Existing studies of LLM-as-judge scoring bias work predominantly at the input-output level: they perturb inputs, measure score deltas, and propose prompt-level mitigations. We argue that the same biases admit a representation-level account in the judge's hidden state, complementary to the input-output view and operationally useful in ways it does not afford. We report three findings, across seven judges, seven bias types, and nine benchmarks. Geometry: baseline judging inputs occupy a tight activation manifold while biased inputs are displaced along a low-dimensional, type-specific subspace that sharpens with depth and is recovered consistently by three families of estimators. Causal control: steering hidden states along this subspace drives scoring in both directions, forward shifts reproducing biased scoring on clean inputs and reverse shifts restoring baseline scoring on biased ones, while matched-norm random directions produce shifts an order of magnitude smaller. Operational: a simple linear projection onto the same bias-direction features anticipates judge failures on three entirely unseen benchmarks, substantially outperforming text-based alternatives. Reading bias as activation geometry, rather than as input-output noise, unifies geometric structure, causal control, and operational prediction within a single framework. The project page is available at https://xzx34.github.io/unfair-judge/
Existing bias auditing methods typically rely on model outputs, requiring costly benchmarks or judge models and potentially missing internal shifts that never appear in generated text. We propose a reference-based method that audits bias in hidden-state representations across related model variants, for example before and after fine-tuning. Because fine-tuning reshapes representation geometry, absolute hidden states are not directly comparable, so we encode each sentence by its similarities to a fixed set of anchor sentences, yielding relative representations in a shared comparison space. There we measure how target groups shift in their association with positive and negative attributes, a quantity we call the Representational Bias Shift $\Delta B$. Across three model families and the WildGuardMix, DecodingTrust and ToxiGen benchmarks, $\Delta B$ correlates with output-level bias change in 15 of the 18 settings we test, reaching $|r| = 0.84$ ($p<0.001$) under full fine-tuning and becoming more model-dependent under parameter-efficient adaptation. Thresholding $\Delta B$ detects checkpoints whose bias increased with ROC AUC between $0.65$ and $0.99$, and on WildGuardMix and DecodingTrust it separates them better than a SEAT-based baseline for all three families. $\Delta B$ is also stable under changes to the anchor set, attribute sets and target templates. Our method requires no task-specific evaluation data and audits a model in about three minutes, using $3$-$50\times$ less compute than the output-level benchmarks considered here. We view it as complementary to output-based auditing rather than a replacement for it.
M. Jeliński, Jan Dubinski, Maciej Chrabaszcz et al.· 0 citations
Large language models are increasingly used as judges to measure social bias in text, yet the passages they judge are often noisy, containing typos, informal spelling, and broken punctuation. The consequences of such surface noise for social bias measurement remain unclear. To investigate this question, we apply five realistic noise conditions at multiple intensity levels to 3,822 stereotype-related responses and compare the resulting bias judgments with those on the original text. We find that such surface noise does not degrade bias measurement symmetrically: it is far more likely to turn neutral judgments into biased ones than biased judgments into neutral ones, by up to a 120x margin. We further observe two non-obvious effects across four LLM judges: in the most fragile judge the distortion is at its purest at mild, realistic noise levels, where erasure is scarcest, and as judges grow robust it attenuates toward parity rather than reversing. Bias measured on noisy text is therefore systematically overestimated, most in the categories that matter most for fairness.
DongHyun Ryu, Jaehyeok Lee, Yeongjun Hwang et al.· 0 citations
This work introduces OptimismBench, which detects directional bias with inverted pairs: each scenario elicits both P(success) and P(failure), and asymmetry between the two framings yields a signed bias score without ground truth.
Seonglae Cho, A. Koshiyama· arXiv.org· 0 citations
The inverse base-rate effect is a robust bias in how people resolve ambiguity between competing categories, and the most prominent theories explain it through prediction error. Across two experiments we progressively removed the elements of the predictive-learning design that supply such error signals: first by moving to observational learning, then to an unsupervised procedure in which category labels were not presented. The effect persisted--the irrational bias is independent of supervised learning procedures. We propose a new theory, OSCAR, that integrates core computational principles of the best-validated models and operates on self-generated feedback akin to pattern completion. OSCAR extends the learning dynamics underlying the response bias to observational and unsupervised procedures. Evaluated on a large preexisting supervised dataset in addition to the two new experiments reported here, OSCAR performs competitively against alternatives, and is the first model that reproduces the pattern of individual differences seen in humans across all three procedures. The model provides an explanation of hitherto unexplained eye-tracking data, something none of the alternative accounts provide.
It is shown the natural way to do this does not work, specify one that survives measurement, then finds that the correction making it work carries more variance than the null it is tested against, and that the correction making it work carries more variance than the null it is tested against.
It is shown that prior scores, even when included only as context metadata, anchor judgments and systematically shift ratings toward their values, and effective mitigation must be validated for the intended model and task or domain.
A. Kapetanović, Kemal Altwlkany, Andro Merćep et al.· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
