Last week saw a landmark in computer Artificial Intelligence (AI), one so significant it really should have made more of a stir. For the first time a computer beat a champion level Go player. No, not that wonderful 60’s travel board game that your Gran has, the even older strategy game ‘Go’. For over 2000 years Go has tested the greatest minds in China, though we in the west only learnt of it in the last century or so. It far outstrips Chess and the like in terms of complexity and ‘strategic depth’, yet most of us can happily learn the full rules over a cup of tea. Minutes to learn, a life time to master; sounds like a good computing challenge, right?   So why did it take nearly 20 of computing’s boom years to progress from winning at Chess to conquering Go, and so what?

In the era of big data and rapid technology innovation perhaps it’s no surprise that the computer would win, but this isn’t a triumph of brute force and sheer speed over us puny humans. The great success of DeepMind and AlphaGo comes from applied ‘Deep Learning’ AI techniques. Put (very) simply, the computer has used general purpose mathematical techniques to learn to play Go. Unlike the computers that beat chess grandmasters in the late 90’s, which learned many thousands of the best games by rote, AlphaGo and DeepMind have built their own understand from comparatively few reference human games and a lot of machine-vs-machine practice. It’s a re-birth of the ideas of ‘neural networks’ some thirty years ago, and it re-opens whole new avenues for computer AI that many thought had died.

There is so much to hope for: a robotic system that could spot failing crops even before the best farmer perhaps; a medical device that not only spots disease but pushes our use of imaging sensors to new heights; robots that can finally usefully interact with us every day and overcome the ‘uncanny valley’. What prospects! Machine vision that saves lives, signal processing that feeds millions. Some great hopes for the future, and reborn from the simple joy of playing games. Go figure!

Matt Lacey