1983 a-more-rational-view-of-logic-or-up-against-the-wall-logic-i
The paper by Alex P. Pentland and Martin A. Fischler is a pointed rebuttal to Nils Nilsson’s AAAI President’s address, which advocated a ‘propositional doctrine’ that positions logic‑like propositional formalisms as the core of artificial intelligence. The authors argue that this view is an instance of ‘Logic Imperialism’ and that AI should instead be defined by its computational problems—reasoning, perception, language, and learning—regardless of the formal tools used. They propose a research paradigm where the problem dictates the representation (a “form follows function” approach) and emphasize that meaning (semantics) must be derived from the embedding of representations in the world, not from syntactic manipulation alone. Logic is acknowledged as a useful universal language but is critiqued as insufficient for general‑purpose reasoning, especially when evidence is incomplete or inconsistent. The paper introduces the concept of isomorphic representations—representations whose structure mirrors domain properties—and illustrates their advantage through examples such as a backpacker load‑division problem and mental imagery, showing how iterative, semantics‑preserving approximations succeed where pure deductive theorem‑proving fails. The authors conclude that while logic will continue to play a role, AI must prioritize rational (inductive) reasoning, semantics, and domain‑appropriate representations, and they issue a warning about misrepresentation of AAAI staff.
Topics
Logic Imperialism: Critique of the Propositional Doctrine
The authors dissect Nilsson’s claim that AI should be centered on propositional, logic‑like formalisms, labeling it as an over‑reaching ‘Logic Imperialism’ that marginalizes other AI research avenues.
Proposed AI Research Paradigm – Form Follows Function
A functional, meaning‑oriented paradigm where the computational problem (reasoning, perception, language, learning) dictates the choice of representation and tools, rather than letting logic dictate the problem.
Logic as a Tool, Not the Core Language of AI
Discussion of logic’s universality and adequacy for description, contrasted with its limited usefulness for many AI problems; logic is positioned as one of many possible universal formalisms.
Semantics and Isomorphic Representations
Argument that meaning arises from the embedding of symbols in the world; introduction of isomorphic representations that preserve domain topology (e.g., images, polynomial models) and enable effective reasoning.
Deductive vs. Inductive Reasoning in AI
Contrast between deductive theorem‑proving (requiring complete, consistent premises) and inductive/approximate reasoning needed for real‑world, noisy, or incomplete evidence.
Illustrative Case Studies: Backpacker Load Division and Mental Imagery
Concrete examples showing failure of pure theorem‑proving and success of isomorphism‑based, iterative approximations; mental imagery highlighted as iconic, isomorphic reasoning.
Conclusion and Call for Rational AI Research
Reaffirmation of AI’s core topics, emphasis on rational (inductive) reasoning, and a warning about misrepresentation of AAAI staff.