Man beats Machine

Jonathan Anthony
Mar 14, 2016 · 3 min read

Well there is a headline which is a true sign of our times. After losing three times in a row, Lee Se-dol beat AlphaGo at the Chinese game of GO.

Perhaps not that he did it, but how he did it, is more interesting. His seventy-eighth move was seen as 10,000:1 in terms of unexpectedness. So having lost three times, Lee Se-dol learnt from his previous games and tried something unexpected. He thought outside the box.

To understand this a little more deeply it is important to understand the role of training sets in AI. The crux of machine learning is a very repetitive form of training, designed to train the network to slowly converge on correct outputs using inputs which produce “known outputs”. By training the network to give the correct outputs for all the known input combinations we have, we then rely on extrapolation and expect the network to give us correct answers for inputs it has not previously seen before.

For example if we want to train a Neural Network to recognise a specific kitten, we show it hundreds of images of kittens including many of our kitten, and “train” the network until it always gets it right. Next we show it a picture of our kitten and another of a different kitten, both of which have not been seen before. If the Neural Net identifies our kitten correctly then we have successfully trained our machine.

Now back to GO, in the case of GO we can create training sets by having the software play against other versions of itself with an element of move randomisation. Computers can play pretty fast so in a relatively short space of time our computer can play millions of games. As with a human, repeated scenarios reinforce the weightings of those situations and so the unexpected manoeuvre can still remain — well … unexpected, and the very human trait of innovation and thinking “outside the box”, is something that is not necessarily learnt by high-repetition training sets.

So back to the real world, I am not yet that feeling that threatened by AI, but I am very very excited about what it could bring to the world of highly repetitive tasks such as quickly analysing millions of cell samples and lab results. Another example is which is using AI to analyse big data from social media in a way that humans could not do just because of the sheer volume.

On a final note one to watch out for — few initiatives have done more to bring the next generation back to coding than Minecraft (with players creating whole new worlds in code) and now Minecraft are opening the doors to allow player generated AI into the platform. This really might be one to watch for to discover the AI coding stars of the future.

adappt Intelligence

Useful insights into AI from the Hub team

Jonathan Anthony

Written by

Software Architect, TensorFlow, Behavioural Analytics, iPhone, Android , TV Studios, Broadcast Playout.

adappt Intelligence

Useful insights into AI from the Hub team

Welcome to a place where words matter. On Medium, smart voices and original ideas take center stage - with no ads in sight. Watch
Follow all the topics you care about, and we’ll deliver the best stories for you to your homepage and inbox. Explore
Get unlimited access to the best stories on Medium — and support writers while you’re at it. Just $5/month. Upgrade