AI systems are trained to perform one task. While such systems are immensely useful, future AI systems will be general, with common sense and general problem solving ability. To train such systems, we created OpenAI Universe, a software platform that allows an agent to interact with any program or website via pixels, keyboard, and mouse, thus providing access to the vast digital universe of existing programs and websites, and therefore, a great variety of tasks. In this talk, I will present Universe, and preliminary results in a research direction that, if pushed to its logical extreme, will result in systems with a general problem solving ability.
Ilya Sutskever received his PhD in 2012 from the University of Toronto working with Geoffrey Hinton. After completing his PhD, he cofounded DNNResearch with Geoffrey Hinton and Alex Krizhevsky which was acquired by Google. He is interested in all aspects of neural networks and their applications.