Part 4: Philosophy of Artificial Intelligence [319] ============================================= Contents -------- 4.1 The Turing Test [18] 4.2 Godelian Arguments (Lucas) [42] 4.3 The Chinese Room (Searle) [49] 4.4 Machine Consciousness, Misc [24] 4.5 Philosophy of Connectionism [98] 4.5a Connectionism and Compositionality (Fodor/Pylyshyn) [20] 4.5b Representation in Connectionism [11] 4.5c Connectionism and Eliminativism [7] 4.5d The Connectionist/Classical Debate [15] 4.5e Subsymbolic Computation (Smolensky) [6] 4.5f Philosophy of Connectionism, Misc [29] 4.5g Foundational Empirical Issues [10] 4.6 Symbols and Symbol Systems [5] 4.7 Methodological Foundations of AI [22] 4.8 Computation and Semantics [12] 4.9 The Frame Problem [11] 4.10 Analog and Digital Processing [5] 4.11 Levels of Analysis (Marr, etc) [9] 4.12 Philosophy of AI, Misc [24] 4.13 Computationalism in Psychological Explanation -- see 2.9. 4.1 The Turing Test [18] ------------------- Turing, A. 1950. Computing machinery and intelligence. Mind 59:433-60. Proposes the Imitation game (Turing test) as a test for intelligence: If a machine can't be told apart from a human in a conversation over a teletype, then it's intelligent. Barresi, J. 1987. Prospects for the Cyberiad: Certain limits on human self-knowledge in the cybernetic age. Journal for the Theory of Social Behavior 17:19-46. Block, N. 1981. Psychologism and behaviorism. Philosophical Review 90:5-43. A look-up table could pass the Turing test, and surely isn't intelligent. The TT errs in testing behavior and not mechanisms. A nice, thorough paper. Dennett, D.C. 1984. Can machines think? In (M. Shafto, ed) _How We Know_. Harper & Row. Defending the Turing test as a good test for intelligence. French, R.M. 1990. Subcognition and the limits of the Turing test. Mind 99:53-66. The Turing Test is too hard, as it requires not intelligence but human intelligence. Any machine could be unmasked through careful questioning, but this wouldn't mean that the machine was unintelligent. Gunderson, K. 1964. The imitation game. Mind 73:234-45. The Turing test is not broad enough: there's much more to thought than the ability to play the imitation game. Harnad, S. 1991. Other bodies, other minds: A machine incarnation of an old philosophical problem. Minds and Machines 1:43-54. On the Total Turing Test (full behavioral equivalence) as a test for mind. Hofstadter, D.R. 1981. A coffee-house conversation on the Turing test. Scientific American. A dialogue on the Turing test. Karelis, C. 1986. Reflections on the Turing test. Journal for the Theory of Social Behavior 16:161-72. Mays, W. 1952. Can machines think? Philosophy 27:148-62. Moor, J.H. 1976. An analysis of Turing's test. Philosophical Studies 30:249-257. The basis of the Turing test is not an operational definition of thinking, but rather an inference to the best explanation. Moor, J.H. 1978. Explaining computer behavior. Philosophical Studies 34:325-7. Reply to Stalker 1978: Mechanistic and mentalistic explanations are no more incompatible than program-based and physical explanations. Rankin, T.L. 1987. The Turing paradigm: A critical assessment. Dialogue 29:50-55. Some obscure remarks on lying, imitation, and the Turing test. Richardson, R.C. 1982. Turing tests for intelligence: Ned Block's defense of psychologism. Philosophical Studies 41:421-6. A weak argument against Block: input/output function doesn't guarantee a capacity to respond sensibly. Rosenberg, J. 1982. Conversation and intelligence. In (B. de Gelder, ed) _Knowledge and Representation_. Routledge & Kegan Paul. Shanon, B. 1989. A simple comment regarding the Turing test. Journal for the Theory of Social Behavior 19:249-56. The Turing test presupposes a representational/computational framework for cognition. Not all phenomena can be captured in teletype communication. Stalker, D.F. 1978. Why machines can't think: A reply to James Moor. Philosophical Studies 34:317-20. Contra Moor 1976: The best explanation of computer behavior is mechanistic, not mentalistic. Stevenson, J.G. 1976. On the imitation game. Philosophia 6:131-33. 4.2 Godelian arguments (Lucas) [42] ------------------------------ Lucas, J.R. 1961. Minds, machines and Godel. Philosophy 36:112-127. Humans can Godelize any given machine, so we're not a machine. Smart, J.J.C. 1961. Godel's theorem, Church's theorem, and mechanism. Synthese 13:105-10. A machine could escape the Godelian argument by inductively ascertaining its own syntax. With comments on the relevance of ingenuity. Whitely, C. 1962. Minds, machines and Godel: A reply to Mr. Lucas. Philosophy 37:61-62. Humans get trapped too: "Lucas cannot consistently assert this formula". George, F. 1962. Minds, machines and Godel: Another reply to Mr. Lucas. Philosophy 37:62-63. Lucas's argument applies only to deductive machines, not inductive ones. Chari, C. 1963. Further comments on minds, machines and Godel. Philosophy 38:175-8. Can't reduce the lawless creative process to computation. Good, I.J. 1967. Human and machine logic. British Journal for the Philosophy of Science 18:145-6. Even humans can't Godelize forever; ordinals, transfinite counting. Lucas, J.R. 1967. Human and machine logic: a rejoinder. British Journal for the Philosophy of Science 19:155-6. Reply to Good 1967: human can always trump machine, so H <> M. (not nec H>M). Good, I.J. 1969. Godel's theorem is a red herring. British Journal for the Philosophy of Science 19:357-8. Rejoinder to Lucas 1967: the role of consistency; non-constructible ordinals. Benacerraf, P. 1967. God, the Devil, and Godel. Monist 51:9-32. Consistency; sharpens Lucas; maybe we are machines, but can't know which. Lucas, J.R. 1968. Satan stultified: A rejoinder to Paul Benacerraf. Monist 52:145-58. Benacerraf 1967 is empty and omega-inconsistent. Reply to arguments based on difficulty of seeing consistency (e.g. Putnam). Fallacious but engaging. Webb, J. 1968. Metamathematics and the philosophy of mind. Philosophy of Science 35:156-78. Lucas, J.R. 1971. Metamathematics and the philosophy of mind: A rejoinder. Philosophy of Science 38:310-13. Lewis, D. 1969. Lucas against mechanism. Philosophy 44:231-3. Lucas cannot produce all of `Lucas arithmetic'. Coder, D. 1969. Godel's theorem and mechanism. Philosophy 44:234-7. Only mathematicians understand Godel; and TM's can go wrong. Weak. Lucas, J.R. 1970. Mechanism: A rejoinder. Philosophy 45:149-51. To Lewis 1969: dialectic nature. To Coder 1969: whatever. Lucas, J.R. 1970. _The Freedom of the Will_. Oxford University Press. Hanson, W. 1971. Mechanism and Godel's theorem. British Journal for the Philosophy of Science 22:9-16. Chihara, C. 1972. On alleged refutations of mechanism using Godel's incompleteness results. Journal of Philosophy 64:507-26. An analysis of the Lucas/Benacerraf argument. Wang, H. 1974. _From Mathematics to Philosophy_. London. Hutton, A. 1976. This Godel is killing me. Philosophia 3:135-44. Gives a statistical argument to the effect that we cannot know we are consistent; so the Lucas argument cannot go through. Lucas, J.R. 1976. This Godel is killing me: A rejoinder. Philosophia 6:145-8. Contra Hutton, we know -- even if fallibly -- that we are consistent. Dennett, D.C. 1978. The abilities of men and machines. In _Brainstorms_. MIT Press. There is no unique TM which we are -- there could be many. Lewis, D. 1979. Lucas against mechanism II. Canadian Journal of Philosophy 9:373-6. Reply to Lucas 1970: dialectic argument fails as output depends on input. Hofstadter, D.R. 1979. _Godel, Escher, Bach: An Eternal Golden Braid_. Basic Books. Contra Lucas: we can't Godelize forever; and we're not formal on top level. Webb, J. 1980. _Mechanism, Mentalism and Metamathematics_. Kluwer. Bowie, G. 1982. Lucas' number is finally up. Journal of Philosophy Logic, 11:279-85. Lucas's very Godelization procedure makes him inconsistent, unless he has an independent way to see if any TM is consistent, which he doesn't. Right. Slezak, P. 1982. Godel's theorem and the mind. British Journal for the Philosophy of Science 33:41-52. General analysis; Lucas commits type/token error; self-ref paradoxes. Slezak, P. 1983. Descartes's diagonal deduction. British Journal for