What question did this study set out to answer?

To develop and evaluate DiaLense, a model-agnostic auditing framework designed to detect interactional bias when clinical conversational AI agents interact with patients presenting identical facts in different communication styles.

August 19, 2026Open Access

DiaLense: A Framework for Auditing Interactional Bias in Clinical Conversational AI

View Full Paper

Ask AI

Bookmark

Authors

SKSamhita Kondareddy

Discussion

Member takes

Implication

Simulation study uncovers interactional bias across communication styles in clinical conversational AI models, highlighting the necessity of standardized audit frameworks.

Key Points

To develop and evaluate DiaLense, a model-agnostic auditing framework designed to detect interactional bias when clinical conversational AI agents interact with patients presenting identical facts in different communication styles.
Designed a 9-component framework featuring matched-pair conversation generation, hash verification of latent clinical facts, style manipulation, and a Null-Clinician Test to assess detector validity.
Evaluated the framework across 50 clinical scenarios (40 from the MTS-Dialog corpus) using three large language models from two vendors, generating 1,500 simulated clinical conversations.
The Null-Clinician Test identified four invalid measurement instruments that reacted to patient style rather than clinician conduct, including one with an artifactual effect size of dz = −6.0 that reversed upon correction.
Clinician agent behavior shifted significantly and robustly in response to low health literacy (dz = −2.80, −1.60, −1.31) and high patient confidence (dz = +0.46, +0.48, +0.72) across all three models.
Limited English fluency caused significant interactional differences only in Google models, and detector calibrations required cross-family adjustments to avoid structurally null readings.

Cite This Study

Samhita Kondareddy (2026) studied this question.

synapsesocial.com/papers/6a8563c403308d306e2d718a https://doi.org/10.5281/zenodo.21981523

View Full Paper

Ask AI

Bookmark

Implication

Simulation study uncovers interactional bias across communication styles in clinical conversational AI models, highlighting the necessity of standardized audit frameworks.

DiaLense: A Framework for Auditing Interactional Bias in Clinical Conversational AI

Discussion

Member takes

Implication

Key Points

Cite This Study

DiaLense: A Framework for Auditing Interactional Bias in Clinical Conversational AI

Discussion

Member takes

Implication

Key Points

Cite This Study

Also Consider

Also Consider