Microsoft’s AI agents have managed to beat Ms. Pac-Man. The 36-year-old classic has been a difficult nut to crack for AI, but researchers’ efforts eventually paid off.
We all know that the Redmond giant and Google are in a continuous competition. As a quick reminder, just last month, Google’s AlphaGo defeated the best Go players in the world, so Microsoft had to come with a similar achievement.
More about Microsoft’s AI-based system
The AI system that beat Ms. Pac-Man relies on a divide-and-conquer model. The team from Maluuba, a Canadian deep learning startup acquired by Microsoft, used reinforcement learning to achieve the maximum score possible of 999,990. To beat the game, researchers divided the problem into small pieces, which they then distributed among AI agents. In other words, the AI agents worked on different pieces in order to achieve a common goal.
This is an important milestone in AI research that could have broad implications for teaching AI systems to do complex tasks that enhance human capabilities. This divide-and-conquer model will definitely open the door to new possibilities in AI research.
Here’s how this method works: the AI system uses more than 150 agents, working in parallel with the other agents. Agents get rewarded for successfully completing a particular task.
The system also uses a top agent that collects and analyzes the suggestions from all the agents in order to decide where to move Ms. Pac-Man.
The top agent takes into account a plethora of variables, but gives more weight to suggestions that make the best move for everyone. For example, if 140 agents want to go right because that is the best path to their pellet, but two vote to go left because there is a deadly ghost to the right, the top agent decides to go left.
Interestingly enough, researchers confirm that the best results are achieved when each agent acts egotistically, while the top agent behaves very ‘altruistically and empathically’, taking the decision that best serves the interests of all the agents.
Microsoft’s AI success in beating Ms. Pac-Man demonstrates once again that humans have managed to build intelligent algorithms. AI systems need to have human-like intelligence capabilities in order to beat the games, and both Microsoft and Google have been very successful at it lately.Follow The AI Center on social media:
Join me as I track the latest progress in AI research.
Latest posts by Mary Blaut (see all)
- AI powers staff-free stores: Shop without any human interaction - January 18, 2018
- AI helps researchers improve gene editing results - January 13, 2018
- True Emoji is an AI app that uses your expressions to create animated emojis - November 21, 2017