causal

Prediction Fits the Past But May Not Hold in the Future

A model can fit historical data well—correlations that 'predict' for groups—without being causal for any individual; machine learning is unable to make the prediction-versus-causation distinction, so even though it fits the past well it may fail to predict the future because the underlying correlations may not be sustained or should not be.

causalpending

Speaker

Russ Roberts

Evidence Quote

“It doesn't imply that this person in particular, when they go back into that neighborhood, will have that experience. Because there could be a correlation that's not causal.”

Source

Cathy O'Neil on Weapons of Math Destruction— EconTalk
Created: 6/13/2026, 12:25:42 AM

Your notes

Loading notes...