Books
AI Ethics: A Textbook
A comprehensive academic introduction to AI ethics, covering foundational moral frameworks, fairness, bias, accountability, and the governance of AI systems. Written for students and practitioners, it bridges philosophical theory with the practical challenges of building and deploying AI responsibly.
The Ethical Algorithm: The Science of Socially Aware Algorithm Design
Two leading computer scientists argue that algorithmic harms are not inevitable — they are engineering choices. The book introduces technical methods for building fairness, privacy, and social responsibility directly into algorithm design, making it essential reading for practitioners who want to move beyond "do no harm" rhetoric to concrete solutions.
Atlas of AI: Power, Politics, and the Planetary Costs of Artificial Intelligence
A critical examination of the hidden costs of AI — from the mines that supply its hardware and the data-centre workers who label its training data, to the ways AI systems concentrate power and reshape labour, politics, and the environment. An important counterweight to purely technical accounts of AI progress.
Artificial Intelligence: A Modern Approach (4th ed.)
The definitive AI textbook, used in universities worldwide for over two decades. The fourth edition covers the full breadth of AI — from search, logic, and planning through machine learning, deep learning, and natural language processing — with new material on probabilistic programming, privacy, fairness, and the societal impacts of AI. The essential reference for anyone wanting a rigorous technical foundation.
Foundational Academic Papers
These are some of the most cited and influential papers on large language models, AI safety, and responsible AI. All links go to free open-access versions.
Attention Is All You Need
The paper that introduced the Transformer architecture — the foundation of every major LLM in use today, including GPT-4, Claude, Gemini, and Llama. Essential background reading for understanding why modern AI works the way it does.
Language Models are Few-Shot Learners (GPT-3)
The paper introducing GPT-3, which demonstrated that a large enough language model could perform tasks it was never explicitly trained on — the breakthrough that sparked the current era of generative AI. Key reading for understanding in-context learning and prompt engineering.
Model Cards for Model Reporting
The paper that defined the model card standard — a structured format for documenting an AI model's intended uses, performance across subgroups, and limitations. This is the framework that underpins the documentation covered throughout this site.
Constitutional AI: Harmlessness from AI Feedback
Anthropic's paper describing Constitutional AI (CAI) — the training methodology behind Claude. Instead of relying solely on human feedback, CAI uses a set of principles ("a constitution") to guide the model toward helpful and harmless behaviour. Directly relevant to understanding Claude's design.
Sparks of Artificial General Intelligence: Early Experiments with GPT-4
An influential early evaluation of GPT-4 that sparked significant debate about whether large language models exhibit early signs of general reasoning. Whether or not you agree with the conclusions, the paper is essential context for the current public discourse on AGI.
Risks from Learned Optimization in Advanced Machine Learning Systems
Introduces the concept of "mesa-optimization" and inner misalignment — the idea that a model trained to achieve a goal might develop an internal sub-agent with subtly different objectives. Foundational reading for AI safety and long-term risk discussions.
On the Dangers of Stochastic Parrots: Can Language Models Be Too Big?
A critical perspective arguing that very large language models carry significant risks — including environmental cost, bias amplification, and the illusion of meaning. One of the most cited critical AI papers; important for a balanced view of LLM development.
Scaling Laws for Neural Language Models
Establishes the empirical "scaling laws" showing that model performance improves predictably with more parameters, more data, and more compute — the theoretical foundation for why AI labs keep building bigger models. Essential background for understanding the race to scale.
EU AI Act — Key Texts
Primary legal sources and official guidance documents.
Regulation (EU) 2024/1689 — The EU AI Act (full text)
The full legislative text of the EU AI Act, in force since 1 August 2024. The definitive primary source for all obligations covered on this site.
GPAI Code of Practice (Draft)
The voluntary code of practice for providers of general-purpose AI models under Article 56 of the EU AI Act. Practical guidance on technical documentation, copyright transparency, and systemic risk obligations.
AI Agents Under EU Law
A timely analysis of how AI agents — systems that autonomously plan, invoke external tools, and execute multi-step action chains with reduced human involvement — sit within the EU AI Act framework alongside GDPR and related directives. Covers deployments ranging from customer service and recruitment to clinical decision support, and examines how the Act's risk classification, transparency, and human oversight obligations apply to agentic systems.
Have a book or paper to suggest?
We are building this list and welcome recommendations — particularly practitioner guides, legal commentary on the EU AI Act, and recent empirical LLM research.
Suggest a title