What is this? 1998 3 Can Physical Symbol Systems Think? Can a symbolic knowledge base 83 Søren Kierkegaard, 1844, as articulated by Herbert Dreyfus, 1972 The leap. There are times Can symbolic representations account represent human understanding? when a person makes a "leap" to a new sphere of existence, which infuses his or her being with a new order of significance. for human thought? 106 George Lakoff, 1987 The objectivist account of cognition is inconsistent. The objectivist account makes inconsistent assumptions. 1. The meaning of a 2. The meaning of Such leaps are so radical that sentence is a the parts cannot 82 Hubert Dreyfus, 1972 Wow! I'm not the same person afterwards we cannot imagine 105 George Lakoff, 1987 function that assigns ic t be changed ion Creative discoveries I was yesterday! I've made a how life could have ever been Objectivism is in conflict with empirical studies of natural a truth value to the without changing I'm in love! 103 con trad radically restructure creative discovery about otherwise. The History and Status of the Debate — Map 3 of 7 81 The representationalist assumption. human categories. The objectivist paradigm—which influences sentence for each the meaning of The knowledge base assumption. Symbolic human knowledge. Large That changes myself. everything! Symbol structures are internal representations most contemporary cognitive science—rests on a classical theory possible world. the whole. (This data can be organized into a knowledge base that knowledge bases organize and of categories that is disproved by a wide body of empirical evidence (This is a standard is a requirement process data in a relatively of external reality. They are made up of represents the entirety of human understanding. symbols, and are operated on by rules, search, concerning basic-level concepts, kinesthetic image schemas, definition of of any theory of However, when AI researchers develop knowledge fixed manner. Human metaphorical concepts, metonymic models, and others. meaning in meaning.) knowledge, by contrast, is and other psychological processes. Symbolic bases, they generally focus on some particular is representations have a constituent structure, in objectivist domain, such as the domain of furniture, animals, disputed subject to radical restructuring The cat is semantics.) on the basis of "creative is supported by that the meaning of a given representation is a restaurants, and so forth. A knowledge base is by on the mat. An Issue Map™ Publication written in a "representation language" (such as discoveries," which can alter a function of the meaning of its constituent parts. Disputed by "On" is an To see the contradiction, notice the following: Changing the meaning of the parts of a sentence should change the meaning of the sentence This info-mural is one of seven the predicate calculus or LISP). A good knowledge person's entire understanding of embodied the world. Such fundamental "The Front-End Assumption Is Dubious," as a whole (by 2). This implies that the truth value of this sentence will base supports inference, allowing the computer Map 1, Box 74. concept. also change for some possible world (by 1). But, we can construct a to draw conclusions from available information. shifts can take place at personal, conceptual, and 88 Hubert Dreyfus, 1992 sentence in which the meaning of its parts changes but its truth value For example, a knowledge base representing the Notes: remains the same in all possible worlds. So, we have a contradiction. cultural levels. CYC will not be able to domain of furniture should support the inference Note: In this context Dreyfus demonstrate common sense. • Symbol structures in this sense are often Note: In unpacking the claim, Lakoff draws on an argument by Putnam “argumentation maps” in a that a chair is something people sit on. Start Here Can the elements of thinking be Postulates of Dreideggereanism Wood comes from also discusses Thomas Kuhn's notion of paradigm shift. CYC is inconsistent with the phenomenology of skilled coping. The project of encoding human knowledge in a vast database is beset by the referred to as mental representations or classical representations. • For more on this classical AI theory of representation, see Newell and Simon (1976), Fodor (1975), and Pylyshyn (1984). 