causal

Models Drift and Develop Biases Over Time

Machine learning models are trained on a fixed dataset from a specific point in time, and as they absorb more data they can learn biases or behave badly, as the Microsoft Twitter bot did, and this appears in industrial models when initial human assumptions are reweighted.

causalpending

Speaker

Ash

Evidence Quote

“will some biases start coming out in that model and will it start doing crazy stuff?”

Source

Investing in Artificial Intelligence— Invest Like the Best
Created: 11/19/2026, 2:19:59 AM

Your notes

Loading notes...