Steering Smart Software
May 6, 2022
Why is smart software in need of direction? The short answer is that drift fouls up the machinery. If the real world conformed to the lab, life would be easier.
To get a sense for the “opportunities” of steering, take a look at “Cross-Ethnicity/Race Generalization Failure of Behavioral Prediction from Resting-State Functional Connectivity.” Here’s a quick summary: AI makes errors.
For a glimpse of Google’s awareness of this “minor” problem, navigate to “A Loss Curvature Perspective On Training Instability in Deep Learning.” Here’s a quick summary: AI makes errors and we think we can automate around these.
Net net: Certain applications of machine learning to smart software work okay; for example, sorting bruised fruit and ad matching. Other applications? Not so much.
Stephen E Arnold, May 6, 2022
Comments
-
- Subscribe to Beyond Search
Feature archive
News archive
-
