Jamie is excited that AlphaGo, the system built by DeepMind, has beaten the leading player of the game, Go. 

The Go game is played on a board of 19 by 10 squares, so there are literally hundreds of thousands of options for each move.

“Awesome”, he said.

The company, now owned by Google, is the creation of Demis Moussabis. His dad was a Greek Cypriot who ran a toyshop in North London and his mum worked for John Lewis. Demis turned out to be a chess prodigy.

They trained AlphaGo on a library of thousands of actual games played by humans. That got it to an average level. But the big breakthrough came when the system started to play against itself.

While chess can be learned by a machine that can see, at very high speed, a “tree” of future moves and countermoves and choose the best option, Go is just too complex for that and the number of branching options becomes infinitesimal.

The system uses layers of neural networks which must develop something like a game theory, something more intuitive, more like a human brain, but supported by a huge memory and working thousands of times faster. 

I find myself thinking about a practical problem, like one I have at the moment. If the new expressway is built and nearly all the crossings are closed, how much more traffic would there be on the arterial side roads and country lanes?

People don’t always take the shortest route, but sometimes the quickest route, or one that avoids lanes with limited passing places or intermittent obstructions, animals crossing, farm machinery moving.

People have different options and different people choose differently.

A huge number of journeys could be monitored over an extended period and crunched to give a more accurate statistic.

Jamie says: “I admire what DeepMind have done. It’s cool. It’s great showmanship. Winning a game is a long way from getting the right answer to a complex question. But it is a step towards Singularity.”