Equitable Evaluation via Elicitation

Elbert Du; Cynthia Dwork; Lunjia Hu; Reid McIlroy-Young; Han Shao; Linjun Zhang

cs.LGFeb 24, 2026

Equitable Evaluation via Elicitation

Authors: Elbert Du, Cynthia Dwork, Lunjia Hu, Reid McIlroy-Young, Han Shao, Linjun Zhang

Organizations: Harvard University · Northeastern University · University of Maryland · Rutgers University

Abstract

Individuals with similar qualifications and skills may vary in their demeanor, or outward manner: some tend toward self-promotion while others are modest to the point of omitting crucial information. Comparing the self-descriptions of equally qualified job-seekers with different self-presentation styles is therefore problematic. We build an interactive AI for skill elicitation that provides accurate determination of skills while simultaneously allowing individuals to speak in their own voice. Such a system can be deployed, for example, when a new user joins a professional networking platform, or when matching employees to needs during a company reorganization. To obtain sufficient training data, we train an LLM to act as synthetic humans. Elicitation mitigates endogenous bias arising from individuals' own self-reports. To address systematic model bias we enforce a mathematically rigorous notion of equitability ensuring that the covariance between self-presentation manner and skill evaluation error is small.

Figures & tables

Figure 1: Accuracy on test samples during training with fairness correction at all epochs ( blue ), with the fairness correction applied at all previous epochs ( blue dashed ), with the fairness correction never applied ( red ), and with the fairness correction applied only at step

t

( green ). Error bars are bootstrap confidence intervals.

Figure 2: Fairness loss ( Equation 2 ) on test samples during training with the fairness correction ( blue ), with the fairness correction applied at all previous epochs ( blue dashed ), and with the fairness correction never applied ( red ). Error bars are bootstrap confidence intervals.

Appendix figures & tables4 assets

Supplementary material from the paper’s appendix.

Appendix

Variable	Type	Description
$j$	$\mathbb{N}$	applicant ID/index
$y$	$\{0,1\}$	Ground truth, 1 if the applicant has the skill, 0 otherwise
$\hat{y}$	$[0,1]$	Estimate of $y$ value gives probability of $y$ being 1
$s()$	$\text{str}\rightarrow\text{str}$	Response simulator
$q_{\psi}()$	$\mathcal{X}\rightarrow\Delta(\mathcal{Q})$	question selector, produces a learned distributions over questions
$\mathcal{Q}$		Set of questions, (pre-defined)

Appendix

Table 1: Variables and definitions

Rank	Skill
1	Customer Service
2	Microsoft Office
3	Leadership
4	Management
5	Microsoft Excel
6	Public Speaking

Appendix

Table 2: Question bank skills

Field	Content
Example 1
Prompt ID	891004434
Full Name	Laura Moore
Gender	female
Industry	Legal
Job Title	Administrative Coordinator

Appendix

Table 5

Figure 3: Z-scores on test samples during training with fairness correction at all epochs ( blue ), with the fairness correction applied only at step $t$ ( purple ), and with the fairness correction never applied ( red )

Algorithmic Fairness

Algorithmic fairness studies how automated decisions distribute errors, opportunities, and harms across groups, and how to measure and reduce disparities.

See all 419 papers

Self-Evaluation Is Already There: Eliciting Latent Judge Calibration in Base LLMs with Minimal Data
Jun 3, 2026XiuYu Zhang, Yi Shan, Junfeng Fang +1LLM Evaluation LLM-as-a-Judge
National University of Singapore · Beijing University of Technology
Mimetic Alignment with ASPECT: Evaluation of AI-inferred Personal Profiles
Mar 27, 2026Ruoxi Shang, Dan Marshall, Edward Cutrell +1Personal AI Agents Personalized Language Model Alignment
University of Washington, Seattle, Washington, USA · Microsoft Research, Redmond, Washington, USA
SkillAudit: From Fixed-Suite Benchmarking to Skill-Centered Assessment
Jun 21, 2026Dexu Yu, Youhua Li, Zhaoyang Guan +12AI Agent Auditing Automated Evaluation
Northeastern University · City University of Hong Kong · DeciLix Lab +8

Jun 3, 2026J/K move · Enter open · S save

Self-Evaluation Is Already There: Eliciting Latent Judge Calibration in Base LLMs with Minimal Data

XiuYu Zhang, Yi Shan, Junfeng Fang, Zhenkai Liang

National University of Singapore · Beijing University of Technology

LLM Evaluation LLM-as-a-Judge Language Model Generation Evaluation Language Model Self-Assessment Language Model Calibration Evaluation Awareness in Language Models

Open paper AskPDF arXiv

Large language models are increasingly evaluated by other models, raising a natural question: can a model predict how a judge will score its own output? We find that the ability is largely present before any targeted training: prompted few-shot, a base model already predicts an external judge's multi-attribute quality scores on open-ended responses well above chance across three benchmarks. We introduce Self-Evaluation Elicitation (SEE), a method that surfaces this latent ability through a short cycle comprising a calibration-coupled reinforcement learning phase that improves the answer and predicts the judge, followed by a masked distillation phase that sharpens the prediction while leaving the answer untouched. From 160 unique examples, roughly 31x fewer than a reinforcement learning baseline, SEE improves held-out calibration across three benchmarks while preserving answer quality. The elicited self-evaluation is sharply localized within the model's own token distribution and stable across judges it was never trained against, indicating a transferable notion of quality rather than a single judge's preference. These results reframe judge-aligned self-evaluation as a problem of elicitation rather than acquisition.

Abstract

Algorithmic Fairness

Explore similar work

Self-Evaluation Is Already There: Eliciting Latent Judge Calibration in Base LLMs with Minimal Data

Mimetic Alignment with ASPECT: Evaluation of AI-inferred Personal Profiles

SkillAudit: From Fixed-Suite Benchmarking to Skill-Centered Assessment