the Philosophy of Science 34:13-36. Cogito was a diagonal argument; connection to Godel, Lucas, Minsky, Nagel. Boyer, D. 1983. J.R. Lucas, Kurt Godel, and Fred Astaire. Philosophical Quarterly 33:147-59. Remarks on the various ways in which Lucas and a machine might be said to "prove" anything, and the ways in which a machine might simulate Lucas. The argument has all sorts of level confusions, and a bit of circularity. Lucas, J.R. 1984. Lucas against mechanism II: A rejoinder. Canadian Journal of Philosophy 14:189-91. Reply to Lewis 1979. Putnam, H. 1985. Reflexive reflections. Erkenntnis 22:143-153. A generalized Godelian argument: if our prescriptive inductive competence is formalizable, then we could not know that such a formalization is correct. Kirk, R. 1986. Mental machinery and Godel. Synthese. Lucas's argument fails, as theorems by humans don't correspond to outputs of their formal systems. Jacquette, D. 1987. Metamathematical criteria for minds and machines. Erkenntnis 27:1-16. A machine will fail a Turing test if it's asked about Godel sentences. Penrose, R. 1989. _The Emperor's New Mind_. Oxford University Press. We are non-algorithmic as we can see Godel sentences of any algorithm. Penrose, R. 1990. Precis of _The Emperor's New Mind_. Behavioral and Brain Sciences 13:643-705. Much debate over the "non-algorithmic insight" in seeing Godel sentences. Penrose, R. 1992. Setting the scene: The claim and the issues. In (D. Broadbent, ed) _The Simulation of Human Intelligence_. Blackwell. An argument from the halting problem to the nonalgorithmicity of mathematical thought. Addresses objections: that the algorithm is unknowable, unsound, everchanging, environmental, or random. New physical laws may be involved. Lyngzeidetson, A. 1990. Massively parallel distributed processing and a computationalist foundation for cognitive science. British Journal for the Philosophy of Science 41. A Connection Machine might escape the Lucas argument. Bizarre. Martin, J. & Engleman, K. 1990. The mind's I has two eyes. Philosophy 510-16. Contra Hofstadter: Lucas can believe his Whitely sentence. Lucas, J.R. 1990. Mind, machines and Godel: A retrospect. Manuscript. Addresses all the counterarguments. Fun. Tymoczko, T. 1991. Why I am not a Turing Machine: Godel's theorem and the philosophy of mind. In (J. Garfield, ed) _Foundations of Cognitive Science_. Paragon House. Weak defense of Lucas; response to Putnam, Bowie, Dennett. Robinson, W.S. 1992. Penrose and mathematical ability. Analysis 52:80-88. Penrose's argument depends on our knowledge of the validity of the algorithm we use, and here he equivocates between conscious and unconscious algorithms. Yu, Q. 1992. Consistency, mechanicalness, and the logic of the mind. Synthese 90:145-79. 4.3 The Chinese Room (Searle) [49] ----------------------------- Searle, J.R. 1980. Minds, brains and programs. Behavioral and Brain Sciences 3:417-57. Implementing a program is not sufficient for mentality, as someone could e.g. implement a "Chinese-speaking" program without understanding Chinese. So strong AI is false, and no program is sufficient for consciousness. Searle, J.R. 1984. _Minds, Brains and Science_. Harvard University Press. Axiomatizes the argument: Syntax isn't sufficient for semantics, programs are syntactic, minds are semantic, so no program is sufficient for mind. Searle, J.R. 1987. Minds and brains without programs. In (C. Blakemore, ed) _Mindwaves_. Blackwell. More on the arguments against AI, e.g. the Chinese room and considerations about syntax and semantics. Mind is a high-level physical property of brain. Searle, J.R. 1990. Is the brain's mind a computer program? Scientific American 262(1):26-31. On the status of the Chinese Room argument, ten years on. Anderson, D. 1987. Is the Chinese room the real thing? Philosophy 62:389-93. Boden, M. 1988. Escaping from the Chinese Room. In _Computer Models of Mind_. Cambridge University Press. A procedural account of how computers might have understanding and semantics. Ben-Yami, H. 1993. A note on the Chinese room. Synthese 95:169-72. A fully functional Chinese room is impossible, as it (for instance) could not say what the time is. Bynum, T.W. 1985. Artificial intelligence, biology, and intentional states. Metaphilosophy 16:355-77. A chess-playing machine embodied as a robot could have intentional states. Reference requires input/output, computation, and context. Cam, P. 1990. Searle on strong AI. Australasian Journal of Philosophy 68:103-8. Criticizes Searle's "conclusion" that brains are needed for intentionality, notes that even a homunculus has intentional states. A misinterpretation. Carleton, L. 1984. Programs, language understanding, and Searle. Synthese 59:219-30. Arguing against Searle on a number of fronts, somewhat unconvincingly. Chalmers, D.J. 1992. Subsymbolic computation and the Chinese Room. In (J. Dinsmore, ed) _The Symbolic and Connectionist Paradigms: Closing the Gap_. Erlbaum. Gives an account of symbolic vs. subsymbolic computation, and argues that the latter is less vulnerable to the Chinese-room intuition, as representations there are not computational tokens. Churchland, P.M. & Churchland, P.S. 1990. Could a machine think? Scientific American 262(1):32-37. Artificial mentality is possible, not through classical AI but through brain-like AI. Argues the syntax/semantics point using an analogy with electromagnetism and luminance. Cohen, L.J. 1986. What sorts of machines can understand the symbols they use? Aristotelian Society Supplement 60:81-96. Cole, D.J. 1984. Thought and thought experiments. Philosophical Studies 45:431-44. Lots of thought experiments like Searle's, against Searle. Searle's argument is like Leibniz's "mill" argument, with similar level confusions. Nice but patchy. Cole, D.J. 1991. Artificial intelligence and personal identity. Synthese 88:399-417. In the Chinese room, neither the person nor the system understands: a virtual person does. This person isn't the system, just as a normal person isn't a body. Follows from the "Kornese" room, which has two distinct understanders. Copeland, B.J. 1993. The curious case of the Chinese gym. Synthese 95:173-86. Advocates the systems reply, and criticizes Searle's "Chinese Gym" response to connectionism: Searle (like those he accuses) confuses a simulation with the thing being simulated. Nice. Dennett, D.C. 1987. Fast thinking. In _The Intentional Stance_. MIT Press. Argues with Searle on many points. A little weak. Double, R. 1983. Searle, programs and functionalism. Nature and System 5:107-14. The homunculus doesn't have access to the system's intentionality. The syntax/semantics relation is like the neurophysiology/mind relation. Fields, C. 1984. Double on Searle's Chinese Room. Nature and System 6:51-54. Double's argument implies that the brain isn't the basis of intentionality. Fisher, J. 1988. The wrong stuff: Chinese rooms and the nature of understanding. Philosophical Investigations 11:279-99. Fodor, J.A. 1991. Yin and Yang in the Chinese Room. In (D. Rosenthal, ed) _The Nature of Mind_. Oxford University Press. The Chinese room isn't even implementing a Turing machine, because it doesn't use proximal causation. With a reply by Searle. Hanna, P. 1985. Causal powers and cognition. Mind 94:53-63. Argues that Searle is confused, and underestimates computers. Weak. Harnad, S. 1989. Minds, machines and Searle. Journal of Experimental and Theoretical Artificial Intelligence 1:5-25. Non-symbolic function is necessary for mentality. Trying hard to work out a theory of why the Chinese Room shows what it does. Nice but wrong. Dyer, M. 1990. Intentionality and computationalism: minds, machines, Searle and Harnad. Journal of Experimental and Theoretical Artificial Intelligence 2:303-19. Reply to Searle/Harnad: systems reply, level confusions, etc. Harnad, S. 1990. Lost in the hermeneutical hall of mirrors. Journal of Experimental and Theoretical Artificial Intelligence 2:321-27. Reply to Dyer: on the differences between real and as-if intentionality. Dyer, M. 1990. Finding lost minds. Journal of Experimental and Theoretical Artificial Intelligence 2:329-39. Reply to Harnad: symbols, other minds, and physical embodiment