggestions from all the agents and used them to decide where to move Ms. Pac-Man.The top agent took into account how many agents advocated for going in a certain direction, but it also looked at the intensity with which they wanted to make that move. For example, if 100 agents wanted to go right because that was the best path to their pellet, but three wanted to go left because there was a deadly ghost to the right, it would give more weight to the ones who had noticed the ghost and go left.
The research comes from Candian company Maluuba, which Microsoft only just acquired earlier in 2017. According to the team, the results of this research could prove valuable for researchers across the AI field, but particularly in the area of reinforcement learning. Eventually, that could lead to AI that can make decisions on its own, eventually allowing it to take over complex tasks currently performed by humans.
