Algorithmic Bias

Fairness in medical agents is becoming critical as tool-using clinical AI systems orchestrate specialized vision and language modules for tasks such as chest X-ray question answering. While these medical AI agents can improve flexibility, their added pipeline complexity also creates new pathways for demographic bias beyond standalone models. We present DUCK, Decomposing Unfairness in Chest X-ray agents, a systematic audit of fairness in tool-using chest X-ray agents instantiated with MedRAX. To localize where disparities arise, we introduce a stage-wise fairness decomposition that separates end-to-end bias from three agent-specific sources: tool exposure bias, or utility gaps conditioned on tool presence; tool transition bias, or subgroup differences in tool-routing patterns; and model reasoning bias, or subgroup differences in synthesis behaviors. Extensive experiments on tool-using agentic frameworks across five driver backbones reveal that demographic gaps persist in end-to-end performance, with equalized odds up to 20.79% and the lowest fairness-utility tradeoff down to 28.65%. Intermediate behaviors, including tool usage, transition patterns, and reasoning traces, exhibit distinct subgroup disparities that are not predictable from end-to-end evaluation alone. For example, conditioned on segmentation-tool availability, the subgroup utility gap reaches as high as 50%. Our findings underscore the need for process-level fairness auditing and debiasing to ensure the equitable deployment of clinical agentic systems. Code: https://github.com/Nanboy-Ronan/DUCK.

Large language models are increasingly used in decision-making tasks that require them to process information from a variety of sources, including both human experts and other algorithmic agents. How do LLMs weigh the information provided by these different sources? We consider the well-studied phenomenon of algorithm aversion, in which human decision-makers exhibit bias against predictions from algorithms. Drawing upon experimental paradigms from behavioural economics, we evaluate how eightdifferent LLMs delegate decision-making tasks when the delegatee is framed as a human expert or an algorithmic agent. To be inclusive of different evaluation formats, we conduct our study with two task presentations: stated preferences, modeled through direct queries about trust towards either agent, and revealed preferences, modeled through providing in-context examples of the performance of both agents. When prompted to rate the trustworthiness of human experts and algorithms across diverse tasks, LLMs give higher ratings to the human expert, which correlates with prior results from human respondents. However, when shown the performance of a human expert and an algorithm and asked to place an incentivized bet between the two, LLMs disproportionately choose the algorithm, even when it performs demonstrably worse. These discrepant results suggest that LLMs may encode inconsistent biases towards humans and algorithms, which need to be carefully considered when they are deployed in high-stakes scenarios. Furthermore, we discuss the sensitivity of LLMs to task presentation formats that should be broadly scrutinized in evaluation robustness for AI safety.

We study the implicit bias of momentum-based optimizers on smooth homogeneous models. We show that momentum steepest descent algorithms like Muon (spectral norm), MomentumGD ($\ell_2$ norm), and Signum ($\ell_\infty$ norm) are approximate steepest descent trajectories under a decaying learning rate schedule, proving that these algorithms have a bias towards KKT points of the corresponding margin maximization problem. We extend the analysis to Adam (without the stability constant), which maximizes the $\ell_\infty$ margin, and to Muon-Signum and Muon-Adam, which maximize a hybrid norm. Our experiments corroborate the theory and show that the identity of the margin maximized depends on the choice of optimizer. Overall, our results extend earlier lines of work on steepest descent in homogeneous models and momentum-based optimizers in linear models.

Instruction-guided image-to-image (I2I) editors are increasingly used in consumer and professional visual workflows, where trustworthiness depends not only on prompt compliance but also on equitable preservation of identity-relevant attributes. We formalize two failure modes: Soft Erasure, where requested edits are weakly realized or silently suppressed, and Stereotype Replacement, where edits introduce unrequested, stereotype-consistent demographic attributes. Using a controlled benchmark of 5,040 edited portraits, we evaluate these failures across three recent open-weight editors with vision-language model scoring and human evaluation. Our results show that identity-preservation failures are pervasive and demographically uneven. In particular, 62--71% of outputs exhibit skin lightening, with Indian and Black source portraits affected at 72--75%, compared with 44% for White source portraits, indicating output-level drift toward lighter or more White-presenting appearances when identity constraints are underspecified. In a mitigation case study, prompt-level appearance constraints reduce race-change scores for non-White source portraits by up to 1.48 points, while leaving White source portraits largely unchanged, without modifying model weights. These findings show that identity preservation is not a uniform property of I2I portrait editing systems, but an unevenly distributed trustworthiness failure with direct social consequences. At deployment scale, such silent distortions can shape AI-mediated self-representation and reinforce representational disparities. We introduce a controlled audit protocol for fairness-aware evaluation and governance of generative editing systems. Project page: https://seochan99.github.io/i2i-demographic-bias

Equity Bias is a philosophical and practical framework for building more equitable AI systems. It treats AI as an interpretive agent that constructs meaning and knowledge rather than as a neutral calculator. Grounded in Gadamerian hermeneutics and epistemic injustice theory, Equity Bias views bias not as an error to eliminate but as a reflection of whose knowledge is encoded into systems. Since all knowledge is situated, neutral systems are not possible. While traditional approaches aim to reduce or remove bias, Equity Bias instead makes bias transparent and contestable. Its three-stage AI Life Cycle (Equity Archaeology, Co-Creating Meaning, and Ongoing Accountability) turns these principles into practice. Equity Bias gives developers, researchers, policymakers, and communities affected by AI a method for examining whose knowledge a system encodes and who has the power to contest it. By placing the politics of knowledge at the centre of equitable AI design, Equity Bias offers a distinct approach to AI ethics.

