Planning, not so well solved AI problems

by Forrest Sheng Bao

A big area in artificial intelligence is planning, e.g., finding a configuration (mostly temporal) of using different resources to achieve a goal. Maybe you feel this is sorta linear programming you learned in your college algebra class. But the thing is, AI researchers want to find a universal automated way. For decades, they have been wanted to build a planner that can automatically handle all planning problems, humanly or rationally. Therefore, you don't have to model a problem into a constraint programming manually and do the same for a different planning problem.

But this problem has been far from being solved.

Apparently, commercial applications of planning algorithms do not work well. The itinerary below is an example. United Airlines thought a good way to fly from Singapore to China is to do a U-turn in San Francisco over the Pacific Ocean via Japan. A five-year-old would even think this is very inhuman. Plus, it's very ``non-green.''

No comments: