The half-century of AI that came before deep learning, much of it still useful.
9 topics · 10 curated works
Assumes you know the vocabulary.
The Art of Artificial Intelligence: Themes and Case Studies of Knowledge Engineering
Edward A. Feigenbaum · 1977
Argues an expert system's power comes from the volume of domain-specific knowledge it encodes rather than the sophistication of its inference engine,…
+3 more at this level
Primary sources and full treatments.
A Machine-Oriented Logic Based on the Resolution Principle
J. A. Robinson · 1965
Introduces resolution and unification as a single inference rule complete for first-order logic, replacing the many special-case rules earlier…
+5 more at this level
10 works
Edward A. Feigenbaum
Argues an expert system's power comes from the volume of domain-specific knowledge it encodes rather than the sophistication of its inference engine, coining 'knowledge engineering' for the work of extracting that knowledge.
link checked 17 Sept 2026Notes on this copyThomas R. Gruber
Defines an ontology as an explicit specification of a conceptualisation, and argues that portability between knowledge-based systems requires agreeing on that specification independent of any one representation language.
link checked 17 Sept 2026Notes on this copyPatrick Winston (MIT OpenCourseWare)
The symbolic AI curriculum as it stood before deep learning displaced it — search, constraints, representation and learning — and still the clearest account of why representation is the hard part.
link checked 17 Sept 2026Notes on this copyArtur d'Avila Garcez & Luis C. Lamb
Argues neural learning and symbolic reasoning solve complementary problems, and surveys the systems trying to combine them rather than treating the connectionist-symbolic debate as settled in either direction.
link checked 17 Sept 2026Notes on this copyJ. A. Robinson
Introduces resolution and unification as a single inference rule complete for first-order logic, replacing the many special-case rules earlier automated theorem provers needed.
link checked 17 Sept 2026Notes on this copyJohn McCarthy & Patrick J. Hayes
Sets out the situation calculus for reasoning about action, and names the frame problem — stating everything an action does not change — that logic-based AI never solved cleanly.
link checked 17 Sept 2026Notes on this copyRichard E. Fikes & Nils J. Nilsson
Represents planning states and operators so that theorem proving can generate plans directly, introducing the add/delete-list action representation that automated planners still use as a baseline.
link checked 17 Sept 2026Notes on this copyDonald E. Knuth & Ronald W. Moore
Proves alpha-beta pruning explores the minimum possible number of nodes to find a game tree's minimax value under best-case move ordering, and analyses how far real move ordering falls short of that bound.
link checked 17 Sept 2026Notes on this copyAlan K. Mackworth
Defines node, arc and path consistency for constraint networks and gives the AC-3 algorithm, turning constraint propagation into a preprocessing step that shrinks search before it starts.
link checked 17 Sept 2026Notes on this copyMao et al.
Learns visual concepts and word meanings jointly from images and question-answer pairs by compiling questions into symbolic programs that execute over neural scene representations, without ever being told what an object or colour is.
link checked 17 Sept 2026Notes on this copy