AI Paper Atlas

Research map / Large language models / Language, safety and interpretability

Ecosystem-level Analysis of Deployed Machine Learning Reveals Homogeneous Outcomes

Connor Toups, Rishi Bommasani, Kathleen Creel, Sarah H Bana, Dan Jurafsky, Percy Liang

NeurIPS 2023 · Large language models › Language, safety and interpretability · approach: Bias measurement and mitigation · 3 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
Evaluate the ecosystem-wide distribution of outcomes against an independence baseline rather than only individual model accuracyCommercial ML APIs across 11 HAPI datasets · HAPIsystemic and consistent-success outcomes relative to baselineMixed
Improve one commercial APICommercial ML APIs across HAPI datasets · HAPI, 2020–2022improvements on systemic failures relative to potential improvements: just 10%; 27%Hurt
Improve Amazon's sentiment APIAmazon sentiment API · Waimai, 2020–2021net systemic failures: 2.5%; 78; 78No clear effect
Use gross improvements instead of net improvements (ablation)Year-over-year API improvements · HAPIprogress on systemic failuresNo clear effect
Vary the threshold for including model improvements (ablation)Year-over-year API improvements · HAPIunder-improvement on systemic failures: 6 different thresholdsNo clear effect
Include HAM10k predictions (ablation)Dermatology model ecosystem · DDIhomogeneous outcomesHurt
Include HAM10k predictions (ablation)Dermatology model ecosystem · DDI, skin-tone disparityracial disparity in model homogeneityHurt
Stratify FER+ examples by annotator disagreement (ablation)Three commercial FER APIs · FER+homogeneous outcomes relative to baselineNo clear effect
Stratify DDI examples by dermatologist accuracy (ablation)Dermatology models · DDI, instances both dermatologists classify correctlyhomogeneity relative to baselineNo clear effect
Stratify DDI examples by dermatologist accuracy (ablation)Dermatology models · DDI, instances both dermatologists misclassifyhomogeneous outcomes relative to baselineImproved
Replace the instance-level independence baseline with a shared hard/easy-instance baseline (ablation)Theoretical ecosystem outcome model · HAPIfit to observed outcome distributions: 0.2-0.3; 1-4Improved
Examine ecosystems of human dermatologists as well as modelsTwo board-certified dermatologists · DDIhomogeneity relative to model outcomesMixed

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy