Google's DeepMind API Learns human Behaviours
DeepMind Technologies is a British artificial intelligence company founded in September 2010. It was acquired by Google in 2014. The company has created a neural network that learns how to play video games in a fashion similar to that of humans, as well as a Neural Turing Machine, or a neural network that may be able to access an external memory like a conventional Turing machine, resulting in a computer that mimics the short-term memory of the human brain.
Researchers have overcome one of the major stumbling blocks in artificial intelligence with a program that can learn one task after another using skills it acquires on the way
The AI is not capable of the general intelligence that humans draw on when they are faced with new challenges; its use of past lessons is more limited. But the work shows a way around a problem that had to be solved if researchers are ever to build so-called artificial general intelligence (AGI) machines that match human intelligence.
“If we’re going to have computer programs that are more intelligent and more useful, then they will have to have this ability to learn sequentially,” said James Kirkpatrick at DeepMind.
The ability to remember old skills and apply them to new tasks comes naturally to humans. A regular rollerblader might find ice skating a breeze because one skill helps the other. But recreating this ability in computers has proved a huge challenge for AI researchers. AI programs are typically one trick ponies that excel at one task, and one task only.
The problem arises because of the way AIs tend to work. Most AIs are based on programs called neural networks that learn how to perform tasks, such as playing chess or poker, through countless rounds of trial and error. But once a neural network is trained to play chess, it can only learn another game later by overwriting its chess-playing skills. It suffers from what AI researchers call “catastrophic forgetting”.
Without the ability to build one skill on another, AIs will never learn like people, or be flexible enough to master fresh problems the way humans can. “Humans and animals learn things one after the other and it’s a crucial factor which allows them to learn continually and to build upon their previous knowledge,” said Kirkpatrick.
To build the new AI, the researchers drew on studies from neuroscience which show that animals learn continually by preserving brain connections that are known to be important for skills learned in the past. The lessons learned in hiding from prey are crucial for survival, and mice would not last long if the know-how was erased by the skills needed to find food.
The DeepMind AI mirrors the learning brain in a simple way. Before it moves from one task to another, it works out which connections in its neural network have been the most important for the tasks it has learned so far. It then makes these harder to change as it learns the next skill. “If the network can reuse what it has learned then it will do,” said Kirkpatrick.
0 comments