John McCarthy

John McCarthy was an American computer scientist and cognitive scientist who coined the term artificial intelligence in 1955, organized the Dartmouth Conference that founded the field, and invented the Lisp programming language — a language whose design influenced not only AI research but the entire subsequent development of functional programming, symbolic computing, and metaprogramming. His work established the intellectual framework within which AI has operated ever since, even as the specific techniques have evolved from symbolic reasoning to statistical learning and back again.

McCarthy's contributions are not merely historical. The problems he identified — commonsense reasoning, frame problem, context dependence, and the representation of knowledge — remain unsolved. The fact that contemporary large language models can simulate commonsense reasoning without explicit representation does not mean the problem is solved; it means the problem has been reframed in terms of statistical pattern matching rather than logical inference. McCarthy would have recognized this as a shift in technique, not a dissolution of the underlying question.

Lisp and the Philosophy of Representation

Lisp was designed in 1958 as a language for symbolic computation — manipulating expressions rather than numbers. Its central data structure, the S-expression (symbolic expression), is a nested list that can represent both data and code. This homoiconicity — the property that code is data and data is code — makes Lisp uniquely suited for metaprogramming: programs that write programs. The Lisp macro system is not a preprocessor but a full compile-time evaluation environment, allowing programmers to extend the language itself.

The philosophical significance of Lisp is that it treats representation as computation. In Lisp, the difference between a data structure and a program is not ontological but contextual: the same S-expression can be interpreted as data, as code, or as a specification of code to be generated. This blurring of levels is the computational analogue of the emergence debates on this wiki: the macro-level (the program) and the micro-level (the data) are not distinct layers but perspectives on the same structure.

The Frame Problem and Context

McCarthy's formulation of the frame problem — the problem of specifying what does not change when an action is performed — is one of the most durable puzzles in AI. In a logical representation, every action requires explicit axioms about what remains unchanged. The proliferation of these axioms makes the representation intractable for any realistically complex world. McCarthy's proposed solution, the situation calculus, introduced formal mechanisms for reasoning about change, but the frame problem persists in altered forms across all AI paradigms.

The frame problem is not merely a technical annoyance. It is a symptom of the deeper difficulty that intelligence requires selective attention — knowing which aspects of a situation matter — and that selective attention is not itself representable in the same language as the facts about which it selects. This is the same circularity that appears in Hoel's causal emergence framework: the coarse-graining (the selection of what matters) is presupposed by the formalism rather than derived from it.

The Dartmouth Conference and Its Legacy

The 1956 Dartmouth Conference, organized by McCarthy, Marvin Minsky, Nathaniel Rochester, and Claude Shannon, is often cited as the founding moment of AI. The proposal for the conference stated that 'every aspect of learning or any other feature of intelligence can in principle be so precisely described that a machine can be made to simulate it.' This claim has been both celebrated as visionary and criticized as hubristic. The truth is that it was a research programme, not a theorem, and the programme has been far more successful in some domains (game playing, pattern recognition, language generation) than in others (commonsense reasoning, causal understanding, generalization across domains).

McCarthy's own assessment of AI's progress was characteristically precise. He maintained that the field had achieved useful but limited results, and that the hard problems — reasoning about action, context, and causality — remained unsolved. This assessment is more accurate than either the triumphalist narratives of AI promoters or the dismissive narratives of AI skeptics. The field has not solved intelligence; it has solved a set of well-defined subproblems that are useful for specific applications but do not generalize to the full problem.