Last week, Google's artificial intelligence program AlphaGo dominated its match with South Korean world Go champion Lee Sedol, winning with a 4-1 score.
The achievement stunned artificial intelligence experts, who previously thought that Google's computer program would need at least 10 more years before developing enough to be able to beat a human world champion.
What could be scary regarding the computer program is that Google DeepMind CEO Demis Hassabis said that AlphaGo could still improve its performance, as the match with Sedol was able to expose some of its weaknesses.
Computers have long been winning against skilled humans in games - Deep Blue defeated chess legend Garry Kasparov two decades ago, and IBM Watson beat Jeopardy players in 2011. However, the victory of AlphaGo is even more impressive due to the sheer number of possibilities for a single move in the game of Go.
As such, the win by AlphaGo did not only use the computing strength of supercomputers, but also deep learning, which utilizes neural networks that allow computer programs to learn just like humans.
Geoffrey Hinton, the godfather of deep learning who helped in the development of AlphaGo, discussed the meaning of the computer program and hailed its achievement as particularly momentous in an interview with Maclean's.
Hinton thought that AlphaGo's wins were quite exciting, though the team behind the computer program were nervous that AI program could have some weaknesses that would come to surface during the match with Sedol.
The importance of AlphaGo's win over Sedol is the fact that it shows how computers are now also able to rely on its own intuition, which is something that was thought only humans can do. The intuition of AlphaGo was provided by its neural network, giving the program all the possible moves and the intuition on what would be the best move.
Hinton, however, believes that deep learning will not be translated into human-level abilities until systems are developed that have the same number of parameters as the human brain. In comparison, the human brain has 1,000 trillion synapses, while current neural networks only have about 1 billion synapses.
Hinton also answered to fears that artificial intelligence and deep learning will one day allow computers to dominate humanity, stating that the technology is being used for a lot of good things in the world right now. The technology could lead to bad things if used wrongly, but that is due to the involvement of the politics of the technology and not caused by the technology itself.
According to Hinton, preventing artificial intelligence from harming or making humans obsolete should not be done by crippling the technology. Instead, the changes should be made to the political system to ensure that the technology is not used to harm humans.
The development of deep learning and artificial intelligence has been vindicating for Hinton, as he has been working on the technology since the 1980s. At the time, the field of artificial intelligence dismissed the concepts of deep learning and neural networks. What changed since then?
"Mainly the fact that it worked," Hinton said, and for AlphaGo and its match with Sedol, it very much did.
