This is the repository of articles written by John.
A position paper about the relations among artificial intelligence (AI), mathematical logic and the formalization of common-sense knowledge and reasoning.
'Artificial Intelligence and Philosophy' was given at Aaron Sloman's Symposium on philosophy and AI at IJCAI-95. The present version is somewhat improved.
Artificial intelligence (AI) has closer scientific connections with philosophy than do other sciences, because AI shares many concepts with philosophy, e.g. action, consciousness, epistemology (what it is sensible to say about the world), and even free will. This article treats the philosophy of AI but also analyzes some concepts common to philosophy and AI from the standpoint of AI.
This article describes how to express programs with assignment statements and conditional GOTOs in mathematical logic without any programming constructs used in Fortran and Algol.
Concepts that are ill-defined in general can be precise in limited contexts. We propose applications for this idea in AI and in philosophy.
Machine learning research has had its goal the discovery of relations among observations. This is inadequate for science, because there is a reality behind appearance. This article discusses how atoms were discovered, as an example of discovering the reality behind appearance.
This paper was published in Artificial Intelligence in 1986. It gives a better formulation of circumscription and emphasizes making certain functions and predicates variable.
AI needs to deal with objects and predicates that don't admit if-and-only-if definitions. It also needs approximate theories and needs to study the relation between entities at different levels of approximation. This paper is in the Proceedings of KR-2000.
This paper concerns what it means for a machine to have beliefs. This started the dispute about whether thermostats could be considered to have beliefs. It was published in 1979 in an obscure collection and reprinted in my 1990 book 'Formalizing Common Sense'.
Slides discussing what Artificial Intelligence needs from Symbolic Logic
AI Needs more Emphasis on Basic Research was published in AI Magazine in 1983 a a president's message when I was President of AAAI.
This paper, first given in 1961, was published in 1963 in Computer Programming and Formal Systems, edited by P. Braffort and D. Hirschberg and published by North-Holland.
A short story.
This draft article presents a situation calculus formalism featuring events as primary and the usual actions as a special case.
This paper, published in 1982, proposes a language for inter-business inter-computer communication. Most of the ideas in this paper have been re-invented in connection with electronic commerce, specifically in connection with XML.
This paper was presented at the QED meeting in Warsaw in 1995 July. It is a proof in set theory that I think an interactive prover for heavy duty set theory should be able to accept. It uses for a different purpose the same problem as the previous paper.
This essay is an outgrowth of my review of Monty Newborn's Kasparov versus Deep Blue: Computer Chess Comes of Age.
This 1996 article discusses the initial knowledge of the world that makes a baby more competent than a "Lockean baby" would be.
This note is a refutation of that argument that computers can't be conscious.
A humorous June 1999 letter to Christians who also claim to be scientists.
This paper was published in Artificial Intelligence in 1980.
A short note on the need for programs that can translate natural language texts into collections of sentences in mathematical logical language.
Coloring Maps and the Kowalski Doctrine was a 1982 Stanford report. More is known about realizing the Kempe heuristic by making a Prolog that can run in an introspective mode, and I'll put in a note about it when I get a chance.
Concepts of Logical AI has a paragraph each about each of approximately 50 concepts.
This paper was first published in Machine Intelligence 9 in 1979.
Many tasks will require computer programs to examine their own computational structures in ways like those involved in human consciousness and indeed self-consciousness.
Appeared in Proceedings of IJCAI - 1993. This version has an improvement in the way lifting above-theory is treated.
Useful Counterfactuals by Tom Costello and John McCarthy is published in the ETAI (Electronic Transactions on Artificial Intelligence), Vol 3 (1999), Section A.
This paper was given at the AISB Workshop on Artificial Intelligence and Creativity, 1999 April 8 and 9.
Text of the original 1955 Dartmouth Proposal.
Phenomenal Data Mining concerns finding relations between data and phenomena and not just relations within the data. There isn't much AI in the paper - yet, but the idea for phenomenal data mining has somewhat of a philosophical and AI origin.
