Artificial Intelligence Prize for Colin Johnson's Research
5 February 2020
The paper entitled Stepwise Evolutionary Learning using Deep Learned Guidance Functions discussed the design of a new approach to artificial intelligence, where the computer generalises from a collection of solutions to a problem in order to tackle the problem in general. This was applied train the computer to solve the Rubik’s Cube puzzle: the program learned what an almost solved cube looked like, then what a slightly less solved cube looked like, and so on. Then, given a scrambled cube, it could work out how to unscramble the cube step-by-step – reducing the cube to a slightly less scrambled state, then an even less scrambled state, and so on until the problem was solved.
This work potentially opens out new approaches to a wide variety of problems, such as understanding how to untangle proteins in biology, solving complex business planning problems, and removing noise from historical video and audio recordings.