Resources

A sprawling but essential reference. The handbook brings together philosophers, legal scholars, and computer scientists across 40+ chapters covering everything from algorithmic fairness to autonomous weapons. Not a cover-to-cover read, but invaluable for finding rigorous treatments of specific sub-topics. I keep returning to the chapters on moral agency and the ones on justice in automated decision-making.

The paper that launched a thousand debates — and cost Timnit Gebru her position at Google. Bender et al. argue that the rush to build ever-larger language models carries serious risks: environmental costs, encoded biases in training data, and the illusion of understanding where none exists. The “stochastic parrot” metaphor — a system that produces plausible-sounding text without comprehension — has become one of the most cited framings in AI ethics discourse. Required reading for anyone working with or writing about LLMs.

A landmark audit study that exposed dramatic accuracy disparities in commercial facial recognition systems. Buolamwini and Gebru found that these systems performed worst on darker-skinned women — an intersectional failure that single-axis analysis would miss entirely. The paper didn’t just document a problem; it established a methodology for algorithmic auditing that has since been widely adopted. It’s a model of how rigorous empirical work can drive both academic discourse and real-world policy change.

Mitchell et al. propose a standardised framework for documenting machine learning models — their intended use, performance across different demographic groups, and known limitations. Think of model cards as nutrition labels for AI: they don’t solve the problem, but they make the problem visible. The paper has had enormous practical influence; major tech companies now publish model cards, and the concept has been integrated into regulatory frameworks like the EU AI Act. A rare example of an academic paper that directly shaped industry practice.

The companion piece to Model Cards, but focused on the data side. Gebru et al. argue that every dataset used in machine learning should come with a “datasheet” documenting its motivation, composition, collection process, intended uses, and ethical considerations. The parallel to datasheets in electronics is apt — you wouldn’t use a component without knowing its specifications, so why would you train a model on data you haven’t characterised? This paper has become foundational in responsible AI practice and data governance curricula.

A sharp meta-analysis of AI ethics guidelines that asks an uncomfortable question: are these documents actually effective, or are they performative gestures? Hagendorff evaluates over 20 prominent AI ethics frameworks and finds significant gaps — particularly around issues of surveillance, labour rights, and environmental impact. The paper is a useful corrective for anyone tempted to assume that publishing principles is the same as practising them. I’ve returned to this paper repeatedly when thinking about the gap between ethics as aspiration and ethics as action.

A comprehensive database of AI ethics guidelines from around the world. Invaluable for understanding the landscape of soft governance — who’s saying what, and where the consensus lies (and doesn’t).

A searchable repository of real-world AI failures and harms. Every time someone argues that AI risks are theoretical, I point them here. The database documents hundreds of incidents — from facial recognition misidentifications to autonomous vehicle accidents to algorithmic discrimination in hiring. It’s a critical resource for grounding ethics discussions in reality rather than hypotheticals.

An independent research institute based in London, focused on ensuring data and AI work for people and society. Their reports on facial recognition, biometric data, and public attitudes toward AI are consistently rigorous and policy-relevant. I find their work bridges the gap between academic research and practical governance better than most. Their “AI Now” reports are particularly worth reading alongside the AI Now Institute’s work in the US.

A free, accessible introduction to AI ethics from the University of Helsinki — the same team behind the popular Elements of AI course. Covers key concepts like fairness, transparency, and accountability without requiring a technical background. An excellent starting point for anyone entering the field.