Hacker News Reader

any ideas on how this is similar to or differs from the prospects and structure of Darpa's explainable AI (XAI) contract?

https://www.fbo.gov/index?s=opportunity&mode=form&id=1606a25...

curious about the inherent trade-off between predictive power/complexity in ML model and the accuracy of system explanation's inferred by these models

View on HN

IIRC the last XAI approch I read about basically gives you a local linear approximation or Jacobian of the decision surface. You take a look at the biggest terms and call them an explanation. Not too shabby, probably. I don't know if all of the XAI project follows that approach.

Edit: Framing

Reply on news.ycombinator.com