MIT incorporates human intuition in artificial intelligence to help computers plan better

MIT researchers have improved award winning automatic planning software by adding in code that mimics human intuition. The strategies used by high performing human planners were converted into a machine readable form, and then encoded into the automatic planning software. Adding human intuition to the planning software saw an increase in performance between 10 to 15 percent on a challenging set of problems.

The research was conducted by scientists at  Computer Science and Artificial Intelligence Laboratory (CSAIL), which is known for a number of cutting edge artificial intelligence breakthroughs. The results from the finding will be presented at an upcoming conference of the Association for the Advancement of Artificial Intelligence .

Julie Shah, an assistant professor of aeronautics and astronautics at MIT, explained the approach  “In the lab, in other investigations, we’ve seen that for things like planning and scheduling and optimization, there’s usually a small set of people who are truly outstanding at it. Can we take the insights and the high-level strategies from the few people who are truly excellent at it and allow a machine to make use of that to be better at problem-solving than the vast majority of the population?”

The researchers made MIT students themselves solve some challenging problems, such as the most efficient ways to route planes given a list of planes, locations, people and destinations. A simple version of the problem is to make sure everyone arrives at the destination, and no plane flies empty. A more advanced version adds constraints to the problem, such as minimising the fuel use and flight time. Finally, the most advanced version of the problem includes temporal constraints, such as specific take off and landing times for the aircraft.

MIT students were better than the automatic planning software. The researchers quizzed the MIT students on how they approached the problems. The strategies used by the human problem solvers were described in a simple statements that could be understood by machines, in a formal language known as linear temporal logic. The resulting plans from the improved software were more similar in nature to plans that humans came up with.

The researchers are now working on natural language processing based mechanisms to automatically incorporate human strategies into planning software. Instead of humans describing to other humans how a problem was solved, the team hopes to use machines to translate strategies used by humans into linear temporal logic that can directly be fed into machines.