Artificial Intelligence begets Artificial Intelligence

Create an intelligent program that can create, autonomously, an even more intelligent program. The new frontier for Artificial Intelligence. Credit: TechCrunch

An interesting research a few years ago have shown that the brain of a professional basketball player works "less" than the one of an amateur when trying to throw the ball through the "hoop". The reason is that the brain of a professional player, through continuous exercise has modified itself to better manage, with more accuracy and less energy expense, the actions needed to win the basket.

This is notable: the brain does not learn, in this case and in several others like learning to ride a bike, by storing the knowledge somewhere but by rewiring itself. In a way, using computer terms, it is like the brain has the capability of reprogramming the way it works.

Well, the progress in artificial intelligence is aiming at something pretty similar: having an artificial intelligence program able to create more artificial intelligence programs to address new areas (or similar one in a better way).

At Berkeley, Google and MIT researchers are working to create AI software that can create AI software. 

Automatic machine learning is something that has been a reality for a few decades and in the last decade it has become quite effective. Alpha GO learnt to play GO as a master by looking at thousands of games and by playing against itself millions of time. Now, however, the goal is to move one, big, step forward by having a program writing a program that is "better" than the program writing it. If this is achieved we are opening the door to a tremendous increase in capabilities.

Does this mean that in a few years (human) programmers will be out of job? Yes and no. For sure, in this area as in several others, automation is chewing on workforce. At the same time achieving this type of results requires tremendous processing power, and involves very high cost. Google Brain is trying to create -through a program- a program that can be better than humans at recognising images and to do that is using 800 high processing power graphic computers. 

MIT Media Lab is opening its research through an open-sourcing to stimulate other groups to contribute to the effort.

Author - Roberto Saracco

© 2010-2018 EIT Digital IVZW. All rights reserved. Legal notice. Privacy Policy.