Sign In

Communications of the ACM

ACM News

Identifying Artificial Intelligence 'Blind Spots'


View as: Print Mobile App Share: Send by email Share on reddit Share on StumbleUpon Share on Hacker News Share on Tweeter Share on Facebook
AI blind spots

A model by MIT and Microsoft researchers identifies instances where autonomous systems have "learned" from training examples that don't match what's actually happening in the real world, which can be used to identify which learned actions could cause costly errors.

Credit: MIT News

A novel model developed by MIT and Microsoft researchers identifies instances in which autonomous systems have "learned" from training examples that don't match what's actually happening in the real world.

From MIT News
View Full Article

 


 

No entries found

Sign In for Full Access
» Forgot Password? » Create an ACM Web Account
ACM Resources