A condensed version of an invited presentation exploring a number of 'Drosophila' and potential 'Drosophila' for AI
This paper discusses making logical representations of facts that can accept various kinds of modifications easily.
This unpublished draft is a proposal for a new programming language, but it includes the mathematical theory of computation proposal for distinguishing input-output and accomplishment specifications, characterizes input and output statements as speech acts and allows reference to the past in programs.
This talk summarized the epistemological problems I saw at that time (1977).
This paper is a start on a report on experimental computer science (and its relationship to theoretical and applied computer science).
We propose an International Institute for Nuclear Explosions which would conduct research in the scientific and technological uses of nuclear explosions.
A set of expanded notes on formalizing contexts as first class objects.
This paper, written for a special issue of the Journal of Experimental and Theoretical Artificial Intelligence devoted to philosophical questions. Robots will need to consider their own choices in a manner similar to that in which a human contemplates his own free will.
This is a shorter paper than Free Will - Even for Robots. I think it gets to the essence of free will and formalizes it in two short situation calculus formulas.
A draft note exploring the question whether the design of a financial transaction machine would benefit from making it do intentional rule following.
A 1987 summary for the ACM to include in a volume of Turing Award lectures.
Overcoming unexpected obstacles is a note written in 1992 and 1993 describing an important kind of elaboration tolerance. A plan is shown to work by reasoning involving circumscribing a predicate occurs(e,s) asserting that the event e occurs in situation s. If a sentence is added asserting the occurrence of an event that defeats the plan and the circumscription is done again, the plan can no longer be shown to work. In our example, a revised plan including an action that overcomes the obstacle can be shown to work. There is a bug in handling the circumscription that I have not fixed.
An article addressing responses to my 1997 criticism of the expressiveness of modal logic.
This is a 1989 editorial for the CACM. Now that Internet is universally available, its considerations are mostly obsolete.
This article was published in Man and Computer. Proc. int. Conf., Bordeaux 1970, pp. 48-57 (Karger, Basel 1972). I claim quite a few prophet points for it.
A roadmap to human-level AI presented developed from an invited talk at KR-96.
This paper introduced the notion of well-defined problem, but I became convinced that inverting computable functions in general is not a feasible way of doing AI.
An article on the features of the Lemmings computer game that make it suitable as a Drosophila for AI research.
The history of LISP according to McCarthy's memory in 1978, presented at the ACM SIGPLAN History of Programming Languages Conference.
This paper was published in 1980. I put it up since it mostly represents my present opinions. There are some 1999 footnotes.
First published in Psychology Today in 1983, this is a popular article with no formulas.
This paper responds to a request for a note on our approach to formalizing context in mathematical logic that can be compared with John Perry's situation semantics based approach to context. It appeared in a CSLI (Center for Studies in Linguistics and Information) publication.
An incomplete draft of a book incorporating concepts of logical AI, philosophy, common sense reasoning, nonmonotonic reasoning, situation calculus, elaboration tolerance, approximate objects, consciousness and problem solving.
This paper, from 1958, was probably the first on logical AI, i.e. AI in which logic is the method of representing information in computer memory and not just the subject matter of the program.
This paper contains expanded material on context. It is joint work with Sasa Buvac.
This is the basic paper on situation calculus. Joint work with Patrick J. Hayes.
This paper contains a proof of the correctness of a simple compiling algorithm for compiling arithmetic expressions into machine language. It may have been the first such formal proof.
Notes on the potential for using the Missionaries and Cannibals puzzle as a Drosophila for research in elaboration tolerance.
This paper argues that there are better ways, especially for AI, of treating modalities than any kind of modal logic. It appeared in Studia Logica, volume 59, 1997.
This paper, by myself, M. Sato, T. Hayashi and S. Igarashi, was written in the late 1970s.
I think this 1976 memorandum is of modern interest. The problems it raises haven't been solved or even substantially reformulated.
