Demis Hassabis led and Google owned company DeepMind has started developing an AI (artificial intelligence) agent capable of having ‘imagination’ on its owns and solving real world unpredictable scenarios – enabling machines to predict the consequences of their actions before they even make them.
DeepMind was founded in 2010, London by some of the most successful technology entrepreneurs in the world – Demis Hassabis, Mustafa Suleyman and Shane Legg. Having been acquired by Google in 2014, DeepMind is now part of the Alphabet group and currenly employing over 400 top notch techies.
DeepMind created a storm when its narrow AI program AlphaGo defeated Lee Sedol – a world reknowned professional Go player and South Korean Go legend in a five-game match series to win it by 4-1 last year in March. By doing so AlphaGo became the first Computer Go program to beat a human professional Go player without handicaps on a full-sized 19×19 board. AlphaGo AI uses machine learning and neural networks to analyze its probability of winning and pick moves that are likely to lead to success.
Until now, solving the ancient Chinese board game of Go had long been a north star for computer scientists due to its unparalleled complexity and huge number of potential moves. DeepMind’s success came decades earlier than many had predicted.
And it keeps getting better…
The story with AlphaGo became more interesting when the current world no.1 Go player Ke Jie challanged AlphaGo and lost to it this year May and wisely admitted defeat by saying “I used to think that it must have weak points, but now I’m convinced that its understanding of the Go game and its judgment during matches are vastly superior to human beings,” He also said when interviewed after the game “AlphaGo made an unexpected and novel move. It’s very unlikely for human players to think that way,” Ke said. “The single move served multiple purposes, so I totally deserved to lose”.
left: Lee-Sedol right: Ke Jie
In a recent report of a DeepMind it is mentioned that “The real world is complex; rules are not so clearly defined and unpredictable problems often arise. Even for the most intelligent AI agents, imagining in these complex environments is a long and costly process. But with the invention of imagination-augmented agents which is nothing but a neural network that learns to extract information that might be useful for future decisions, an AI agent can now learn different strategies to construct plans and choose from a broad spectrum of strategies.”
DeepMind executed their AI on multiple tasks, including Shokoban, a puzzle game and a space ship navigation game. DeepMind claims that, in both the tests, the imagination augmented AI agent came up on top compared to the imagination-less baselines.
We may very well already started living in an era of sophisticated AIs and its not far since we interact sophisticated AI programs in e-commerce, social media, entertainment and even in educations. Be ready to celebrate the birth of an intellectual being – AI.