of algorithms. Hofstadter, D.R. 1981. Reflections on Searle. In (D. Hofstadter & D. Dennett, eds) _The Mind's I_, pp. 373-382. Basic Books. Searle is committing a level confusion, and understates the complexity of the case. We can move from the CR to a brain (with a demon) by twiddling knobs, and the systems reply should work equally well in both cases. Jacquette, D. 1989. Searle's intentionality thesis. Synthese 80:267-75. Searle's view implies that intentional causation is not efficient causation. Jacquette, D. 1989. Adventures in the Chinese Room. Philosophy and Phenomenological Research 49:605-23. If we had microfunctional correspondence, the CR argument would fail. With points about the status of abstract/biological intentionality. A bit weak. Searle, J.R. 1989. Reply to Jacquette. Philosophy and Phenomenological Research 49:701-8. Jacquette misses the point of the argument. Also, biological and abstract intentionality are quite compatible. Jacquette, D. 1990. Fear and loathing (and other intentional states) in Searle's Chinese Room. Philosophical Psychology 3:287-304. Reply to Searle on CR, central control, biological intentionality & dualism. Jahren, N. 1990. Can semantics be syntactic? Synthese 82:309-28. Against Rapaport's Korean Room argument -- syntax isn't enough. Korb, K. 1991. Searle's AI program. Journal of Experimental and Theoretical Artificial Intelligence 3:283-96. The Chinese room doesn't succeed as an argument about semantics. At best it might succeed as an argument about consciousness. Maloney, J.C. 1987. The right stuff. Synthese 70:349-72. Defends Searle against all kinds of objections. Exhaustive but flawed. Moor, J.H. 1988. The pseudorealization fallacy and the Chinese Room argument. In (J. Fetzer, ed) _Aspects of AI_. D. Reidel. Computational systems must also meet performance criteria. Newton, N. 1989. Machine understanding and the Chinese Room. Philosophical Psychology 2:207-15. A program can possess intentionality, even if not consciousness. Obermeier, K.K. 1983. Wittgenstein on language and artificial intelligence: The Chinese-room thought-experiment revisited. Synthese 56:339-50. Rapaport, W. 1984. Searle's experiments with thought. Philosophy of Science 53:271-9. Comments on Cole, and some general material on syntax and semantics. Rey, G. 1986. What's really going on in Searle's `Chinese Room'. Philosophical Studies 50:169-85. Recommends the systems reply, and a causal account of semantics. Discusses the relevance of wide and narrow notions of content, and the tension between Searle's positive and negative proposals. Roberts, L. 1990. Searle's extension of the Chinese Room to connectionist machines. Journal of Experimental and Theoretical Artificial Intelligence 2:185-7. In arguing against the relevance of the serial/parallel distinction to mental states, Searle becomes a formalist. A nice point. Russow, L-M. 1984. Unlocking the Chinese Room. Nature and System 6:221-8. Searle's presence in the room destroys the integrity of the system, so that it is no longer a proper implementation of the program. Seidel, A. 1988. Searle on the biological basis of cognition. Analysis 48:26-28. Seidel, A. 1989. Chinese Rooms A, B and C. Pacific Philosophical Quarterly 20:167-73. A person running the program, with interpretations at hand, would understand. Point-missing. Sharvy, R. 1985. Searle on programs and intentionality. Canadian Journal of Philosophy Supplement 11:39-54. Argues against Searle, but misses the point for the most part. Sloman, A. 1986. Did Searle attack Strong Strong AI or Weak Strong AI? In (Cohn & Thomas, eds) _Artificial Intelligence and its Applications_. Chichester. Suits, D. 1989. Out of the Chinese Room. Computing and Philosophy Newsletter 4:1-7. Story about homunculi within homunculi. Fun. Thagard, P. 1986. The emergence of meaning: An escape from Searle's Chinese Room. Behaviorism 14:139-46. Get semantics computationally via induction and functional roles. Weiss, T. 1991. Closing the Chinese room. Ratio 3:165-81. Searle-in-the-room isn't in a position to know about the system's first-person states. Intrinsic intentionality is an incoherent notion. Whitmer, J.M. 1983. Intentionality, artificial intelligence, and the causal powers of the brain. Auslegung 10:194-210. Defending Searle's position, with remarks on the "causal powers" argument. 4.4 Machine Consciousness (other) [24] --------------------------------- Barnes, E. 1991. The causal history of computational activity: Maudlin and Olympia. Journal of Philosophy 88:304-16. Response to Maudlin 1989. True computation needs active, not passive causation, so Maudlin's machine isn't really computing. Cohen, L.J. 1955. Can there be artificial minds? Analysis 16:36-41. Subservience to known or knowable rules is incompatible with mentality. Dennett, D.C. 1985. Can machines think? In _How We Know_ (Shafto). Defends the Turing Test, among other things. Gunderson, K. 1968. Robots, consciousness and programmed behaviour. British Journal for the Philosophy of Science 19:109-22. Gunderson, K. 1969. Cybernetics and mind-body problems. Inquiry 12:406-19. Gunderson, K. 1971. _Mentality and Machines_. Doubleday. Kirk, R. 1986. Sentience, causation and some robots. Australasian Journal of Philosophy 64:308-21. One could model brain states with monadic states and appropriate connections. But surely that's not intelligent -- the causation has the wrong form. Nice. Mackay, D. 1951. Mind-life behavior in artifacts. British Journal for the Philosophy of Science 2:105-21. Maudlin, T. 1989. Computation and consciousness. Journal of Philosophy 86:407-32. Computational state is not sufficient for consciousness, as it can be instantiated by a mostly inert object. A very nice thought-experiment, raising questions about the relevance of counterfactuals to consciousness. Mays, W. 1952. Can machines think? Philosophy 27:148-62. McGinn, C. 1987. Could a machine be conscious? In (C. Blakemore & S. Greenfield, ed) _Mindwaves_. Blackwell. Reprinted in _The Problem of Consciousness_ (Blackwell, 1980). Of course, as we are machines. But what *sort* of machines are conscious, and in virtue of what properties? Remarks on artefacts, life, functionalism, and computationalism. So far, we don't know what makes the brain conscious. Negley, G. 1951. Cybernetics and theories of mind. Journal of Philosophy 48:574-82. Puccetti, R. 1966. Can humans think? Analysis. Puccetti, R. 1967. On thinking machines and feeling machines. British Journal for the Philosophy of Science 18:39-51. Machines can think but can't feel, so aren't persons. Putnam, H. 1960. Minds and machines. In (S. Hook, ed) _Dimensions of Mind_. New York University Press. Reprinted in _Mind, Language, and Reality_ (Cambridge University Press, 1975). Suggests that the mind-body problem is precisely analogous to the relationship between logical and structural states of a Turing Machine, Putnam, H. 1964. Robots: machines or artificially created life? Journal of Philosophy 61:668-91. Reprinted in _Mind, Language, and Reality_ (Cambridge University Press, 1975). Various arguments and counter-arguments re machine consciousness and civil liberties. Problems of machine consciousness are analogous to problems of human consciousness. The structural basis of the two may well be the same. Putnam, H. 1967. The mental life of some machines. In (H. Castaneda, ed) _Intentionality, Minds and Perception_. Wayne State University Press. Reprinted in _Mind, Language, and Reality_ (Cambridge University Press, 1975). More on TMs: explaining their psychology via preference functions. Scriven, M. 1953. The mechanical concept of mind. Mind. To speak of a conscious machine is to commit a semantic mistake. Consciousness presupposes life and non-mechanism. Later retracted. Scriven, M. 1960. The compleat robot: A prolegomena to androidology. In (S. Hook, ed) _Dimensions of Mind_. New York University Press. A machine could possess every characteristic of human thought: e.g. freedom, creativity, learning, understanding, perceiving, feeling. Stubenberg, L. 1992. What is it like to be Oscar? Synthese 90:1-26. Argues that AI systems like Pollock's Oscar needn't be conscious. Blindsight tells us that complex perceptual processing can go on unconsciously. Thompson, D. 1965. Can a machine be conscious? British Journal for the Philosophy of Science 16:36. Accepting machine consciousness would have few philosophical consequences, whereas rejecting it would tend to commit one to epiphenomenalism. Turing, A. 1950. Computing machinery and intelligence. Mind 59:433-60. Proposing an operational criterion (the "Turing test") for whether a machine could think: indistinguishability from humans in conversation over teletype. With replies to objections (consciousness, theology, originality, etc). van de Vete, D. 1971. The problem of robot consciousness. Philosophy and Phenomenological Research 32:149-65. Ziff, P. 1959. The feelings of robots. Analysis. Of course robots can't think: they're not alive, so this gives us good reason not to rely on behavior. With replies by J.J.C. Smart, N. Smart. 