DeepMind's AlphaGo artificial intelligence has won the final match of the Go series against world champion Lee Sedol.
During a week long competition in South Korea, the AI from the London-based company forced Go world champion Lee Se-dol to admit defeat in almost every match played.
The final match, which lasted around five hours, saw Lee resign as it reached its conclusion. DeepMind's AlphaGo AI made an early error but managed to recover to clinch victory.
DeepMind co-founder Demis Hassabis expressed his delight at the result. "AlphaGo wins game 5! One of the most incredible games ever," Hassabis tweeted straight after the result. "To comeback from the initial big mistake against Lee Sedol was mind-blowing!!!"
When 3-0 down, Lee managed to inflict one consolation victory against the AlphaGo software. Before the series began, the 17-time world champion believed he would win the series 4-1, or 5-0.
Overall victory for DeepMind marks a significant milestone in the development of artificial intelligence. The 3,000-year-old Chinese board game has proved notoriously hard to master for AI developers due to the sheer number of possible moves.
DeepMind was purchased by Google for an estimated £400 million in 2014. In January the company said it had already beaten the European champion of Go for the first time. Research published in the journal Nature confirmed the company had solved one of the "grand challenges" of AI by beating Fan Hui 5-0.
The AlphaGo AI, which aims to mimic how a human brain learns, effectively defeated humans at Go by a process of trial and error.
It combines advanced search techniques with neural networking, allowing it to both think creatively and take advantage of huge amounts of data about previously played games of Go.
Using 12 layers of neural networks, it selects its next move using just one element of its system while the others predict how the rest of the game will play out. Predicting the future in this way, it adjusts its strategy and moves gradually towards victory.
This article was originally published by WIRED UK