This 1994 article was intended to be superseded by Combining Narratives by McCarthy and Costello. However, I have become attached to some of the constructions of the original article that were omitted from the new version.
This paper is by John McCarthy and Tom Costello. It was presented at KR-98 in Trento, Italy. A major feature is that sentences describing two independent narratives can be combined just by conjoining the sentences. Sentences can be added later to establish relations between them.
Commentary on individual transportation and the environment.
A draft of a story by John McCarthy.
This review of Computer Chess Comes of Age by Monty Newborn appeared in Science on 1997 June 6.
This is a 1982 article. Its observations are mainly confirmed, but some of the measures it advocates are still not implemented.
It is often inadequate that a theory be consistent, i.e. have models. It should have enough models. We discuss parameterizing the set of models in the special case of propositional satisfiability.
Phenomenal data mining concerns establishing relations among the phenomena underlying data. This is paper updated for inclusion in the SIGKDD Explorations.
This appeared in Logical Foundations for Cognitive Agents: Contributions in Honor of Ray Reiter, edited by H. J. Levesque and F. Pirri, Springer-Verlag, 1999.
A collection of philosophical presuppositions and scientific premises of logical AI.
A proposal to encourage great students to become even better.
Here are some ideas about the future of scientific publication. It discusses the advantages, technological problems and economic problems of getting all scientific publication on line. The document may be referenced, but it is subject to revision.
Formalization of two Puzzles Involving Knowledge involves formalization of facts about knowledge including both knowing what and knowing that, how to assume and prove non-knowledge, joint knowledge and the effect of learning a fact on the set of facts then known. It uses the Kripke possible worlds formalism directly.
This was the original paper on LISP.
An example of a problem quickly solvable by present chess programs that illustrates the use of ideas required for human level intelligence. The solution to the problem requires the idea of making two threats with one line of play - an idea useful in more domains than chess.
This is a science fiction story about what household robots might look like.
A collection of situation calculus formulas relating to robot soccer.
Roofs and Boxes is an example to illustrate that extrapolating past experience to predict the future usually involves recognitions of phenomena in the world and not just the sequence of inputs. The problem is too hard for now.
Here are some of my sayings, which may appeal to some.
Here are the slides presented at the DARPA conference on self-awareness in 2004 April
This is a new (2001 August) article on situation calculus. It differs from previous approaches in three ways. It distinguishes internal events that happen spontaneously from external events (actions). It also treats processes, e.g. a buzzer, that do not settle down. The non-monotonic reasoning is circumscription done situation by situation.
Some Expert Systems Need Common Sense was first published in Computer Culture: The Scientific, Intellectual and Social Impact of the Computer, Heinz Pagels, ed. vol. 426, Annals of the New York Academy of Sciences.
There is nothing so cruel and anti-life as a vegetarian. A humorous essay.
This is a 1985 note aimed at regarding STRIPS as a proof strategy for an interactive theorem prover using a situation calculus formalism. It doesn't quite get there.
A discussion of the potential for robotics to allow workers in developing nations to provide services in richer nations.
This is a partial account of a conversation I had with Edward Teller, relevant to Teller's opinion of Heisenberg's role in the German quest for the atom bomb. It's certainly not worth an article, but maybe it will turn up when some historian does a Google search.
This Stanford AI Memo of 1964 illustrated the fact that a theorem may not be easy to prove if the proof involves an idea not expressible in the language in which the theorem is stated.
Towards a Mathematical Science of Computation was given at the congress IFIP-62 and published in the proceedings of that conference. It extends the results of A Basis for a Mathematical Theory of Computation which was first given in 1961.
My 1953 proof that the function is everywhere continuous and nowhere differentiable is just 13 lines. I've added some remarks to the note in the American Mathematical Monthly.
I get many email inquiries about what artificial intelligence is all about. This is a first attempt at answering them on a layman's level or beginning student's level.
This invited commentary appeared in Volume 2, Issue 4 (1995) of the Journal of Consciousness Studies.