When AI Tech Turned a Corner — Google’s Deep Mind AlphaGo Beating Lee Sedol
I meant to write this blog several months ago. However as a compulsive procrastinator I kept putting it to another day, until today.
On January 27, an article in Nature reported on a computer that had beat a human player at Go. It is an ancient board game that has long been viewed as a hard nut to crack for Artificial Intelligence (AI).
Till then, computers had already beaten the best human players of backgammon, draughts, and chess. To win at Go, computers needed a massive handicap. So DeepMind, a London-based Google AI company went to the lab and created AlphaGo, a computer that could outplay humans.
Faced with defeat
Till then, human grandmasters had maintained an edge no matter how agile the computer system was. That’s perhaps why top AI experts outside Google whether there was going to be a breakthrough anytime soon. In fact, for them, it was another decade before a machine could outplay the top human players.
So, Fan Hui, the European Go champion was matched up with a computer. Hardly did he know he was faced with a defeat. Apart from outplaying Fan Hui, AlphaGo defeated the best available silicon-valley-based programs. It won 99.8 percent of all the game played.
More complex than chess
However, the programmers of AlphaGo weren’t quite there yet. They had to match up the program with world Go grandmaster Lee Sedol to prove they had the best AI machine.
The match happened in March 2016. Out of the five games played AlphaGo won four proving it had mastered a game this ancient game. Of course, computers had beaten the best humans at chess, Othello, and checkers. Go, however, is an exponentially more complex than chess. It requires its best human players to have a greater degree of intuition.
Its significant
This win is significant as the technologies behind AlphaGo are already being used in real life applications. Many of these technologies are being used to identify faces in photos and recognise commands spoken into smartphones. Some techniques at the heart of Google’s AI might just be used to remake everything from robotics to scientific research.
For a grandmaster, Lee Sedol was so overwhelmed that he couldn’t return to finish within one minute of his AI rival. But he did prove his mettle by leading in the first half of game five. This is given an amateurish error by the AI machine. What followed was an edge-of-the-seat drama in the second half, which the machine eventually won.
The algorithm
AlphaGo’s software was developed by a team of 20 persons at DeepMind Technologies, a Google company. Interestingly, there is nothing novel or insightful about AlphaGo’s algorithms. The software is a combination of machine learning techniques and neural network algorithms. It’s the work of incredible software engineering. The software runs on powerful but fairly standards hardware consisting of 48 central processing units (CPUs). It is supported by eight graphical processing units (GPUs) that are powerful enough to run certain operations in mathematics. It’s the same technology that recognizes photos posted on Facebook and automatically labels them. Has the penny may have finally dropped for AI Tech?
