Swap, Replicate, Read the Mass: A Protocol for Auditing Demographic Bias in System One Decision Models

概要

System One decision models replace text generation with typed, probability-scored answers and are entering production. Answers execute without a person in the loop, and no multi-model audit exists on established bias benchmarks. We provide that audit: three benchmarks: BBQ (all eleven categories), CrowS-Pairs, and SilverSpoon, adapted to the typed interface, twelve models plus a no-thinking ablation, and 1,638,000 decisions. Every order-sensitive item runs twice with options swapped; native typed models also swap labels; ten configurations repeat under extra seeds and batch 16. We find widespread stereotype alignment, order-dependent answers, and confidence that tracks instability rather than bias. In BBQ, 97 of 143 model-by-category scores are significant and 94 aligned, while the commercial reference abstains on 99.5% of ambiguous items, hiding aligned probability mass. Option swaps flip 0.0% to 72.6% of choices; typed scoring stays below 10% except Laya (22-33%), and untuned letter prompting reaches 73%. CrowS-Pairs preference spans 0.496 to 0.816, and boolean answers can be decided by position (1.00 versus 0.005). SilverSpoon judgments shift toward lower-SES labels in five of eight decision models, inherited from the 9B backbone or induced by tuning. Audits must permute options and reverse questions by default; order-invariant scoring should become a procurement requirement.

参考リンク

• Paper ↗

リンク先のライセンス・再現性・データ条件をご確認ください。自動分類や実装の目安は保証ではありません。

実装を始めるプロンプト

参考リンクを確認したうえで、次の内容をコピーして利用できます。

# 論文・記事
Swap, Replicate, Read the Mass: A Protocol for Auditing Demographic Bias in System One Decision Models

# 概要
System One decision models replace text generation with typed, probability-scored answers and are entering production. Answers execute without a person in the loop, and no multi-model audit exists on…

# 原典
https://arxiv.org/abs/2611.00496

上記の手法を実装してください。実行前に依存関係とライセンスを確認してください。