Technology & Engineering
Artificial Intelligence
Systems that learn from data — the fastest-moving field in this index, and the one with the worst signal-to-noise ratio.
12 fields · 120 topics · 14 curated works
Fields within Artificial Intelligence
Machine Learning Foundations
14The statistical learning theory underneath everything else here.
Supervised Learning · Unsupervised Learning · Bias-Variance Tradeoff · Overfitting & Regularisation · Cross-Validation
Deep Learning
13Neural networks, and why depth turned out to matter so much.
Backpropagation · Multilayer Perceptrons · Convolutional Neural Networks · Recurrent Networks & LSTMs · Transformers & Attention
Natural Language Processing
11Getting machines to handle language, which turned out to be most of the problem.
Tokenisation · Word & Sentence Embeddings · Language Modelling · Machine Translation · Question Answering
Computer Vision
10Extracting structure from images.
Image Classification · Object Detection · Semantic Segmentation · Pose Estimation · Optical Flow
Reinforcement Learning
10Learning by acting, with delayed and sparse feedback.
Markov Decision Processes · Dynamic Programming Methods · Q-Learning · Policy Gradient Methods · Actor-Critic Algorithms
Generative Models
9Models that produce samples rather than labels.
Autoencoders & VAEs · Generative Adversarial Networks · Diffusion Models · Autoregressive Models · Normalising Flows
AI Alignment & Safety
10Whether these systems do what we intend — the field's most consequential open problem.
The Alignment Problem · Specification Gaming · Reward Hacking · Interpretability for Safety · Scalable Oversight
Probabilistic & Bayesian Methods
8Learning with explicit uncertainty.
Graphical Models · Bayesian Networks · Hidden Markov Models · Gaussian Processes · Variational Methods
Machine Learning Systems
10The engineering that turns a model into a product.
Training Infrastructure · Distributed Training · Model Serving & Inference · Quantisation & Distillation · Data Pipelines
Classical & Symbolic AI
8The half-century of AI that came before deep learning, much of it still useful.
Search Algorithms · Constraint Satisfaction · Knowledge Representation · Expert Systems · Automated Planning
AI Ethics, Policy & Society
9Who is accountable when the model is wrong.
Algorithmic Bias & Fairness · Transparency & Explainability · Privacy & Data Rights · Labour Market Effects · Autonomous Weapons
History & Philosophy of AI
8Recurring hype cycles, and the questions that never resolved.
The Dartmouth Conference · The AI Winters · The Turing Test · The Chinese Room Argument · Symbolic vs Connectionist Debate
Reading in Artificial Intelligence
14Start here
No prior grounding assumed.
- PaperComputing Machinery and IntelligenceAlan Turing, 1950· 28 pages
Replaces 'can machines think' with an operational imitation game, and answers the standard objections in advance.
- EssayThe Bitter LessonRichard Sutton, 2019· ≈1,100 words· Free
General methods that leverage computation consistently beat hand-engineered human knowledge, and researchers keep relearning this.
Then
Assumes you know the vocabulary.
- PaperHidden Technical Debt in Machine Learning SystemsSculley et al., 2015
The model is a tiny box in a large diagram; almost all the cost and almost all the failures live in the plumbing around it.
- PaperConcrete Problems in AI SafetyAmodei et al., 2016· 29 pages· Free
Reframes AI safety as five tractable engineering problems rather than a speculative long-term concern.
- EssayComputer Science as Empirical Inquiry: Symbols and SearchAllen Newell & Herbert A. Simon, 1976
States the physical symbol system hypothesis — that symbol manipulation is necessary and sufficient for intelligence — the claim the field spent fifty years testing.
- PaperOn the Dangers of Stochastic ParrotsBender, Gebru, McMillan-Major & Mitchell, 2021· 14 pages
Argues that scaling language models compounds environmental cost, opaque training data and the illusion of understanding.
Go deeper
Primary sources and full treatments.
- BookSpeech and Language ProcessingDaniel Jurafsky & James H. Martin, 2000· Free
The standard NLP text across three decades of the field changing underneath it, which makes its revisions a history of the subject.
- PaperImageNet Classification with Deep Convolutional Neural NetworksKrizhevsky, Sutskever & Hinton, 2012· 9 pages
AlexNet's error rate on ImageNet made deep learning credible and redirected the entire field within a year.
- PaperAttention Is All You NeedVaswani et al., 2017· 15 pages· Free
Replaces recurrence with self-attention, producing the Transformer architecture that underpins essentially all current language models.
- BookReinforcement Learning: An IntroductionRichard Sutton & Andrew Barto, 1998· Free
The standard text: builds the whole field from the bandit problem up to function approximation and policy gradients.
- BookThe Elements of Statistical LearningTrevor Hastie, Robert Tibshirani & Jerome Friedman, 2001
The reference account of supervised and unsupervised learning as statistics, written before the field renamed itself.
- BookProbabilistic Graphical ModelsDaphne Koller & Nir Friedman, 2009
The comprehensive treatment of representing uncertainty as structure, which is what most machine learning quietly approximates.
- PaperGenerative Adversarial NetworksGoodfellow et al., 2014· 9 pages· Free
Trains a generator against a discriminator so that neither needs an explicit likelihood, and set the direction of generative modelling for a decade.
- BookDeep LearningGoodfellow, Bengio & Courville, 2016· Free
The reference textbook for the mathematical foundations underneath modern neural network practice.