4.5 Philosophy of Connectionism [98] ------------------------------- 4.5a Connectionism and Compositionality (Fodor/Pylyshyn) [20] -------------------------------------------------------- Fodor, J.A. & Pylyshyn, Z.W. 1988. Connectionism and cognitive architecture. Cognition 28:3-71. Connectionist models can't explain cognitive systematicity and productivity, as their representations lack compositional structure. The allures of connectionism are illusory; it's best used as an implementation strategy. Butler, K. 1991. Towards a connectionist cognitive architecture. Mind and Language 6:252-72. Connectionism can make do with unstructured representations, as long have they have the right causal relations between them. Butler, K. 1993. Connectionism, classical cognitivism, and the relation between cognitive and implementational levels of analysis. Philosophical Psychology 6:321-33. Contra Chalmers 1993, F&P's argument doesn't apply at the implementational level. Contra Chater and Oaksford 1990, connectionism can't be purely implementational, but some implementational details can be relevant. Butler, K. 1993. On Clark on systematicity and connectionism. British Journal for the Philosophy of Science 44:37-44. Argues against Clark on holism and the conceptual truth of systematicity. Chalmers, D.J. 1990. Syntactic transformations on distributed representations. Connection Science 2:53-62. An experimental demonstration that connectionist models can handle structure-sensitive operations in a non-classical way, transforming structured representations of active sentences to passive sentences. Chalmers, D.J. 1993. Connectionism and compositionality: Why Fodor and Pylyshyn were wrong. Philosophical Psychology 6:305-319. Points out a structural flaw in F&P's argument, and traces the problem to a lack of appreciation of distributed representation. With some empirical results on structure sensitive processing, and some remarks on explanation. Chater, N. & Oaksford, M. 1990. Autonomy, implementation and cognitive architecture: A reply to Fodor and Pylyshyn. Cognition 34:93-107. Implementation can make a difference at the algorithmic level. Fodor, J.A. & McLaughlin, B.P. 1990. Connectionism and the problem of systematicity: Why Smolensky's solution doesn't work. Cognition 35:183-205. Smolensky's weak compositionality is useless; and tensor product architecture can't support systematicity, as nonexistent tokens can't play a causal role. Hadley, R.F. 1994. Compositionality and systematicity in connectionist language learning. Mind and Language. Argues that existing connectionist models do not achieve an adequate systematicity in learning; they fail to generalize to handle structures with novel constituents. Hawthorne, J. 1989. On the compatibility of connectionist and classical models. Philosophical Psychology 2:5-16. Localist connectionist models may not be able to handle structured presentation, but appropriate distributed models can. Horgan, T. & Tienson, J. 1991. Structured representations in connectionist systems? In (Davis, ed) _Connectionism: Theory and Practice_. A discussion of how connectionism might achieve "effective syntax" without implementing a classical system. Matthews, R.J. 1994. Three-card monte: explanation, implementation, and systematicity. Synthese. F&P deal a sucker bet: on their terms, connectionism could never give a a non-implementational explanation of systematicity, as the notions are construed in a manner specific to classical architectures. McLaughlin, B.P. 1992. Systematicity, conceptual truth, and evolution. In _Philosophy and the Cognitive Sciences_. Against responses to Fodor and Pylyshyn claiming that cognitive theories needn't explain systematicity. Contra Clark, the conceptual truth of systematicity won't help. Contra others, nor will evolution. McLaughlin, B.P. 1993. The connectionism/classicism battle to win souls. Philosophical Studies. Argues that no connectionist model so far has come close to explaining systematicity. Considers the models of Elman, Chalmers, and Smolensky. Pollack, J.B. 1990. Recursive distributed representations. Artificial Intelligence 46:77-105. Develops a connectionist architecture -- recursive auto-associative memory -- that can recursively represent compositional structures in distributed form. Smolensky, P. 1987. The constituent structure of connectionist mental states. Southern Journal of Philosophy Supplement 26:137-60. F&P ignore distributed representation and interaction effects. Smolensky, P. 1990. Tensor product variable binding and the representation of symbolic structures in connectionist systems. Artificial Intelligence 46:159-216. Develops a connectionist architecture that represents compositional structures as tensor products of distributed representations. Smolensky, P. 1991. Connectionism, constituency and the language of thought. In (B. Loewer & G. Rey, eds) _Meaning in Mind: Fodor and his Critics_. Blackwell. Connx can do compositionality its own way, either weak compositionality (with context effects) or strong compositionality (tensor products). van Gelder, T. 1990. Compositionality: A connectionist variation on a classical theme. Cognitive Science 14:355-84. Connectionism can do compositionality functionally. All one needs is the right functional relation between representations; physical concatenation is not necessary. van Gelder, T. 1991. Classical questions, radical answers. In (T. Horgan & J. Tienson, eds) _Connectionism and the Philosophy of Mind_. Kluwer. On connectionism as a Kuhnian paradigm shift in cognitive science, with emphasis on the implications of functional compositionality and distributed representations. 4.5b Representation in Connectionism [11] ------------------------------------ Clark, A. 1989. Connectionism, non-conceptual content, and representational redescription. Manuscript. On some troubles connx has with higher-order knowledge. Contrasts Cussins and Karmiloff-Smith on development. Subsymbols without symbols are blind. Clark, A. 1993. _Associative Engines: Connectionism, Concepts, and Representational Change_. MIT Press. Clark, A. & Karmiloff-Smith, A. 1994. The cognizer's innards: A psychological and philosophical perspective on the development of thought. Mind and Language. On the importance of representational redescription, and on limits of connx networks in cross-domain knowledge transfer. What, above behavior, does a true believer need: conceptual combination, real-world fluency? Cummins, R. 1991. The role of representation in connectionist explanation of cognitive capacities. In (W. Ramsey, S. Stich, & D. Rumelhart, eds) _Philosophy and Connectionist Theory_. Erlbaum. Connectionism isn't really radical. There's no new concept of representation or of learning, and cognition can still be the manipulation of semantically structured representations. Cussins, A. 1990. The connectionist construction of concepts. In (M. Boden, ed) _The Philosophy of AI_. Oxford University Press. Connx builds concepts up from the nonconceptual level. From nonconceptual content (e.g. perceptual experiences) to the emergence of objectivity. Goschke, T. & Koppelberg, D. 1990. Connectionism and the semantic content of internal representation. Review of International Philosophy 44:87-103. Goschke, T. & Koppelberg, D. 1991. The concept of representation and the representation of concepts in connectionist models. In (W. Ramsey, S. Stich, & D. Rumelhart, eds) _Philosophy and Connectionist Theory_. Erlbaum. On correlational semantics and context-dependent representations. Hatfield, G. 1991. Representation and rule-instantiation in connectionist systems. In (T. Horgan & J. Tienson, eds) _Connectionism and the Philosophy of Mind_. Kluwer. Some remarks on psychology & physiology. Even connectionism uses psychological concepts. Hatfield, G. 1991. Representation in perception and cognition: Connectionist affordances. In (W. Ramsey, S. Stich, & D. Rumelhart, eds) _Philosophy and Connectionist Theory_. Erlbaum. Tye, M. 1987. Representation in pictorialism and connectionism. Southern Journal of Philosophy Supplement 26:163-184. Pictorialism isn't compatible with language of thought, but connx might be. van Gelder, T. 1991. What is the D in PDP? In (W. Ramsey, S. Stich, & D. Rumelhart, eds) _Philosophy and Connectionist Theory_. Erlbaum. Argues that distributed representation is best analyzed in terms of superposition of representation, not in terms of extendedness. 