104 As articulated by George Lakoff, 1987 (1981), which extends Lowenheim-Skolem's theorem from first-order logic to higher-order logic. series that explores Turing’s represented in discrete symbolic form? Dreideggereanism is Hubert Dreyfus's application of Heideggerean phenomenology to issues in AI and philosophy of mind. 1. Our basic way of being in the world is coping with equipment (rather than relating to is supported by is made of Chair Tree is 84 John McCarthy and Patrick J. Hayes, 1969 The frame problem. General reasoning requires that a system make relevant inferences while excluding irrelevant inferences. This process requires "frame axioms," which specify those properties Unmapped Territory Additional following problems. • Human beings cope successfully without consulting facts in a knowledge base. • Skills and know-how resist • Much of the debate between connectionism and classical AI is focused on the issue of mental representation. One of AI's major charges against connectionism is is supported by The objectivist account of cognition. Symbol structures are meaningful by virtue of their is disputed by is supported by 1 Alan Turing, 1950 the world by way of mental representations). is made of disputed of the world that remain unchanged when an action is carried out. that connectionist networks can't model correspondence with entities I believe that at the end 41 Hubert Dreyfus, 1972 by frame problem representation as propositional and categories in the world. question: “Can computers think The context antinomy. For a machine to understand sentences in a is made of Specifying such axioms in advance cannot be carried out in any knowing-that. constituent structure. See the "Can arguments • The more human beings know, the 2. Coping skills can't be formalized. of the century ... one natural language, it must place those sentences in a context. However, is used for Table simple way. connectionist networks exhibit systematicity?" They are mirrors of nature, 107 Leopold Lowenheim, 1915; Thoralf Skolem, 1922 Yes, machines can will be able to speak of 32 The symbolic data assumption. The elements because machines use explicit bits of data, they run up against an antinomy. 3. Coping takes place against a background of general familiarity. Sitting on Plastic Notes: • McCarthy and Hayes are not, strictly speaking, disputing the knowledge base assumption. They raise the frame problem as an more quickly they bring the relevant knowledge to bear in a situation. In a computer, the reverse is true. arguments on Map 5. which "re-present" or "make present again" what exists in external reality. Without their The Lowenheim-Skolem theorem. If a countable collection of sentences in a first-order language describes a model (that is, some state of affairs in machines thinking is the world), then it describes more than one model. (or will be able to) of thinking are represented in discrete symbolic form. 4. Familiarity is a kind of coping, but it is not directed at any particular task. Heidegger important (and solvable) issue for AI researchers to deal with. connection to the world, and/or will they ever be able disputed Note: Hilary Putnam's version (1981) of this argument employs a strengthened Beliefs, desires, goals, values, plans, and so forth are Either Or calls this background familiarity an understanding of being. • This definition of the frame problem is itself the subject of dispute. symbol structures are without expecting to be is supported by represented by such data and by their combination into meaningless. by version of the Lowenheim-Skolem theorem and claims that the theorem shows that any statement in natural language has an unintended interpretation think. A computational contradicted. higher-order symbol structures like knowledge bases. is disputed There is a broader context of facts that determines which facts are relevant to a given There is an ultimate context that requires no interpretation, 5. On the basis of familiarity, human beings are able to determine what is relevant to what, what to pay attention to, and what to do in any given situation. 85 Francisco Varela, Evan Thompson, and Eleanor Rosch, 1991 I built this without a if it has any interpretation at all. database! to?” Argumentation mapping is Independent domains can't be aggregated to model common system can possess all Tree Chair by sentence, in which case in which case we are forced to postulate a set 6. Expertise consists of responding to a situation similar to one that has occurred in the past in a way similar to a response that has worked in the past. is disputed by sense. Current models from the representationalist perspective attempt to reconstruct common sense from the combination of a variety of small The cat is on the mat is disputed 108 Rodney Brooks, 1991 The world is its own best representation. Modeling intelligence requires that a system be directly interfaced with important elements of Table that context must itself be of facts that have fixed relevance, regardless of the 7. Similarity is basic and cannot be analyzed in terms of shared features. 8. Expertise is arrived at in 5 stages, beginning with rule-like responses to specific domains, but the world isn't composed of separate discrete domains of knowledge that can be represented in isolation from each other. To provide an adequate theory of common sense, we must provide an account of is by the world through sensors and actuators. These "layers" of interaction with the environment approximate the a method that provides: interpreted, so that we face situation. But no such set of sensorimotor dynamics of the human body and allow the human thinking or an infinite regress of broader and broader contexts. facts exist. features and ending with responses to whole situations. 