Research

Paper

AI LLM March 12, 2026

Managing Cognitive Bias in Human Labeling Operations for Rare-Event AI: Evidence from a Field Experiment

Authors

Gunnar P. Epping, Andrew Caplin, Erik Duhaime, William R. Holmes, Daniel Martin, Jennifer S. Trueblood

Abstract

Many operational AI systems depend on large-scale human annotation to detect rare but consequential events (e.g., fraud, defects, and medical abnormalities). When positives are rare, the prevalence effect induces systematic cognitive biases that inflate misses and can propagate through the AI lifecycle via biased training labels. We analyze prior experimental evidence and run a field experiment on DiagnosUs, a medical crowdsourcing platform, in which we hold the true prevalence in the unlabeled stream fixed (20% blasts) while varying (i) the prevalence of positives in the gold-standard feedback stream (20% vs. 50%) and (ii) the response interface (binary labels vs. elicited probabilities). We then post-process probabilistic labels using a linear-in-log-odds recalibration approach at the worker and crowd levels, and train convolutional neural networks on the resulting labels. Balanced feedback and probabilistic elicitation reduce rare-event misses, and pipeline-level recalibration substantially improves both classification performance and probabilistic calibration; these gains carry through to downstream CNN reliability out of sample.

Metadata

arXiv ID: 2603.11511
Provider: ARXIV
Primary Category: cs.HC
Published: 2026-03-12
Fetched: 2026-03-14 05:03

Related papers

Raw Data (Debug)
{
  "raw_xml": "<entry>\n    <id>http://arxiv.org/abs/2603.11511v1</id>\n    <title>Managing Cognitive Bias in Human Labeling Operations for Rare-Event AI: Evidence from a Field Experiment</title>\n    <updated>2026-03-12T03:55:31Z</updated>\n    <link href='https://arxiv.org/abs/2603.11511v1' rel='alternate' type='text/html'/>\n    <link href='https://arxiv.org/pdf/2603.11511v1' rel='related' title='pdf' type='application/pdf'/>\n    <summary>Many operational AI systems depend on large-scale human annotation to detect rare but consequential events (e.g., fraud, defects, and medical abnormalities). When positives are rare, the prevalence effect induces systematic cognitive biases that inflate misses and can propagate through the AI lifecycle via biased training labels. We analyze prior experimental evidence and run a field experiment on DiagnosUs, a medical crowdsourcing platform, in which we hold the true prevalence in the unlabeled stream fixed (20% blasts) while varying (i) the prevalence of positives in the gold-standard feedback stream (20% vs. 50%) and (ii) the response interface (binary labels vs. elicited probabilities). We then post-process probabilistic labels using a linear-in-log-odds recalibration approach at the worker and crowd levels, and train convolutional neural networks on the resulting labels. Balanced feedback and probabilistic elicitation reduce rare-event misses, and pipeline-level recalibration substantially improves both classification performance and probabilistic calibration; these gains carry through to downstream CNN reliability out of sample.</summary>\n    <category scheme='http://arxiv.org/schemas/atom' term='cs.HC'/>\n    <category scheme='http://arxiv.org/schemas/atom' term='econ.GN'/>\n    <published>2026-03-12T03:55:31Z</published>\n    <arxiv:primary_category term='cs.HC'/>\n    <author>\n      <name>Gunnar P. Epping</name>\n    </author>\n    <author>\n      <name>Andrew Caplin</name>\n    </author>\n    <author>\n      <name>Erik Duhaime</name>\n    </author>\n    <author>\n      <name>William R. Holmes</name>\n    </author>\n    <author>\n      <name>Daniel Martin</name>\n    </author>\n    <author>\n      <name>Jennifer S. Trueblood</name>\n    </author>\n  </entry>"
}