She explains some essential tenants and goals, including
She describes their work as a mix between academia and industry. Their mission is to build something that can solve any task at least as good as any human can. However, she explains that while there’s a large group of researchers who think that we just need to put together single AIs who can perform one task and see what they can do in combination, she comments that such systems are still brittle—a little bit of noise can throw the whole algorithm off.
She adds that given how the complicated the natural world is, she doesn’t think we can come up with enough narrow AIs to handle problems. At this point she brings in her neuroscience, trying to create a model able to make unsupervised transfers of learning as the brain does.
For example, we as humans understand the abstract notion of a paddle and a ball and keeping the ball in the air and can transfer those ideas to another game. She adds that if we can get a computer to do that same transfer, that’s a huge leap forward. She further describes some of the ways she’s trying to get to that point.
To keep up to date in this field of research, she recommends following blogs in the discipline, such as Google’s DeepMind blog: https://deepmind.com/blog, and finding AI research scientists on twitter.