9. Situations cannot be specified in terms of features. context-dependent know-how. disputed by Postulates of Experiential Realism 1. Experiential realism is "experiential" in that it focuses on: world to act as its own representation. This removes the necessity of using the inefficient global representations favored by AI and has the advantages of: • actual and potential experiences • better response time, because no global representation has understanding. Adapted from Dreyfus (1972, 1991) and Dreyfus and Dreyfus (1986). 86 We've got to "bite the • genetically acquired makeup of the organism to be changed whenever the environment changes; - a method for portraying major Alan Turing is supported by In Either Case Explicit data cannot account for the understanding of natural language. Humans avoid this antinomy because they recognize the present situation as a continuation of past situations, and on that basis determine what 40 Martin Heidegger, 1927 is The problem of commonsense knowledge. Human commonsense knowledge is so vast that it can never be adequately represented by propositional data in a knowledge base. Dreyfus factors this problem into three parts. "1. How everyday knowledge must be organized so that one can bullet" and enter in lots of information. 87 Douglas Lenat and Edward A. Feigenbaum, 1991 Computers will be able to demonstrate is supported by • the organism's interactions in the social and physical environment (p. xv). 2. Experiential realism is "realist" in postulating that: • there is a real world • reality places constraints on concepts • increased robustness, because the system is less likely to crash due to some unpredictable change in the world. Rodney Brooks disputed • truth goes beyond mere internal coherence philosophical, political, and is relevant to understanding a sentence. The spatiality of equipment. make inferences from it. common sense with a large enough database. Human beings organize space into by 2. How skills or know-how can be represented as knowing-that. is A large database containing more than 10 million is supported by • there is stable knowledge of the world (p. xv). is supported by areas of nearness and "farness" 3. How relevant knowledge can be brought to bear in particular disputed statements of facts about everyday life, history, relative to their needs and situations" (1992, p. xviii). by physics, and so forth will be able to exhibit common 3. There is more to thought than just representation. Thought is also: concerns. They do not organize This last problem has been called the access problem, that is, sense. • embodied, in that "the structures used to put together our conceptual systems grow out of 37 Hubert Dreyfus, 1972 is supported by space into a three-dimensional the problem of how to efficiently access data in a knowledge base. bodily experience and make sense in terms of it" (p. xiv); pragmatic debates is supported by 33 As articulated by Hubert Dreyfus, 1972 Reductionistic science paradigm. To model and understand the world, divide phenomena into simple elements and relations. Scientists, from Galileo onwards, is disputed by is disputed by Explicit values cannot organize a field of experience. Human interest organizes a field of experience that cannot be captured by explicit goals and values. • Specific goals must be checked at preset intervals, but human concerns pervade experience. system of discrete coordinates. Martin Heidegger The access problem is also relevant to heuristic search (see the "Heuristic Search" arguments on this map), because in large knowledge bases the problem of accessing information is also the problem of searching for it. com • mon • sense knowl • edge: Our everyday understanding of the world. Generally, such knowledge consists in pretheoretical information that seems obvious Implemented Model 89 Doug Lenat and R. V. Guha, 1990 • imaginative, in that concepts not grounded directly in experience (e.g., metaphorical concepts) employ conceptual structures that go beyond the literal representation of reality. 4. Classical categories are inadequate. They are like containers: their members are either in or out. Cognitive models, on the other hand, obey nonclassical "fuzzy" logics, exhibiting Implemented Model - a summary of an ongoing, • When concerns are made explicit they lose their pervasive character. 