4.5c Connectionism and Eliminativism [7] ------------------------------------ Ramsey, W., Stich, S.P. & Garon, J. 1991. Connectionism, eliminativism and the future of folk psychology. In (W. Ramsey, S. Stich, & D. Rumelhart, eds) _Philosophy and Connectionist Theory_. Erlbaum. Connectionism implies eliminativism, as connectionist systems do not have functionally discrete contentful states, and folk psychology is committed to functional discreteness of propositional attitudes. Bickle, J. 1993. Connectionism, eliminativism, and the semantic view of theories. Erkenntnis. Outlines the semantic view of scientific theories, and applies it to the connectionism/eliminativism debate. There's no reason why folk psychology shouldn't be reducible, in a homogeneous or heterogeneous way. Clapin, H. 1991. Connectionism isn't magic. Minds and Machines 1:167-84. Commentary on Ramsey/Stich/Garon. Connectionism has symbols that interact, and has propositional modularity in processing if not in storage. Clark, A. 1989. Beyond eliminativism. Mind and Language 4:251-79. Connectionism needn't imply eliminativism, as higher levels may have a causal role, if not causal completeness. Also, it may not tell the whole story. Clark, A. 1990. Connectionist minds. Proceedings of the Aristotelian Society 90:83-102. Responding to eliminativist challenge via cluster analysis and recurrence. Davies, M. 1991. Concepts, connectionism, and the language of thought. (W. Ramsey, S. Stich, & D. Rumelhart, eds) _Philosophy and Connectionist Theory_. Erlbaum. Argues that our conception of thought requires causal systematicity, which requires a language of thought. Connectionist systems are not causally systematic, so connectionism leads to eliminativism. Forster, M. & Saidel, E. 1993. Connectionism and the fate of folk psychology. Philosophical Psychology. Contra Ramsey, Stich, and Garon, connectionist representations can be seen to be functionally discrete on an appropriate analysis of causal relevance. 4.5d The Connectionist/Classical Debate [15] --------------------------------------- Adams, F., Aizawa, K. & Fuller, G. 1992. Rules in programming languages and networks. In (J. Dinsmore, ed) _The Symbolic and Connectionist Paradigms: Closing the Gap_. Erlbaum. The distinction between programming languages and networks is neutral on rule-following, etc, so there's nothing really new about connectionism. Bringsjord, S. 1991. Is the connectionist-logicist debate one of AI's wonderful red herrings? Journal of Theoretical and Experimental Artificial Intelligence 3:319-49. A detailed analysis purporting to show that connectionism and "logicism" are compatible, as Turing machines can do everything a neural network can. Entertaining, but misunderstands subsymbolic processing. Broadbent, D. 1985. A question of levels: Comment on McClelland and Rumelhart. Journal of Experimental Psychology: General 114:189-92. Distributed models are at the implementational, not computational, level. Chandrasekaran, B., Goel, A. & Allemang, D. 1988. Connectionism and information-processing abstractions. AI Magazine 24-34. Connectionism won't affect AI too much, as AI is concerned with the information-processing (task) level. With greater modularity, connectionism will look more like traditional AI. Corbi, J.E. 1993. Classical and connectionist models: Levels of description. Synthese 95: 141-68. Dennett, D.C. 1986. The logical geography of computational approaches: A view from the east pole. In (M. Brand & R. Harnish, eds) _The Representation of Knowledge and Belief_. University of Arizona Press. Drawing the battle-lines: High Church Computationalism at the "East Pole", New Connectionism, Zen Holism, etc, at various locations on the "West Coast". With remarks on connectionism, and on AI as thought-experimentation. Dennett, D.C. 1991. Mother Nature versus the walking encyclopedia. In (W. Ramsey, S. Stich, & D. Rumelhart, eds) _Philosophy and Connectionist Theory_. Erlbaum. Reiterating the value of connectionism, especially biological plausibility. Dinsmore, J. (ed) 1992. _The Symbolic and Connectionist Paradigms: Closing the Gap_. Erlbaum. Dyer, M. 1991. Connectionism versus symbolism in high-level cognition. In (T. Horgan & J. Tienson, eds) _Connectionism and the Philosophy of Mind_. Kluwer. Garson, J. 1991. What connectionists cannot do: The threat to Classical AI. In (T. Horgan & J. Tienson, eds) _Connectionism and the Philosophy of Mind_. Kluwer. Connx/classicism aren't necessarily incompatible on symbolic discreteness, causal role, functional discreteness, constituency, representation of rules. Horgan, T. & Tienson, J. 1987. Settling into a new paradigm. Southern Journal of Philosophy Supplement 26:97-113. On connectionism, basketball, and representation without rules. Responses to the "syntactic" and "semantic" arguments against connectionism. Nice. Horgan, T. & Tienson, J. 1989. Representation without rules. Philosophical Perspectives 17:147-74. Cognition uses structured representations without high-level rules, and connectionism is better at accounting for this. With remarks on exceptions to psychological laws, and the crisis in traditional AI. McClelland, J.L. & Rumelhart, D.E. 1985. Levels indeed! A response to Broadbent. Journal of Experimental Psychology: General 114:193-7. Response to Broadbent 1985: Distributed models are at the algorithmic level. Elucidating the low-level/high-level relation via various analogies. McLaughlin, B.P. & Warfield, F. 1994. The allures of connectionism. Synthese. Argues that symbolic systems such as decision trees are as good at learning and pattern recognition as connectionist networks, and it is just as plausible that they are implemented in the brain. Rey, G. 1991. An explanatory budget for connectionism and eliminativism. In (T. Horgan & J. Tienson, eds) _Connectionism and the Philosophy of Mind_. Kluwer. Challenges connectionism to explain things that the classical approach seems to handle better: the structure, systematicity, causal role, and grain of propositional attitudes, their rational relations, and conceptual stability. 4.5e Subsymbolic Computation (Smolensky) [6] ---------------------------------------- Smolensky, P. 1988. On the proper treatment of connectionism. Behavioral and Brain Sciences 11:1-23. Connectionism offers a complete account at the subsymbolic level, rather than an approximate account at the symbolic level. Chalmers, D.J. 1992. Subsymbolic computation and the Chinese Room. In (J. Dinsmore, ed) _The Symbolic and Connectionist Paradigms: Closing the Gap_. Erlbaum. Explicates the distinction between symbolic and subsymbolic computation, and argues that connectionism can better handle the emergence of semantics from syntax, doe to the non-atomic nature of its representations. Hofstadter, D.R. 1983. Artificial intelligence: Subcognition as computation. In (F. Machlup, ed) _The Study of Information: Interdisciplinary Messages_. Wiley. Reprinted as "Waking up from the Boolean dream" in _Metamagical Themas_. Basic Books. AI needs statistical emergence. For real semantics, symbols must be decomposable, complex, autonomous -- i.e. active. Marinov, M. 1993. On the spuriousness of the symbolic/subsymbolic distinction. Minds and Machines 3:253-70. Argues with Smolensky: symbolic systems such as decision trees have all the positive features of neural networks (flexibility, lack of brittleness), and can represent concepts as sets of subconcepts. With a reply by Clark. Rosenberg, J. 1990. Treating connectionism properly: Reflections on Smolensky. Psychological Research 52:163. Rejects Smolensky's PTC, as the proper interaction of the microscopic and macroscopic levels would take a "miracle". Smolensky, P. 1987. Connectionist AI, symbolic AI, and the brain. AI Review 1:95-109. On connectionist networks as subsymbolic dynamic systems. 