December 28, 2019

Ray Dalio

Ray Dalio was recently featured on Lex Fridman’s Artificial Intelligence Podcast. I think he offers a good perspective on the limits of AI. A very valuable perspective given the amount of hype around AI today.

“If the future can be different from the past, and you don’t have deep understanding [of the cause effect relationships that are leading you to place that bet], you should not rely on AI.” -Ray Dalio

When you look at the cases where machine learning has been particularly successful (e.g. computer vision, natural language processing) these domains, for the most part, have datasets that do not change much over time. Also they have cause-effect relationships that seem to make sense (e.g. in image classification an image is only likely to be classified as a dog if there is a dog in the picture).

There are many domains where these two criteria do not hold. One that stands out is the financial markets. What influences the direction of the market in the future may very well be different from the past, and many reasonable people would disagree on what exactly drives market fundamentals or valuations.

Ray Dalio’s comments can be found from 46:24-58:55 of the following video.