39 Martin Heidegger, 1927 degrees of membership. have analyzed the world using this reductionist program when explicitly stated: for example, our knowledge that tables generally have 4 legs CYC. CYC is a massive database and have made significant progress. • Human needs only become explicit after they have been fulfilled. is supported by Humans beings are called to take a stand on who they are. Human beings and are something people put things on, or that people generally bring gifts to birthday containing millions of statements and a take a stand on who they are by living "for the sake of" being teachers, doctors, and Out Peripheral 109 Cog Shop, 1998 so forth. Such "for-the-sake-of-whiches" structure human activity in a pervasive and 91 George Lakoff, 1987 parties. It has been said that commonsense knowledge consists not of the information complex inference engine that represents COG. Led by Rodney Brooks, the Cog Shop nondeterminate way. The predicate calculus contained in an encyclopedia, but rather in all the information necessary to read and commonsense knowledge well enough to In has developed a humanoid robot composed of a intelligently process data. Central cannot capture human understand an encyclopedia in the first place. series of interconnected (but independently major philosophical debate of 34 Christine A. Skarda and Walter J. Freeman, 1987 reasoning. Many Nonsymbolic explanations of smelling. Rabbits discriminate odors by a process of chaotic self-organizing activity at the level functional) systems controlled by a set of parallel is supported by History of the Symbolic Data Assumption knowledge-based systems 5. It is important to focus on conceptual structures and cognitive models, which involve a variety processors. COG currently consists of a trunk, of the neural tissue. Symbols and symbol structures don't explain this process as well as the mathematics of nonlinear dynamics and encode information with of phenomena. 2 Allen Newell and Herbert Simon, 1976 A physical symbol system has the head, arms, and a sophisticated visual system. • Prototype effects: Differences among category members such that some members are chaos does. some version of the Physical symbol systems can think. necessary and sufficient means for Note: Skarda and Freeman use similar reasoning to dispute "The Rule-Following Assumption," (Box 46) (because neurons don't follow The idea that the world can be understood as composed of discrete elements has a long history in the philosophical is 90 James Lighthill, 1973 Planned additions include hands, vocalization Thinking in a physical symbol system is a formal computational general intelligent action. By predicate calculus, which Past disappointments. The number of facts necessary to capture the entirety of commonsense knowledge is more central than others. ability, a vestibular system, and skin. COG is rules) and the central control aspect of "The Brain Has a von Neumann Architecture," (Box 12) (because neural activity is self-organizing). tradition. disputed the 20th century is based on the classical = process characterized by "necessary" we mean that any They also argue against representationalism in general, with its assumption of the existence of plans, goals, scripts, and so forth. insurmountably large. Machine agents can only cope successfully in limited domains, such as the game of checkers, where a intended to be as fully a part of the real world as • rule-governed symbol manipulation Rule-governed by concept of a category. But Central: Thinking = manipulation of symbolic system that exhibits general Plato, for example, believed that all true knowledge (knowledge of piety, justice, the good, etc.) must be statable small number of facts exhaustively describe the agent's world. To scale up from such artificial worlds to the real world simply Peripheral: possible, both by having a body and by interacting • the drawing of inferences from large knowledge bases intelligence will prove upon analysis in explicit definitions that would act like rules telling us how to behave. René Descartes proposed that certainty the classical view of is supported by by adding more and more facts is not possible, because this leads to a combinatorial explosion of the number of ways four-legged bean bag in a real (as opposed to a "toy") environment. • heuristic search of data structures representational structures categories has been chair to be a physical symbol system. By in knowledge could only be obtained by relying on what he called "clear and distinct ideas," that is, ideas that in which elements can be grouped in a knowledge base. Because of these problems, AI workers have little more than "past • operations on representational structures "sufficient" we mean that any is are clear in themselves and distinct from other ideas. disproved by empirical disappointments" to their credit. • planning and goal-directed activity 38 John Dewey, 1922 evidence. • Basic level categorization: Categories that are cognitively basic are "'in the middle' of a - a new way of doing intellec- physical symbol system of sufficient disputed Goals pervade Supported by The symbolic processes that constitute thinking are formal in that Note: Also, see sidebar, "The Lighthill Report," Box 74. general-to-specific hierarchy" (p. 13). size can be organized further to by activity. Humans Gottfried Leibniz, one of the originators of the idea that a machine could think, also introduced the notion of "Postulates of Experiential they are independent