AlphaGo Zero, the AI programme that can teach itself

AlphaGo Zero

Last year, Google DeepMind created its AI programme AlphaGo, the machine that was able to beat world champion Lee Sedol at Go, a complex game of strategy and intuition. Unlike AlphaGo, which was programmed with millions of moves of past masters in order for it to predict and adjust its game plan accordingly, the new AlphaGo Zero has now been created, and is able to learn from scratch.

 

AlphaGo Zero was taught how to play the game, but was given no additional instructions. It was able to learn all the best moves as well as develop its own strategies simply by playing millions of games against itself, something that would have taken humans thousands of years to do.

 

The new AlphaGo Zero has led people to be sceptical of its abilities, with belief that AI is being used to overtake and replace humans. However, the idea behind the new machine is not to defeat humans, instead it is to enable a programme to learn for itself what knowledge is and hopefully find solutions to problems a lot quicker than humans potentially could. It was found to operate in quite a different way compared to a human player during the game of Go, so there is a great interest to see what else it can achieve.

 

This method of blank slate learning, in which the programme becomes its own teacher, needs no human knowledge, data or any intervention, has the ability to be transplanted from the game of Go to any other domain.

 

“If similar techniques can be applied to other structured problems, such as protein folding, reducing energy consumption or searching for revolutionary new materials, the resulting breakthroughs have the potential to drive forward human understanding and positively impact all of our lives.” Said DeepMind.

 

Misfolded proteins are responsible for many devastating diseases, including Alzheimer’s, Parkinson’s and cystic fibrosis. If AI can help solve big health problems like these then this will be a huge breakthrough.

 

Having said this, the game of Go is governed by a certain set of rules which makes it much easier to understand and master than more abstract issues like these. It is an enormous step, but AI programme AlphaGo Zero is a significant step from the original AlphaGo, so we are on the right path despite the time it may take.

 

Source: http://www.telegraph.co.uk/science/2017/10/18/alphago-zero-google-deepmind-supercomputer-learns-3000-years/

Leave a Reply