Google DeepMind artificial intelligence (AI) technology can play soccer with an ant. The AI technology may be implemented to real products.
The DeepMind AI technology is very smart, and earlier this year, DeepMind's AlphaGo system was applauded worldwide for defeating Lee Sedol, who is the strongest human Go player. Lee Sedol has won 18 world titles, but the Go player lost 4 to 1 against the Google AI. The company says the game was watched by about 200 million people.
"The game of Go is the most challenging of classic games. Despite decades of effort, prior methods had only achieved amateur level performance. We developed a deep RL [reinforcement learning] algorithm that learns both a value network (which predicts the winner) and a policy network (which selects actions) through games of self-play. Our program AlphaGo combined these deep neural networks with a state-of-the-art tree search," says David Silver of Google DeepMind.
It is a big achievement for the Google AI to learn and master Go. However, mastering Go is just the beginning for the DeepMind AI.
Google has now taught a digital ant to play soccer. It may sound dumb, but the technique can help the company understand reinforcement-based learning process. The Google AI learned how to move the ant and also made it kick a ball in the goal.
Google does not elaborate on the significance of the DeepMind AI technology. However, the company suggests that the technology will benefit "robotic manipulation." A robot with legs can start walking and adapt to unseen conditions without any instructions. With the technology, a robot can also safely grab unfamiliar objects.
"The algorithms we build are capable of learning for themselves directly from raw experience or data, and are general in that they can perform well across a wide variety of tasks straight out of the box," says Google DeepMind.
Silver says the team has also built a massively distributed deep reinforcement learning system known as Gorila, which utilizes the company's Cloud platform to speed up training time.
He adds that Google DeepMind has achieved significant results in some of the most difficult Atari games such as Montezuma's Revenge. However, Atari games are only limited to 2D video games. The team has also introduced 3D navigation and puzzle-solving environments called Labyrinth, which will be released under an open-source license in the coming months.
Silver says DeepMind's reinforcement learning agents have demonstrated great progress on a number of challenging tasks. The team aims to improve learning capabilities of the agent and use them for making a positive impact on our society such as the healthcare field.
Photo: Robert Scoble | Flickr