of any particular physical instantiation. In exhibit general intelligence (p. 16). a monad, which, as a basic component of reality "is nothing but a simple substance that enters into composites— humans, thinking is instantiated in the neurons of the brain. In experience goals as Realism," on this map. Postulates of Subsumption Architecture pervasive elements of simple, that is, without parts" (Leibniz, 1714, p. 1). Note: This argument summarizes a number of Lighthill's opinions, and it represents in early form versions of the problem of computers, thinking is realized in silicon circuits. toy worlds, brittleness, combinatorial explosion, and commonsense knowledge, all of which are well-known problems today. Notes: present activity, rather 1. A subsumption architecture machine (SAM) has sensors and actuators for The British Empiricists (notably, John Locke, George Berkeley, and David Hume) held that all knowledge tual history. • This is a standard interpretation of the field but it is by no means than as fixed ends to specific basic level general interaction with the real world. be worked toward. comes to us through experience in the form of discrete "simple ideas," which combine to form "complex ideas." striped chair furniture shared by everyone. This summary is meant to emphasize those The notion that ideas are discrete continued in the work of Hume, who claimed that "every distinct perception • Kinesthetic image schemas: Recurrent structures of ordinary bodily experience. 35 John Barnden, 1987 2. A SAM is comprised of "layers," that is, "activity-producing subsystems" (p. aspects of artificial intelligence research that are relevant to which enters into the composition of the mind, is a distinct existence, and is different, distinguishable, and Patterns of activity may be "fuzzed" symbols. What does it this map. separable from every other perception, either contemporary or successive" (1739, p. 259). 92 Hubert Dreyfus, 1972; James Lighthill, 1973; and others 146). These layers interact with each other and with the world. • The physical symbol systems hypothesis and functionalism are The "distinctive, stereotypic state of activity" (p. 174) Combinatorial explosion of knowledge. Representing all of the information relevant to an open-ended domain, or to human closely related. The physical symbol systems hypothesis that Skarda and Freeman correlate with the inhalation 36 Christine A. Skarda and Walter J. Freeman, 1987 3. Each layer is a complete functional unit; that is, it functions autonomously of In humans, The same symbol of a learned odor may correspond to the symbolic output In the 19th century, Gottlob Frege invented the first complete formalization of predicate logic. Frege regarded commonsense understanding in general, is an impossible task, because it results in a combinatorial explosion of relevant information. proposes an architecture for simulating and studying intelligence. Fuzzed symbols are not classical symbols. The patterns of thoughts as discrete entities composed of concepts and relations between them. In this century, Bertrand is The number of facts that must be encoded to scale up from a series of small, independent domains to the totality of commonsense other layers, without depending on any central control unit. symbol systems systems can also be of the olfactory system, which the brain then uses in disputed Functionalism is a philosophical position that is used to justify activity that Barnden correlates with symbols are not like classical Russell founded logical atomism, in which the world is understood in terms of atomic propositions that are knowledge is insurmountably large. • Metaphorical concepts: Cross-domain mappings where knowledge from one domain of are instantiated in instantiated in a classical symbol-manipulation style. It is consistent is by 4. Layers can be added gradually to SAMs, which can thus develop incrementally. the brain. computer. this architecture. Functionalism was developed in part as a symbols, which can be composed and manipulated by well-understood either true or false. Ludwig Wittgenstein articulated his early version of logical atomism as follows. "1. The Notes: • Combinatorial explosion also affects the problem of searching databases. See "Combinatorial Explosion of Search," Box 58. response to behaviorism (see the "Is the test, behaviorally or with AI to allow that these symbols "embody a certain disputed the conceptual system is projected onto knowledge in another domain. "We must incrementally build up the capabilities of intelligent systems, having logical operations. Dynamic patterns of neural activity are context world is all that is the case. 1.1. The world is the totality of facts, not of things. 