Researchers at San Francisco-based lab, OpenAI, developed a program recently to study the ability of AI to learn new techniques and strategies through 500 simulations of hide and seek. The programmers at the lab set of an enclosed 3-D space in which movable and immovable barricades were scattered throughout with one team of three being the seekers and the other team of three being the hiders. What was discovered over the course of the testing is that the AI began with crude tactics that did not yield many results for the hiders, but as the testing progressed the AI were able to conduct more advanced strategies.
The term that describes this evolution of strategies is called "emergent intelligence." In other words, the AI were able to, without directions from the programmers, develop strategies and counter strategies in order to adapt to any given situation. The goal of such testing is to develop AI that may be able to solve the seemingly unsolvable problems of the world. Whether or not such advancement will yield results remains to be seen.
The MIT Technology Review Article
No comments:
Post a Comment