What if machines could not only learn, but learn by themselves without the input of a human mastermind? It’s a radical idea, one that Hollywood has been spinning out for years in Sci-Fi movies, but thus far, has remained largely a fantasy.
Yoshua Bengio not only believes it will be a reality in the future, but thinks he’s discovered a key puzzle piece to finally making Artificial Intelligence a success.
It’s called Deep Learning and basically gives computers intuition, allowing them to form their own decisions when developing a new concept or idea. The AI techniques of the 70s and 80s failed primarily because they lacked this and focused on machines that could explain each step through reasoning. According to Bengio it’s much easier to train machines to make the right choices by intuition.
The exciting revelation will hopefully allow computers to learn without much input from humans, which in the past involved “labeled data.” If scientists wanted a computer to recognize a car, they first had to show it a car. Bengio hopes to eliminate this step completely.
“Today’s models can be trained on huge quantities of data, but that’s not enough. We need to discover learning algorithms that can take better advantage of all this unlabeled data that’s sitting out there,” says Bengio.
It’s an exciting idea and one that Bengio cites as a breakthrough of his career. Since the brainstorming session of mathematical algorithms first came to him at an academic conference last May he’s been working with his team at the University of Montreal to test the data in small sets. These new models he says can learn to spew out accurate images such as a human face, without human interruption. If the model is shown only part of a nose or a few words of sentence, it can fill in the rest by itself.
It looks like what Steven Spielberg and James Cameron had only dreamed of in their movies, will one day be a real thing in computer science. While the opportunity for economic success is of course evident, Bengio cites that he’s only doing the experiments to find out why, not put a product on the shelf tomorrow. “Once you have that understanding, you can answer questions – you can do all sorts of useful things that are economically valuable.”