1.11. The world is determined • This problem has been raised in the context of the Turing test (see "Combinatorial Explosion Makes the All-Possible-Conversations operationally construed, a legitimate intelligence test?" arguments amount of 'fuzz'" (p. 174). That is, the states of activity by complete systems at each step of the way and thus automatically ensure that produced by the olfactory system may correspond to dependent and are only roughly correlated with events in the world. by the facts, and by their being all the facts" (1922, p. 5). on Map 2). For more on functionalism, see the "Can functional Note: For discussion of nonclassical symbolic representations, see the Machine Impossible," Map 2, Box 103). She ran like the pieces and their interfaces are valid" (p. 140). states generate consciousness?" arguments on Map 6. rough approximations of classical discrete symbols. the wind ... Allen Newell Herbert Simon "Can connectionist networks exhibit systematicity?" arguments on Map 5. Much of this history was first recounted (in the context of artificial intelligence) by Dreyfus (1972). 5. Each layer enables a specific behavior (e.g., obstacle avoidance, exploration, aluminum-can retrieval, etc.). 6. Any given layer can "subsume" the action of attached layers in order to further contain? • Metonymic concepts: Taking one aspect or part of its own goals, without completely wresting control from the attached layers. something and using it to stand for the thing as a whole The ham 7. A SAM interacts with the "real world," that is, the dynamic world that humans or for some other part of it. sandwich at act in; it does not act in a refined, abstracted, or limited "toy world." "At each table 10 ... step we should build complete intelligent systems that we let loose in the real world with real sensing and real action. Anything less provides a candidate Can physical 43 George Lakoff, 1987 Motivational factors cause different rates of Does mental processing 58 Adapted from George Lakoff (1987). Lakoff's theory draws on the work of a wide range of thinkers, including Mark Johnson, Eleanor Rosch, Ludwig Wittgenstein, and Lotfi Zadeh. with which we can delude ourselves" (p. 140). From Brooks (1991). "Subsumption architecture machine" (SAM) is our own term. Is the relation between hardware Postulates of the Physical Symbol Systems Hypothesis learning. Humans learn more efficiently when rely on heuristic search? Combinatorial explosion of search. When the number of paths in a search space grows exponentially, a combinatorial explosion results: the search becomes is supported by Does thinking symbol systems motivated by supplementary 1. A physical symbol system knowledge, for example, by too long to be carried out, given time and memory • is physical (that is, made up of some physical matter) and software similar to that constraints. To make such searches more efficient, require a body? is supported by metaphorically motivated • is a specific kind of system (that is, a set of components functioning through time in some definable knowledge. But for methods of estimation called "heuristics" have been invented. manner) that manipulates instances of symbols. 2. Symbols can be thought of as elements that are connected and governed by a set of relations called a symbol learn as computers the opposite is the case: they work less efficiently when forced to 57 Heuristic search hypothesis: The solutions to problems are represented as symbol structures. A physical symbol Note: Combinatorial explosion also affects the problem of representing commonsense knowledge. See Does the situated action paradigm between human brains and minds? The heuristic search assumption. Symbolic data is system exercises its intelligence in problem solving by "Combinatorial Explosion of Knowledge," Box 92. structure. The physical instances of the elements (or tokens) are manipulated in the system. deal with supplementary searched using various methods of estimation or "heuristics," search—that is, by generating and progressively knowledge. For example, it 94 George Lakoff, 1987 modifying symbol structures until it produces a solution show that computers can't think? which make the search more efficient. AI models lack the feature of the is humans do? 3. An information process is any process that has symbol structures for at least some of its inputs or outputs. is easier for us to learn about structure (1976, p. 120). disputed the flow of electricity given embodiment of concepts. There is 4. An information processing system is a physical symbol system that consists of information processes. by our knowledge of flowing is supported by 93 evidence that the body is involved in many processes that are considered to be Altogether the seven maps: waters, but in a computer is The disembodied mind assumption. George Lakoff disputed Thinking is an abstract process that does not pure information processing in classical 5. Symbol structures are classified into such knowledge just adds 4 Graham Button, Jeff Coulter, John R. E. Lee, and • data structures more complexity for it to by require the presence of a body. AI. These processes include recent is 3 • programs. deal with. Notes: is discoveries concerning basic-level disputed 