4.5f Philosophy of Connectionism, Misc. [29] --------------------------------------- Bechtel, W. 1985. Are the new PDP models of cognition cognitivist or associationist? Behaviorism 13:53-61. Bechtel, W. 1986. What happens to accounts of mind-brain relations if we forgo an architecture of rules and representations? Philosophy of Science Association 159-71. On the relationship between connectionism, symbol processing, psychology and neuroscience. Bechtel, W. 1987. Connectionism and the philosophy of mind. Southern Journal of Philosophy Supplement 26:17-41. Reprinted in (W. Lycan, ed) _Mind and Cognition (Blackwell, 1990). Lots of questions about connectionism. Bechtel, W. 1988. Connectionism and rules and representation systems: Are they compatible? Philosophical Psychology 1:5-16. There's room for both styles within a single mind. The rule-based level needn't be autonomous; the connectionist level plays a role in pattern recognition, concepts, and so on. Bechtel, W. & Abrahamson, A. 1990. Beyond the exclusively propositional era. Synthese 82:223-53. An account of the shift from propositions to pattern recognition in the study of cognition: knowing-how, imagery, categorization, connectionism. Bechtel, W., & Abrahamsen, A.A. 1992. Connectionism and the future of folk psychology. In (R. Burton, ed) _Minds: Natural and Artificial_. SUNY Press. Bradshaw, D.E. 1991. Connectionism and the specter of representationalism. In (T. Horgan & J. Tienson, eds) _Connectionism and the Philosophy of Mind_. Kluwer. Argues that connectionism allows for a more plausible epistemology of perception, compatible with direct realism rather than representationalism. With remarks on Fodor and Pylshyn's argument against Gibson. Churchland, P.M. 1989. On the nature of theories: A neurocomputational perspective. Minnesota Studies in the Philosophy of Science 14. Reprinted in _A Neurocomputational Perspective_ (MIT Press, 1989). Connectionism will revolutionize our review of scientific theories: From the deductive-nomological view to descent in weight-space. Some cute analogies. Churchland, P.M. 1989. On the nature of explanation: A PDP approach. In _A Neurocomputational Perspective_. MIT Press. We achieve explanatory understanding not through the manipulation of propositions but through the activation of prototypes. Churchland, P.S. & Sejnowski, T. 1989. Neural representation and neural computation. In (L. Nadel, ed) _Neural Connections, Mental Computations_. MIT Press. Implications of connectionism and neuroscience for our concept of mind. Clark, A. 1989. _Microcognition_. MIT Press. All kinds of stuff on connectionism and philosophy. Clark, A. 1990. Connectionism, competence and explanation. British Journal for the Philosophy of Science 41:195-222. Connectionism separates processing from competence. Instead of hopping down Marr's levels (theory->process), connectionism goes (1) task (2) low-level performance (3) extract theory from process. Cute. Clark, A. 1991. In defense of explicit rules. In (W. Ramsey, S. Stich, & D. Rumelhart, eds) _Philosophy and Connectionist Theory_. Erlbaum. Argues that we need explicit rules for flexibility, adaptibility, and representational redescription. With remarks on eliminativism. Cummins, R. & Schwarz, G. 1987. Radical connectionism. Southern Journal of Philosophy Supplement 26:43-61. On computation and representation in AI and connectionism, and on problems for radical connectionism in reconciling these without denying representation or embracing mystery. Cummins, R. & Schwarz, G. 1991. Connectionism, computation, and cognition. In (T. Horgan & J. Tienson, eds) _Connectionism and the Philosophy of Mind_. Kluwer. Explicates computationalism, and discusses ways in which connectionism might end up non-computational: if causal states cross-classify representational states, or if transitions between representations aren't computable. Davies, M. 1989. Connectionism, modularity and tacit knowledge. British Journal for the Philosophy of Science 40:541-55. Argues that connectionist networks don't have tacit knowledge of modular theories (as representations lack the appropriate structure, etc.). Globus, G.G. 1992. Derrida and connectionism: Differance in neural nets. Philosophical Psychology 5:183-97. Hatfield, G. 1990. Gibsonian representations and connectionist symbol-processing: prospects for unification. Psychological Research 52:243-52. Gibson is compatible with connx: rule-instantiation without rule-following. Horgan, T. & Tienson, J. (eds) 1991. _Connectionism and the Philosophy of Mind_. Kluwer. Horgan, T. & Tienson, J. 1992. Cognitive systems as dynamic systems. Topoi 11:27-43. Humphreys, G.W. 1986. Information-processing systems which embody computational rules: The connectionist approach. Mind and Language 1:201-12. Legg, C.R. 1988. Connectionism and physiological psychology: A marriage made in heaven? Philosophical Psychology 1:263-78. Lloyd, D. 1989. Parallel distributed processing and cognition: Only connect? In _Simple Minds_. MIT Press. An overview: local/distributed/featural representations; explanation in connectionism (how to avoid big mush); relation to neuroscience; explicit representations of rules vs weight matrices. Lycan, W.G. 1991. Homuncular functionalism meets PDP. In (W. Ramsey, S. Stich, & D. Rumelhart, eds) _Philosophy and Connectionist Theory_. Erlbaum. On various ways in which connectionism relates to representational homuncular functionalism, e.g. on implementation, eliminativism, and explanation. Ramsey, W. & Stich, S.P. 1990. Connectionism and three levels of nativism. Synthese 82:177-205. How connectionism bears on the nativism debate. Conclusion: not too much. Ramsey, W., Stich, S.P. & Rumelhart, D. (eds) 1991. _Philosophy and Connectionist Theory_. Erlbaum. Rosenberg, J. 1989. Connectionism and cognition. Bielefeld Report. Criticism of Churchland's connectionist epistemology. Shanon, B. 1992. Are connectionist models cognitive? Philosophical Psychology. In some senses of "cognitive", yes; in other senses, no. Phenomenological, theoretical, and sociological perspectives. Toward meaning-laden models. Sterelny, K. 1990. Connectionism. In _The Representational Theory of Mind_. Blackwell. 4.5g Foundational Empirical Issues [10] ---------------------------------- Cliff, D. 1990. Computational neuroethology: A provisional manifesto. Manuscript. Criticizes connectionism for not being sufficiently rooted in neuroscience, and for not being grounded in the world. Dawson, M.R.W., & Schopflocher, D.P. 1992. Autonomous processing in parallel distributed processing networks. Philosophical Psychology 5:199-219. Hadley, R. 1990. Connectionism, rule-following, and symbolic manipulation. Proc AAAI. Some rules are learnt so quickly that representation must be explicit. Hanson, S. & Burr, D. 1990. What connectionist models learn. Behavioral and Brain Sciences. What's new to connx is not learning or representation but their interaction. Kaplan, S., Weaver, M. & French, R.M. 1990. Active symbols and internal models: Towards a cognitive connectionism. AI and Society. Addresses behaviorist/associationist charges. Connectionism needs recurrent circuits to support active symbols. Kirsh, D. 1987. Putting a price on cognition. Southern Journal of Philosophy Supplement 26:119-35. On the importance of variable binding, and why it's hard with connectionism. Lachter, J. & Bever, T. 1988. The relation between linguistic structure and associative theories of language learning. Cognition 28:195-247. Criticism of connectionist language models. They build in too much. Nelson, R. 1989. Philosophical issues in Edelman's neural darwinism. Journal of Experimental and Theoretical Artificial Intelligence 1:195-208. On the relationship between ND, PDP and AI. All are computational. Oaksford, M., Chater, N. & Stenning, K. 1990. Connectionism, classical cognitive science and experimental psychology. AI and Society. Connectionism is better at explaining empirical findings about mind. Pinker, S. & Prince, A. 1988. On language and connectionism. Cognition 28:73-193. Extremely thorough criticism of the R&M past-tense-learning model, with arguments on why connectionism can't handle linguistic rules. 