|Google's DeepMind Demis Hassabis Gives The Strachey Lecture|
|Written by Mike James|
|Wednesday, 30 March 2016|
You can't have missed the amazing advances in AI that have been happening at Google's DeepMind. First a neural network learns to play arcade games and then convincingly beat's the world's top Go expert. The founder and CEO of DeepMind, Demis Hassabis, gave this year's Strachey Lecture in Oxford and had a lot to say.
The Strachey Lectures are a series of lectures organized by Oxford University in honor of Christopher Strachey who, among other computing achievements, was the first Professor of Computation at Oxford.
This year's Strachey Lecture was given as a live stream on February 24th but you can watch it as a video.
Dr. Demis Hassabis is the Co-Founder and CEO of DeepMind, the world’s leading General Artificial Intelligence (AI) company, which was acquired by Google in 2014 in their largest ever European acquisition. Demis will draw on his eclectic experiences as an AI researcher, neuroscientist and videogames designer to discuss what is happening at the cutting edge of AI research, its future impact on fields such as science and healthcare, and how developing AI may help us better understand the human mind.
The video covers some philosophical points but there is a lot of general information about what is going on at DeepMind.
The key idea is the bringing together of deep neural networks with an older technique - reinforcement learning. The idea is that the neural network isn't told how close it is to a desired result, which is what happens in supervised learning. Instead the network is provided with a reward that tells it how well it is doing. The network then modifies its behavior to increase the reward. This is an old idea and it has been tried many times in the past with mixed results. What has changed is mainly that now we can make use of very deep networks that in the past would have been impossible to train just because we didn't have the computing power.
Now that we do have the computing power, deep reinforcement learning seems to be powerful enough to learn computer games just by playing them, to navigate mazes and learn to play Go at a grand master level.
Watch the video to find out more;
I'm not sure that the question of the future has been answered, but you can tell that there will be a deep reinforcement network in it. Perhaps this is the start of realizing the goal of strong AI.
or email your comment to: firstname.lastname@example.org
|Last Updated ( Wednesday, 30 March 2016 )|