Synthetic faces from text-to-image (T2I) models pervade digital media, yet their demographic biases under emotionally conditioned prompts remain poorly understood. We aim to systematically audit how emotionally conditioned prompts affect demographic and perceived-attractiveness biases in synthetic faces generated by T2I models, with particular attention to intersectional patterns and cross-ecosystem differences across model families. We audited eight (4 Western and 4 Chinese) T2I models and generated 56,000 faces under seven prompt conditions: a neutral baseline and six emotion conditions. We quantified biases in gender, race, age, and perceived attractiveness using information-theoretic divergence metrics. We further conducted intersectional analyses across combined demographic attributes and compared patterns between the Western and Chinese model groups to assess cross-ecosystem consistency and divergence in bias behavior. All models show strong overrepresentation of young faces, and most also overrepresent White-coded individuals. Intersectional analysis reveals compound underrepresentation or near-erasure of specific demographic combinations, such as young x female x Black faces, which are largely absent across models and are not captured by single-attribute audits. Emotion prompts act as additional demographic selectors: negatively valenced emotions (including sadness and fear) consistently shift outputs toward White, middle-aged, male-coded faces. This produces a valence-driven mapping that is also associated with lower perceived attractiveness in generated faces. These findings indicate that demographic bias in T2I face generation is both pervasive and shaped by emotional conditioning. They underscore the need for intersectional, emotion-conditioned, and multilingual demographic audits as part of standard pre-deployment evaluation practices.

Artificial Intelligence (AI) systems increasingly shape consequential decisions, creating value but also potential harms for individuals, social groups, and society. This has prompted calls for proactive approaches that anticipate harms early in the AI lifecycle. Although prior research identifies AI biases as sources of harm, the associations between particular lifecycle biases and harms remain insufficiently understood. We introduce ECHO, a systematic, context-sensitive, and participatory framework that anchors early harm anticipation in lifecycle biases and elicits their perceived associations with potential harms.ECHO identifies domain-specific stakeholders, instantiates biases through vignettes, collects harm judgements from human participants and a large language model (LLM), and organises them into descriptive and inferential ethical matrices. Applied to disease diagnosis and hiring, ECHO surfaced non-uniform, context-sensitive bias--harm patterns indicating which harms were perceived as plausible consequences of particular AI biases. The theoretical interpretability of these patterns and the inferential support for specific associations strengthen the plausibility of the mappings. By linking stakeholder-specific anticipated harms to lifecycle biases, ECHO supports source-level harm anticipation and provides structured input to subsequent AI governance actions

This PhD thesis investigates the societal impact of machine learning (ML). ML increasingly informs consequential decisions and recommendations, significantly affecting many aspects of our lives. As these data-driven systems are often developed without explicit fairness considerations, they carry the risk of discriminatory effects. The contributions in this thesis enable more appropriate measurement of fairness in ML systems, systematic decomposition of ML systems to anticipate bias dynamics, and effective interventions that reduce algorithmic discrimination while maintaining system utility. I conclude by discussing ongoing challenges and future research directions as ML systems, including generative artificial intelligence, become increasingly integrated into society. This work offers a foundation for ensuring that ML's societal impact aligns with broader social values.

We treat the internals of generative models as mechanistic objects rather than black boxes. We introduce Attribution Graphs (AGs), which extend GradCAM++ to circuit-level representations, and Causal Probing, a do-calculus intervention method for identifying causal latent structures, enabling detection and correction of spurious correlations, demographic biases, and misaligned decision circuits during training. We further propose the Cognitive Alignment Score (CAS), quantifying agreement between model-internal representations and human concepts, a saliency-first privacy mechanism sharing only thresholded attribution nodes, a bias-aware regularizer aligning subgroup statistics, and a Reveal-to-Revise loop integrating attribution signals into parameter updates without separate fine-tuning. Evaluated on CelebA, FairFace, Jigsaw, and HateXplain, our method achieves 94.1% accuracy, 92.3% macro F1, 79.4% IoU-XAI, and 12.7 FID at 72--76% adversarial robustness, while reducing subgroup disparity $Δ_{\mathrm{bias}}$ by 41%, demonstrating that mechanistic interpretability, fairness, and generative performance can be jointly optimized.

Uncertainty in artificial intelligence (AI) predictions raises pressing legal and ethical questions for AI-assisted decision-making. This article examines two uncertainty-based algorithmic interventions that act as guardrails for human-AI interaction: selective abstention, which withholds high-uncertainty predictions from human decision-makers, and selective friction, which presents such predictions together with salient warnings about the model's uncertainty. Prior work suggests that uncertainty-based abstention can exacerbate disparities where under-represented groups are more likely to receive uncertain predictions. We provide, to our knowledge, the first doctrinal analysis of uncertainty-based algorithmic interventions under laws from the United Kingdom and examine their consequences through two AI-assisted case studies: consumer credit and risk of reoffending. We show that the use of uncertainty thresholds, though formally neutral, can generate discriminatory effects. We argue that both interventions pose risks of unlawful discrimination, but that selective friction is legally preferable. It preserves access to the prediction and is more likely to satisfy proportionality under the Equality Act 2010. Whether selective friction also improves decision quality in practice is uncertain. We identify conditions under which it may improve or worsen decision quality.

Despite growing empirical evidence of bias amplification in machine learning, its theoretical foundations remain poorly understood. We develop a formal framework for majority-minority learning tasks, showing how standard training can favor majority groups and produce stereotypical predictors that neglect minority-specific features. Assuming population and variance imbalance, our analysis reveals three key findings: (i) the close proximity between ``full-data'' and stereotypical predictors, (ii) the dominance of a region where training the entire model tends to merely learn the majority traits, and (iii) a lower bound on the additional training required. Our results are illustrated through experiments in deep learning for tabular and image classification tasks.