4.6 Symbols and Symbol Systems [5] ------------------------------ Harnad, S. 1990. The symbol grounding problem. Physica D 42:335-346. AI symbols are empty and meaningless. They need to be "grounded" in something, e.g. sensory projection. Maybe connectionism can do the trick? Harnad, S. 1992. Connecting object to symbol in modeling cognition. In (A. Clark & R. Lutz, eds) _Connectionism in Context_. Springer-Verlag. On the limitations of symbol systems, and the potential for grounding symbols in sensory icons and categorical perception, e.g. with neural networks. Kosslyn, S.M. & Hatfield, G. 1984. Representation without symbol systems. Social Research 51:1019-1045. Newell, A. 1980. Physical symbol systems. Cognitive Science 4:135-83. Newell, A. & Simon, H.A. 1981. Computer science as empirical inquiry: Symbols and search. Communications of the Association for Computing Machinery 19:113-26. Reprinted in (J. Haugeland, ed) _Mind Design_. MIT Press. On computer science, AI, & the Physical Symbol System Hypothesis. 4.7 Methodological Foundations of AI [21] ------------------------------------ Brooks, R. 1991. Intelligence without representation. Artificial Intelligence 47:139-159. We don't need explicit representation; the world can do the job instead. Use embodied, complete systems, starting simple and working incrementally. Chalmers, D.J., French, R.M. & Hofstadter, D.R. 1992. High-level perception, representation, and analogy: A critique of AI methodology. Journal of Experimental and Theoretical Artificial Intelligence. AI must integrate perception and cognition in the interest of flexible representation. Current models ignore perception and the development of representation, but this cannot be separated from later cognitive processes. Clark, A. 1986. A biological metaphor. Mind and Language 1:45-64. AI should look at biology. Clark, A. 1987. The kludge in the machine. Mind and Language 2:277-300. Clark, A. and Toribio, J. 1994. Doing without representing. Synthese. A discussion of anti-representationalism in situated robotics and the dynamic systems movement (Brooks, Beer, van Gelder). These arguments appeal to overly simple domains, and a modest notion of representation survives. Dreyfus, H.L. 1979. A framework for misrepresenting knowledge. In (M. Ringle, ed) _Philosophical Perspectives in Artificial Intelligence_. Humanities Press. On the problems with context-free symbolic representation. Dreyfus, H.L. 1981. From micro-worlds to knowledge: AI at an impasse. In (J. Haugeland, ed) _Mind Design_. MIT Press. Micro-worlds don't test true understanding, and frames and scripts are doomed to leave out too much. Explicit representation can't capture intelligence. Hadley, R. 1991. The many uses of `belief' in AI. Minds and Machines 1:55-74. Various AI approaches to belief: syntactic, propositional/meaning-based, information, tractability, discoverability, and degree of confidence. Haugeland, J. (ed) 1981. _Mind Design_. MIT Press. 12 papers on the foundations of AI and cognitive science. Haugeland, J. 1981. Semantic engines: An introduction to mind design. In (J. Haugeland, ed) _Mind Design_. MIT Press. Kirsh, D. 1991. Foundations of AI: The big issues. Artificial Intelligence 47:3-30. Identifying the dividing lines: pre-eminence of knowledge, embodiment, language-like kinematics, role of learning, uniformity of architecture. Kirsh, D. 1991. Today the earwig, tomorrow man? Artificial Intelligence 47:161-184. Marr, D. 1977. Artificial intelligence: A personal view. Artificial Intelligence 9:37-48. AI usually comes up with Type 2 (algorithmic) theories, when Type 1 (info processing) theories might be more useful -- at least if they exist. McDermott, D. 1981. Artificial intelligence meets natural stupidity. In (J. Haugeland, ed) _Mind Design_. MIT Press. Problems in AI methodology: wishful mnemonics, oversimplifying natural language concepts, and never implementing programs. Entertaining. McDermott, D. 1978. Tarskian semantics, or no notation without denotation. Cognitive Science 2:277-82. On the virtues of denotational semantics for AI. Notation without denotation, as found in many AI systems, leads to castles in the air. McDermott, D. 1987. A critique of pure reason. Computational Intelligence 3:151-60. Criticism of logicism (i.e. reliance on deduction) in AI. Nilsson, N. 1991. Logic and artificial intelligence. Artificial Intelligence 47:31-56. Birnbaum, L. 1991. Rigor mortis: A response to Nilsson's `Logic and artificial intelligence'. Artificial Intelligence 47:57-78. Partridge, D. & Wilks, Y. (eds) 1990. _The Foundations of Artificial Intelligence: A Sourcebook_. Cambridge University Press. Lots of papers on various aspects of AI methodology. Quite thorough. Pylyshyn, Z.W. 1979. Complexity and the study of artificial and human intelligence. In (M. Ringle, ed) _Philosophical Perspectives in Artificial Intelligence_. Humanities Press. Ringle, M. (ed) 1979. _Philosophical Perspectives in Artificial Intelligence_. Humanities Press. 10 papers on philosophy of AI, psychology and knowledge representation. Wilks, Y. 1990. Form and content in semantics. Synthese 82:329-51. Criticism of McDermott's views on semantics, logic and natural language. 4.8 Computation and Semantics [12] ------------------------------ Fodor, J.A. 1978. Tom Swift and his procedural grandmother. Cognition 6:229-47. Reprinted in _RePresentations_ (MIT Press, 1980). Against procedural semantics; it's a rerun of verificationism. Hadley, R. 1990. Truth conditions and procedural semantics. In (P. Hanson, ed) _Information, Language and Cognition_. University of British Columbia Press. Johnson-Laird, P. 1977. Procedural semantics. Cognition 5:189-214. Johnson-Laird, P. 1978. What's wrong with Grandma's guide to procedural semantics: A reply to Jerry Fodor. Cognition 9:249-61. Dietrich, E. 1988. Computers, intentionality, and the new dualism. Manuscript. Rapaport, W. 1988. Syntactic semantics: Foundations of computational natural language understanding. In (J. Fetzer, ed) _Aspects of AI_. Kluwer. Smith, B. 1988. On the semantics of clocks. In (J. Fetzer, ed) _Aspects of AI_. Kluwer. Smith, B. 1987. The correspondence continuum. CSLI-87-71. Wilks, Y. 1982. Some thoughts on procedural semantics. In (W. Lehnert, ed) _Strategies for Natural Language Processing_. Erlbaum. Winograd, T. 1985. Moving the semantic fulcrum. Linguistics and Philosophy 8:91-104. Woods, W. 1981. Procedural semantics as a theory of meaning. In (A. Joshi, B. Weber, & I. Sag) _Elements of Discourse Understanding_. Cambridge University Press. Woods, W. 1986. Problems in procedural semantics. In (Z. Pylyshyn & W. Demopolous, eds) _Meaning and Cognitive Structure_. Ablex. With commentaries by Haugeland, J.D. Fodor. 