Stockertim Dictionary · Technology · Human resources
Feedback loop
Also called self-reinforcing loop, runaway feedback loop
In short
A cycle in which a model's own past decisions become its next training data, so whatever pattern it started with grows stronger over time.
- Advanced
- Concept
- Checked
In detail
A feedback loop forms when a system's outputs shape the data it later learns from. In recruitment, a screening model decides who is interviewed; only interviewed people can be hired; only hired people produce performance data; and that data trains the next version of the model. People the model screened out never get the chance to show they would have done well, so the model never learns it was wrong about them.
The loop narrows the pool. If an early version slightly favoured one type of background, the next version sees even more successful people from that background and favours it more strongly. The same can happen in sourcing tools that learn from which profiles recruiters click, or in performance tools where flagged employees receive less support and then perform worse.
Breaking the loop takes deliberate design: keep a share of decisions independent of the model, for example by reviewing a random sample of screened-out applications; retrain on outcomes that are not purely the model's own choices; monitor selection rates by group over successive versions; and treat a steadily narrowing applicant mix as a warning sign.
Worked example
A made-up firm, Oakfield Tech, reviews a random 5% of the 2,000 applications its tool rejects each quarter.
- Applications reviewed: 2,000 × 0.05 = 100
- Found suitable on review: 12 / 100 = 0.12
12% of sampled rejections were suitable, evidence that the tool is missing good candidates and that retraining only on its own picks would deepen the gap.
In practice
Model owners include holdout or random-review samples in live processes and compare each new model version with the last on selection rates and outcomes by group before release.
Watch out
- Improving accuracy on the model's own picks can hide growing blind spots.
- Loops also form through people: managers who trust the score act on it, which shapes the next data.
Learn it properly
Official sources
- Towards a Standard for Identifying and Managing Bias in Artificial Intelligence, NIST SP 1270 · National Institute of Standards and Technology
- Artificial Intelligence Risk Management Framework (AI RMF 1.0), NIST AI 100-1 · National Institute of Standards and Technology
Facts last checked on 3 November 2026.
Education, not advice: the Stockertim Dictionary explains what terms mean. It never recommends buying or selling anything.