Alphago

AlphaGo defeats Go champion: what this means for artificial intelligence and machine learning

How significant is it that Google’s artificial intelligence-powered AlphaGo beat a world number one at his own game?

It is the world’s oldest current board game but it might be the setting for a new major development in the technological revolution.

Google’s AlphaGo has defeated the world’s number one Go player Ke Jie using some of the most highly developed artificial intelligence and machine learning available.

What is Go and why is it relevant to AI?

Go was invented in China more than 2,500 years ago and is considered to be one of the world’s most complex games.

It is a strategy game played on a 19×19 gridded board on which players take turns placing black and white stones on the grid in order to capture each other’s territory.

AI researchers favor it because there are a much larger number of outcomes compared to other strategic games such as chess. In fact, it is claimed that there are more potential moves in Go than there are atoms in the universe. To illustrate, one round of chess has 400 possible board positions compared with 129,960 in Go.

Go’s complexity lies in its need for sophisticated evaluation functions. In chess, material value outweighs other aspects of knowledge – the player with more important pieces left is generally winning. However, Go has a pattern-based knowledge which is incredibly complex to program for.

How did they develop the technology for AlphaGo?

When it first hit the headlines in March 2016 after beating 18-time world champion Lee Sedol, AlphaGo combined advanced tree search with deep neural networks to process its moves. These neural networks were able to process the playing board through 12 different network layers – each with millions of connections. While the “policy network” selected the next move to play, the “value network” predicted the winner of the game.

This year’s version differs through its increased efficiency. AlphaGo now runs on a single Tenor Processing Unit machine, using ten times less computing power than before while learning at a faster rate. Therefore, it is able to learn the game almost entirely by playing against itself, rather than relying on the data generated by humans. This technique of deep reinforcement learning, in theory, allows systems to focus on a wider range of tasks beyond Go.

What does this mean for AI and the future?

Google has long claimed that its DeepMind technology can address socio-economic challenges. It has already reached a five-year agreement with the UK’s National Health Service to develop its healthcare app. In addition, DeepMind is being used to reduce energy consumption in data centers by 15%, making cooling systems 40% more efficient.

There are numerous examples of AI and machine learning already bolstering struggling services. Artificial intelligence is performing better than standard medical guidelines in predicting heart attacks for example. The University of Nottingham analyzed electronic medical records from 2005 using four machine-learning algorithms to make predictions based on their 2015 records.

Using Appropriate Use Criteria (with 1.0 being the highest) the algorithms registered predictive scores of 0.745-0.764 compared with 0.728 for the general guidelines. Or in a more tangible sense, from a sample of 83,000 records, the best algorithm could have saved an additional 355 patients’ lives.

It is not inconceivable to expect these breakthroughs to become more common and more apparent as technologies such as DeepMind continue to develop.

Here sooner than we think

Like a number of tech commentators, Google believed it would take AlphaGo far longer to conquer the board game world than it did.

DeepMind has already turned its attention to StarCraft II, a complex strategy game considered even more challenging than Go due to the number of variables attached to it. For example, players are unable to see the whole map at once, giving them less information to work with when planning moves.

Of course, AI companies have far loftier ambitions. Given how AlphaGo went from an amateur Go player to virtually unbeatable against humankind within just two years, it is clear that an end game of AI providing us with tangible benefits to society is closer than we think.

Inbenta is a leader in natural language processing and artificial intelligence for customer support, e-commerce and conversational chatbots, providing an easy-to-deploy solution that improves customer satisfaction, reduces support costs, and increases revenue.

Interested in finding out more? Our team of experts is available to show you how Inbenta can benefit your company.

Let’s get in touch