4.9 The Frame Problem [11] ---------------------- Dennett, D.C. 1984. Cognitive wheels: The frame problem of AI. In (Hookaway, ed) _Minds, Machines and Evolution_. Cambridge University Press. General overview. Dreyfus, H.L. & Dreyfus, S. 1987. How to stop worrying about the frame problem even though it's computationally insoluble. In (Z. Pylyshyn, ed) _The Robot's Dilemma_. Ablex. FP is an artifact of computational explicitness. Contrast human commonsense know-how, with relevance built in. Comparison to expert/novice distinction. Fodor, J.A. 1987. Modules, frames, fridgeons, sleeping dogs, and the music of the spheres. In (Z. Pylyshyn, ed) _The Robot's Dilemma_. Ablex. FP is Hamlet's problem: when to stop thinking. It's equivalent to the general problem of non-demonstrative inference. Haugeland, J. 1987. An overview of the frame problem. In (Z. Pylyshyn, ed) _The Robot's Dilemma_. Ablex. The FP may be a consequence of the explicit/implicit rep distinction. Use "complicit" reps instead, and changes will be carried along intrinsically. Hayes, P. 1987. What the frame problem is and isn't. In (Z. Pylyshyn, ed) _The Robot's Dilemma_. Ablex. FP is a relatively narrow problem, Some, e.g. Fodor, wrongly equate FP with the "Generalized AI Problem". Janlert, L. 1987. Modeling change: The frame problem. In (Z. Pylyshyn, ed) _The Robot's Dilemma_. Ablex. Lormand, E. 1990. Framing the frame problem. Synthese 82:353-74. Criticizes Dennett's, Haugeland's and Fodor's construals of the FP. Maloney, J.C. 1988. In praise of narrow minds. In (J. Fetzer, ed) _Aspects of AI_. D. Reidel. McCarthy, J. & Hayes, P. 1969. Some philosophical problems from the standpoint of artificial intelligence. In (Meltzer & Michie, eds) _Machine Intelligence 4_. Edinburgh University Press. McDermott, D. 1987. We've been framed: Or, Why AI is innocent of the frame problem. In (Z. Pylyshyn, ed) _The Robot's Dilemma_. Ablex. Solve frame problem by using the sleeping-dog strategy -- keeping things fixed unless there's a reason to suppose otherwise. Pylyshyn, Z.W. (ed) 1987. _The Robot's Dilemma_. Ablex. Lots of papers on the frame problem. 4.10 Analog and Digital Processing [5] ---------------------------------- Demopoulos, W. 1987. On some fundamental distinctions of computationalism. Synthese 70:79-96. Fodor, J.A. & Block, N. 1973. Cognitivism and the analog/digital distinction. Manuscript. Goodman, N. 1968. _Languages of Art_. Bobbs-Merrill. Haugeland, J. 1981. Analog and analog. Philosophical Topics 12:213-26. Lewis, D. 1971. Analog and digital. Nous 5:321-7. 4.11 Levels of Analysis (Marr, etc) [9] ----------------------------------- Foster, C. 1990. _Algorithms, abstraction and implementation_. Academic Press. Outlines a theory of the equivalence of algorithms. Horgan, T. & Tienson, J. 1992. Levels of description in nonclassical cognitive science. Philosophy 34, Supplement. Generalizes Marr's levels to: cognitive state-transitions, mathematical state-transitions, implementation. Discusses these with respect to connectionism, dynamical systems, and computation below the cognitive level. Houng, Y. 1990. Classicism, connectionism and the concept of level. Dissertation, Indiana University. On levels of organization vs. levels of analysis. Marr, D. 1982. _Vision_. Freeman. Defines computational, algorithmic and implementational levels. McClamrock, R. 1990. Marr's three levels: a re-evaluation. Minds and Machines 1:185-196. On different kinds of level-shifts: organizational and contextual changes. There are more than three levels available. Newell, A. 1982. The knowledge level. Artificial Intelligence 18:81-132. Newell, A. 1986. The symbol level and the knowledge level. In (Z. Pylyshyn & W. Demopolous, eds) _Meaning and Cognitive Structure_. Ablex. With commentaries by Smith, Dennett. Peacocke, C. 1986. Explanation in computational psychology: Language, perception and level 1.5. Mind and Language 1:101-23. Psychological explanation is typically somewhere *between* the computational and algorithmic levels. Sticklen, J. 1989. Problem-solving architectures at the knowledge level. Journal of Experimental and Theoretical Artificial Intelligence 1:233-247. 4.12 Philosophy of AI, Misc [24] --------------------------- Buchanan, B. 1988. AI as an experimental science. In (J. Fetzer, ed) _Aspects of AI_. D. Reidel. Burks, A.W. 1973. Logic, computers, and men. Proceedings and Addresses of the American Philosophical Association 46:39-57. Arguing that a finite deterministic automaton can perform all natural human functions. With remarks on the logical organization of computers. Dennett, D.C. 1978. AI as philosophy and as psychology. In (M. Ringle, ed) _Philosophical Perspectives on Artificial Intelligence_. Humanities Press. Reprinted in _Brainstorms (MIT Press, 1978). AI as detailed armchair psychology and as thought-experimental epistemology. Implications for mind: e.g. a solution to the problem of homuncular regress. Dretske, F. 1985. Machines and the mental. Proceedings and Addresses of the American Philosophical Association 59:23-33. Dretske, F. 1993. Can intelligence be artificial? Philosophical Studies 71:201-16. Intelligence requires not just action or thought, but the governance of action by thought, which requires a history. "Wired-up" systems lack the explanatory connection between thought and action, so are not intelligent. Dreyfus, H.L. 1972. _What Computers Can't Do_. Harper and Row. Computers follow rules, people don't. Dreyfus, H.L. & Dreyfus, S.E. 1988. Making a mind versus modelling the brain: AI at a crossroads. Daedalus. History of AI (boo) and connectionism (qualified hooray). And Husserl/ Heidegger/Wittgenstein, of course. Quite nice. Dunlop, G. 1990. Conceptual dependency as the language of thought. Synthese 82:275-96. Relates Schank's CD to Fodor's LOT. Glymour, C. 1988. AI is philosophy. In (J. Fetzer, ed) _Aspects of AI_. D. Reidel. Haugeland, J. 1979. Understanding natural language. Journal of Philosophy 76:619-32. Reprinted in (W. Lycan, ed) _Mind and Cognition (Blackwell, 1990). AI will need holism: interpretational, common-sense, situational, existential. Henley, T.B. 1990. Natural problems and artificial intelligence. Behavior and Philosophy 18:43-55. On the philosophical importance of criteria for intelligence. With remarks on Searle, the Turing test, attitudes to AI, and ethical considerations. Krellenstein, M. 1987. A reply to `Parallel computation and the mind-body problem'. Cognitive Science 11:155-7. Thagard 1986 is wrong: speed and the like make no fundamental difference. With Thagard's reply: it makes a difference in practice, if not in principle. Kukla, A. 1989. Is AI an empirical science? Analysis 49:56-60. No, AI is an a priori science that uses empirical methods; its status is similar to that of mathematics. Manning, R.C. 1987. Why Sherlock Holmes can't be replaced by an expert system. Philosophical Studies 51:19-28. An expert system would lack Holmes' ability to raise the right questions, sort out relevant data, and determine what data are in need of explanation. McCarthy, J. 1979. Ascribing mental qualities to machines. In (M. Ringle, ed) _Philosophical Perspectives in Artificial Intelligence_. Humanities Press. Preston, B. 1993. Heidegger and artificial intelligence. Philosophy and Phenomenological Research 53:43-69. On the non-represented background to everyday activity, and environmental interaction in cognition. Criticizes cognitivism, connectionism, looks at Agre/Chapman/Brooks, ethology, anthropology for support. Robinson, W.S. 1991. Rationalism, expertise, and the Dreyfuses' critique of AI research. Southern Journal of Philosophy 29:271-90. Defending limited rationalism: i.e. a theory of intelligence below the conceptual level but above the neuronal level. Robinson, W.S. 1992. _Computers, Minds, and Robots_. Temple University Press. Russell, S. 1991. Inductive learning by machines. Philosophical Studies 64:37-64. A nice paper on the relationship between techniques of theory formation from machine learning and philosophical problems of induction and knowledge. Sloman, A. 1978. _The Computer Revolution in Philosophy_. Harvester. All about how the computer should change the way we think about the mind. Thagard, P. 1986. Parallel computation and the mind-body problem. Cognitive Science 10:301-18. Parallelism does make a difference. Some somewhat anti-functionalist points. Thagard, P. 1990. Philosophy and machine learning. Canadian Journal of Philosophy 20:261-76. Thagard, P. 1991. Philosophical and computational models of explanation. Philosophical Studies 64:87-104. A comparison of philosophical and AI approaches to explanation: deductive, statistical, schematic, analogical, causal, and linguistic. Winograd, T. & Flores, F. 1987. _Understanding Computers and Cognition_. Addison-Wesley. -- Compiled by David Chalmers, Department of Philosophy, Washington University. (c) 1994 David J. Chalmers.