Textbook of AI

The Mechanisation of Thought Processes symposium was held at the National Physical Laboratory in Teddington, England, on 24–27 November 1958. Sometimes called "the British Dartmouth", it brought together a wide cross-section of the world's pioneers of artificial intelligence and machine intelligence and produced a two-volume HM Stationery Office proceedings (1959) that contains several foundational papers of the field.

Context

The NPL had a deep history with computation, Alan Turing's ACE design had been done there a decade earlier, and by the mid-1950s the laboratory was a natural British centre for early AI. The 1956 Dartmouth workshop had taken place two years earlier in the United States; the NPL symposium consolidated the cross-Atlantic AI community in the late 1950s and was deliberately international. American, British, Israeli, and continental European participants attended.

Notable papers

The proceedings include several papers that are now landmarks:

• Oliver Selfridge, "Pandemonium: a paradigm for learning". Selfridge's Pandemonium model proposed pattern recognition as a hierarchy of "demons", feature detectors that "shout" with intensity proportional to how well their feature matches the input, with higher-level demons combining the shouts of lower ones. It is the most explicit early statement of distributed pattern recognition and an ancestor of modern hierarchical feature learning in convolutional networks.
• John McCarthy, "Programs with Common Sense". McCarthy introduced the Advice Taker thought experiment: a system that could be told facts and rules in declarative form and would reason about them logically. The paper became the foundational document of declarative, knowledge-based AI and shaped logic programming, expert systems, and the entire knowledge-representation tradition.
• Marvin Minsky, "Some methods of heuristic programming and artificial intelligence". Minsky surveyed heuristic-driven theorem-proving and search, prefiguring the later "heuristic search" school.
• Contributions from Yehoshua Bar-Hillel (machine translation, including his later-famous critique of fully-automatic high-quality translation), Warren McCulloch (on neural mechanisms), Donald MacKay, Albert Uttley, and Gordon Pask (on self-organising systems).

Legacy

Because Selfridge's Pandemonium paper, McCarthy's Programs with Common Sense, and Minsky's heuristic-programming survey all appear together in the proceedings, the NPL volume is one of the most concentrated single sources of foundational AI ideas. The two volumes (the second a companion of discussions) are now collector's items but have been digitised in part. The symposium also marks an important moment in British AI, Britain was, in 1958, very much a co-equal partner in the field, before the Lighthill Report (1973) and the consequent funding collapse hollowed out much of the British AI research base for over a decade.