As we have seen, Google is taking machine learning very seriously. In 2014 they bought a UK company called DeepMind for US$525 million. In 2016 DeepMind lost US$162 million. Part of that loss was down to the development of Google’s machine learning computer program called AlphaGo. AlphaGo was designed to beat a professional human Go player (Go is a strategic board game in which one player must use moves to populate more space on the board than the other). In 2017 AlphaGo beat Ke Jie 3 nil – Ke Jie was the number 1 ranked player in the world!
Why is this significant?
I can remember a computer program called Deep Blue beating Gary Kasparov at chess in 1997. The significance comes from the complexity of Go. After the first 2 moves in chess there are 400 possible moves, but in Go, there are close to 130,000!
In a more directly relatable example, Google uses machine learning to aid speech recognition. Four years ago, Google had a 20% error rate when translating voice to text. Two years ago, this was an 8% error rate, and it’s now a 4.9 % error rate. The more data it feeds on, the better it gets. This is of particular relevance to AdWords as now 20% of all searches in the Google app are vocal.