This research formalizes bias in machine learning systems as a symmetry-breaking operation, defining fairness as invariance under counterfactual sensitive attribute switching. It implements loss-based regularization as a symmetry-restoring mechanism, achieving over 90% bias violation reduction with approximately 5% accuracy cost.
This research paper explores the disconnect between fair outputs of language models and their latent internal biases in high-stakes decisions like mortgage underwriting. It demonstrates that while LLMs may show no output bias, they retain and amplify demographic representations which can cause decision reversals, and this bias is asymmetric.
This article emphasizes the non-optional nature of responsible AI development due to its impact on decisions and emerging regulations. It provides practical techniques, including code examples for quantifying bias using standard fairness metrics, to build responsible AI applications.
This research investigates the trustworthiness and fairness of nonparametric deep survival models for analyzing Alzheimer's Disease (AD) progression. It addresses the lack of studies considering learned bias in existing deep learning models for AD and proposes novel fairness metrics to ensure reliable predictions.
FASE is a Fairness-Aware Spatiotemporal Event Graph framework designed to integrate crime prediction with fairness-constrained patrol allocation to mitigate racial disparities in predictive policing. It utilizes a spatiotemporal graph neural network and a multivariate Hawkes process to model crime incidents in Baltimore, addressing data bias through a closed-loop deployment simulator.
This paper investigates counterfactual unfairness in LLMs by analyzing how their responses to humor change when swapping speaker and addressee identities. Experiments reveal consistent relational disparities, where jokes told by privileged speakers are more often refused or judged as malicious by the models.
A survey on gender bias in Natural Language Processing analyzes how gender stereotypes are perpetuated in AI models. The study discusses methods to mitigate these biases and explores challenges in creating more equitable NLP systems.
Pesquisadores introduziram AttriBench, um novo dataset demograficamente balanceado para atribuição de citações. Testes em 11 LLMs revelaram disparidades sistemáticas de precisão baseadas em raça e gênero, expondo um viés de atribuição e uma nova métrica de justiça na IA.
This article argues that fairness evaluation, specifically demographic parity, is a critical and non-negotiable deployment constraint for AI systems in child safety. Disproportionate flagging harms users, creates legal risks, and undermines trust, while also missing threats in underrepresented groups due to biased datasets.
This paper proposes a lightweight, inference-time framework to mitigate demographic representational bias in text-to-image models like Stable Diffusion, without requiring model retraining. The approach allows users to select their own fairness specifications to generate more equitable outputs across professions.
The `FairMind` software prototype automates causal fairness analysis at the dataset level, addressing the lack of fairness consideration in most AutoML frameworks. It utilizes the standard fairness model and LLMs to generate accurate reports on fairness based on counterfactual causal effects.
A new multi-domain red teaming framework was developed to evaluate the safety, robustness, and fairness of medical Large Language Models (LLMs) across 690 clinically grounded scenarios. The research revealed substantial performance variance and critical failures in safety-critical scenarios, even in high-performing systems.
The rapid development of Language Diffusion Models (LDMs) introduces new trustworthiness challenges. TrustLDM is a comprehensive benchmark evaluating safety, privacy, and fairness in LDMs, revealing that their alignment behavior degrades noticeably with malicious contexts, regardless of context length.
Este trabalho propõe um novo modelo para treinar Redes Neurais Gráficas (GNNs) sensíveis à justiça, aprimorando o framework CAF. A abordagem utiliza uma estratégia de treinamento em duas fases, editando o grafo para ajustar a homofilia e integrando perdas contrastivas e ambientais modificadas para melhorar a predição e a justiça.
The article discusses the complex issue of accountability when AI makes mistakes, illustrating how blame fragments among various stakeholders like banks, vendors, and data providers. It defines Responsible AI as the practice of building and deploying AI systems that are fair, transparent, safe, and accountable.