110 Terry Winograd and Fernando Flores, 1986 The biological assumption. The brain is the hardware (or is Wes Sharrock, 1995 42 Note: The electricity Newell and Simon • The notion of embodiment is used by many disputed concepts, kinesthetic image schemas, and by The representational tradition is flawed. Classical symbolic AI places too much emphasis on the role of representations in human intelligence. Actual human "wetware") on which the software of the mind is run. Thinking disputed Neurons cannot represent rules of ordinary language. AI assumes Humans learn by situated action theorists. See, for example, the experiential basis of metaphorical - summarize over 800 major that rules are ultimately represented in the brain. But neurons don't provide a 6. A program is a symbol structure that designates the sequence of information processes (including inputs and example is drawn from by thinking only becomes symbolic and representational when normal modes of cognition break down. Future work in AI should concern itself with interactive design, is a symbolic process that is implemented in the neurons of the by outputs) that will be executed by the elementary information processes of the processor. adding symbolic data Gentner and Gentner (1982). heu • ris • tic search: A search that uses special "COG," Box 109. concepts. Tree • On the notion that the mind is disembodied, brain and that can also be implemented in the circuits of a digital symbolic medium in which rules can be inspected and modified, so they are to a knowledge base. Note: Also, see sidebar, "Postulates of task-specific projects, and the role of computers as machines for the facilitation of communication, rather than with representational symbol systems. not appropriate as a medium for the formulation of rules for ordinary language. knowledge about a problem domain to find solutions Note: Winograd and Flores are influenced by Dreyfus's approach. See sidebar, "Postulates of Dreideggereanism," on this map and "The Critique of Artificial Reason," computer. is supported by Both machines and people more efficiently. For example, a search of possible see the "Is the brain a computer?" arguments Experiential Realism," on this map. Note: Also, see the "Is the brain a computer?" arguments on Map We don't use neurons like we use rules. 7. Memory is the component of an information processing system that stores symbol structures. learn by adding symbolically Chair 44 Hubert Dreyfus and Stuart Dreyfus, is Box 76. Table disputed moves in a chess game could be aided by a set of on Map 1 and the "Can functional states 1, the "Is biological naturalism valid?" arguments on Map 4, the Note: See the "Do humans use rules as physical symbol systems do?" arguments encoded information to a 1986 generate consciousness?" arguments on 8. Elementary information processes are transformations that a processor can perform upon symbol structures by heuristics that tell the computer to avoid useless lines of moves in the debates threaded "Are connectionist networks like human neural networks?" on this map. Also, see sidebar, "Postulates of Ordinary Language," on this map. knowledge base. is Computers never move beyond 60 Zenon Pylyshyn, 1974 Map 6. (e.g., comparing and determining equality, deleting, placing in memory, retrieving from memory, etc.). explicit rules. In acquiring skills such attack, to maintain center control, and so forth. arguments on Map 5, and sidebar, "Formal Systems: An Overview," disputed The best heuristics on Map 7. as the ability to drive a car or play chess, 95 John Haugeland, 1995 9. A processor is a component of an information processing system that consists of: by aren't just trial and error. The mind–body–world system. Mind, body, and world Postulates of Situated Action is supported by • a fixed set of elementary information processes, Input humans initially use explicit rules and then Heuristic searches don't is supported by communicate vast amounts of information to one another • a short-term memory that stores the input and output symbol structures of the elementary information advance through a series of stages in which necessarily rely on trial and performance becomes increasingly skilled, across "wide-bandwidth" channels—so much so that they into claims, rebuttals, and processes, and 59 Hubert Dreyfus, 1972 error. Programs can be 1. "Situated actions ... [are] actions taken in does not are integrated into a single system. For example, as a person • an interpreter that determines the sequence of elementary information processes to be executed is fluid, and habituated. At the highest levels Trial and error is different from essential discrimination. Humans are able to intuitively structured so that a heuristic the context of particular concrete 7 David Rumelhart, of expertise, the rules are no longer require drives to San Jose, his or her mind doesn't operate like a similar James McClelland, and FARG, 1986 as a function of the symbol structures in short-term memory. grasp what is essential or inessential about a problem. Symbol systems lack the capacity for method moves the search classical symbol system, solving problems by communicating circumstances" (p. viii–ix). to consulted. Computers, on the other hand, "essential discrimination," and proceed blindly by brute-force trial and error. Adding heuristic is ever closer to a problem 2. "All activity, even the most analytic, is Neurons receive thousands of are tied to the use of explicit rules and can't comparatively tiny instructions at "narrow-bandwidth" times more input than logic gates. 