Knowledge MaxxingFeedback loop
A feedback loop occurs when a system's own past decisions influence the data it later learns from, potentially reinforcing and amplifying an early pattern over time.
Prerequisite: Distributional impact
What matters
Because the system's own outputs shape its future training data, an early bias can compound over repeated cycles rather than staying fixed at its original level.
Example
A predictive policing tool that sends more patrols to an area may generate more recorded incidents there simply from increased presence, reinforcing the model's original prediction.
Common misconception
A feedback loop is not a one-time bias baked in at training; it's an ongoing cycle where the system's own outputs continue to shape the data it learns from next.
Recall question
Why can a feedback loop cause an early bias to grow over time rather than stay constant?