DeepMind’s Demis Hassabis interview on TedTalk

https://www.youtube.com/watch?v=I5FrFq3W25U

Some interesting points to me.

He thinks chess should be part of the school curriculum.  It teaches phenomenal skills.

What would happen if AI just read Wikipedia? The AI would not be grounded as it is not living in the real world.  Some of our assumptions about intelligence may not hold.  AI would not know, for instance that a dog has 4 legs.  ChatGP3 tends to allucinate.

Alphafold’s immediate goal is to solve the protein folding problem.  It’s immediate application is acceleration of drug discovery in designing new molecules and new compounds.

What is the most unsolved problem in your field that you would like to see?

The notion of abstract concepts of conceptual knowledge is quite rudimentary so far.  It’s transfer learning or analogical reasoning in psychology.  We, as humans, do something like this effortlessly in our own brain.  In a domain, find the underlying structure, and apply it to another domain.  AI systems don’t do that in a satisfactory way.

What’s missing is true creativity.  AI can invent new game strategies but cannot invent a game of GO.  Not there yet.

The ultimate goal is to understand the whole universe around us.

Leave a Reply

Your email address will not be published. Required fields are marked *