Hybrid Reward Architecture This is the name given to the methodology applied by the team at Maluuba. The architecture consists of a super agent which is responsible to carry out decisions. This super agent is assisted by over 150 normal agents which provide input data for the super agent. With respect to this game, the agents worked in parallel with some looking for pellets while the rest kept an eye out for ghosts. The super agent used the data replayed by them to decide the direction to move Ms. Pac-Man in. Rahul Mehrotra, a program manager at Maluuba, had this to say when quizzed on why the team choose a game from the 1980’s for their research, “A lot of companies working on AI use games to build intelligent algorithms because there’s a lot of human-like intelligence capabilities that you need to beat the games. ” There has been a paper published on this research and accomplishment wherein the author Harm van Seijen states that “(this breakthrough) enables us to make further progress in solving these really complex problems.” Mehrotra also mentioned a possible real-life implementation of their methods and architecture with a company using them in their sales department by using multiple agents to target clients and use the data by these agents to improve the probabilities of making a sale. Source: Neowin
