Friday, May 14, 2021

A far ranging AI discussion with Irving Wladawsky-Berger

I've known Irving since his days at IBM running Linux strategy. Since he "retired," he's been busy with many things, including a number of roles at MIT, where I've kept touch with him including through the MIT Sloan CIO Symposium. We were emailing back and forth last week and discovered that that we've been on something of a similar wavelength with respect to AI. Irving just wrote a blog post "Will AI Ever Be Smarter Than a Baby?" which delved into some of the same topics and concerns that I covered in a presentation at devconf.cz earlier this year.

In this discussion, we explored the question of the nature of intelligence, the answer to which seems to go well beyond what is covered by deep learning (which to put it way too simplistically is in some respects a 1980s technique enabled by modern hardware). 

Among the topics that we explore in this podcast:

The two notions of intelligence. Classifying/recognizing/predicting data and explaining/understanding/modeling the world, which is complementary but potentially much more powerful. 

Whether we need to bring in a stronger element of human cognition (or really even learning/problem solving as we see in the animal kingdom) to take the next steps? And the related work in cognitive science by researchers like Alison Gopnik at Berkeley and Josh Tenenbaum at MIT.

Have we been seduced by great but bounded progress? Can we get to Level 5 autonomous driving?

What will the next 10 years look like?

Listen to the podcast - MP3 [40:29]