Skill UI

Provides interpretable explanations for ML predictions using SHAP, LIME, and other techniques, helping debug model behavior, highlight feature importance, and communicate fairness insights to stakeholders.

Assesses AI/ML models and datasets for ethical risks, fairness concerns, and bias, generating reports and mitigation guidance via the ai-ethics-validator plugin whenever a user requests an ethics review or bias detection.

Conducts a comprehensive, structured ethical review for any AI or ML model, feature, or product. It assesses risks related to fairness, bias, transparency, privacy, and societal impact, providing a risk tier score, necessary mitigations, and governance documentation suitable for pre-deployment compliance.

A specialized guide for authors targeting the ACM Conference on Fairness, Accountability, and Transparency (FAccT). It helps authors strategically frame their responsible AI research, ensuring claims are rigorously grounded in both technical measurement and social context. Covers necessary evidence bars, scope calibration, and submission best practices for this niche venue.

A comprehensive guide detailing best practices for designing and auditing empirical evaluations of Large Language Models (LLMs). It addresses critical pitfalls such as data contamination, baseline unfairness, model version drift, and decoding parameter sensitivity, ensuring that published results are robust, reproducible, and scientifically sound for academic publication.

A comprehensive guideline for researchers developing robot-learning papers. It details best practices for experimental design, emphasizing statistical rigor, handling stochasticity (training seeds, initial states), bridging the sim-to-real gap, and ensuring fairness across baseline policy families (BC/RL/VLA). This resource helps ensure claims are supported by credible evidence.

A comprehensive guide for designing, auditing, and executing rigorous experimental programs for top-tier computer vision papers (e.g., CVPR). It covers essential scientific components, including establishing fair benchmarks, structured ablation studies, documenting failure modes, quantifying computational efficiency, and demonstrating cross-dataset generalization, ensuring the research is reproducible and robust.

A comprehensive guide for structuring the empirical evidence section of top-tier NLP conference papers (e.g., EACL). It mandates rigorous controls, including fair baselines (tuned models, explicit LLM prompts), statistical significance testing (variance, CI), contamination checks, and detailed human evaluation protocols. Use this to elevate your research from mere assertion to scientifically audited proof.

A comprehensive guide for designing and auditing high-stakes computer vision experiments for top-tier conferences. It covers critical areas like 'matched-substrate fairness,' 'six-month-staleness testing,' and rigorous ablation studies, ensuring that claims are robust, reproducible, and withstand intense peer review scrutiny.

A comprehensive guide to preparing accountability artifacts for AI research papers, particularly for venues like FAccT. It covers the necessary documentation genres—including datasheets for datasets, model cards for models, and impact assessments—to ensure transparency, verifiability, and accountability regarding claims of fairness and harm mitigation in AI systems. The focus is on making research outputs rigorously traceable and ethically responsible.

This guide outlines the rigorous methodological standards for empirical work in AI fairness, accountability, and transparency (FAccT). It advises researchers on matching specific claims (e.g., disparate harm) to appropriate evidence types—including disaggregated metrics, coded qualitative data, and documented ethical procedures. It stresses the importance of ethical review, handling protected attributes, and acknowledging limitations to ensure high scientific and ethical rigor.

This guide helps researchers determine the most appropriate academic venue for their responsible AI work. It provides detailed criteria and rigorous testing methods (like the Delete-the-equity test) to assess whether fairness, accountability, or transparency is a primary, core contribution, guiding authors away from merely 'bolted-on' ethical framing.