10. The external environment of the system consists of "readable" stimuli. Reading consists of creating internal rules to the system is only a stopgap measure. Humans recognize what is necessary automatically, disputed solution without the need for fundamentally concrete and embodied" (p. 113 move beyond them to more flexible forms by transducers. In a trip to San Jose, the brain, the fingers, and viii). The situated action paradigm. This approach to 125 Martin Heidegger, 1927 Neurons are connected to 1,000–100,000 symbol structures in memory that designate external stimuli. Writing is the operation of emitting the by zeroing in on the essential nature of a problem. redundant backtracking. counterrebuttals of expertise. the road are in constant wide-bandwidth "collaboration," 3. Even "purposeful actions are inevitably artificial intelligence emphasizes the situated and Representations are not involved in other neurons. Logic gates are connected responses to the external environment that are commanded by the internal symbol structures. Note: Dreyfus derives his notion of essential discrimination from the work of Gestalt psychologist Such programs approximate acting as a single, integrated system. to only a handful of other logic gates. Max Wertheimer. human "zeroing in." Zenon Pylyshyn situated action" (p. viii). embodied character of cognition, and the role of concernful activity. In ongoing concernful activity no is Notes: affordances in the acquisition of perceptual information representations are necessary. A carpenter hammers nails • Haugeland supports his view by citing Dreyfus (see "The The difference indicates that the brain Adapted from Newell and Simon (1972, chap. 2). 4. No action is ever "fully anticipated" by plans disputed because the "circumstances of our actions and the control of behavior. without having any explicit representation of the hammer. It is only in does not use the kind of logical circuitry by Body Is Essential to Human Intelligence," Box 96), Gibson is found in digital computers. Proponents include Jerry Fodor, Allen Newell, Herbert Simon, John McCarthy, Zenon Pylyshyn, early Marvin ... are continuously changing around us" (p. cases of "breakdown" that the hammer emerges from the background of (see "Affordances are Features of the Environment," Box - 97-130 arguments and rebuttals is supported by is ix). equipment and is represented as an object with properties. For example, disputed Minsky, Doug Lenat, Edward Feigenbaum, and Pat Hayes. disputed 114), and Brooks (see sidebar, "Postulates of Subsumption is 5. All "our actions, while systematic, are never is supported by if the hammer slips from the carpenter's grasp it might then be seen as an by by Architecture," on this map). Expert disputed object with the property of being too light, too slippery, and so forth. But • Haugeland also argues, with some help from Brooks, that planned in the strong sense" (p. ix). Do physical symbol systems by 6. All or most of life is "primarily ad hoc is supported by prior to the breakdown, the carpenter had no representation of the hammer. General Structure of an Information Processing System the mind–body–world system does not use classical activity (p. ix). Note: See "Computers Never Move Beyond Explicit Rules," Box 44, 8 Jack Copeland, 1993 45 Joseph Rychlak, 1991 representations. Therefore, Haugeland's argument also which clarifies the role representations play in the development of skill. per map 5 Neurons operate like logic gates. The neurons of the brain are is Neurons are diversely structured. Computer logic gates consist of variations on a single structure. The brain, by contrast, consists of many different kinds of neurons. Information Receptors Processor Memory Learning is process of interpretation. Most AI theorists model learning on a Lockean paradigm that takes repetition and contiguity of perceptions as the primary way in which new concepts are learned. For example, we learn how to spell by repeatedly seeing how words are spelled and which letters are Unmapped Territory Novice play chess as humans do? 96 Hubert Dreyfus, 1972 The body is essential to human disputes the representationalist assumption. From Suchman (1987). 124 Lucille Suchman, 1987 Situated action can explain the use of plans without representations. Plans are enacted in the course of practical activities, and are best explained by ethnomethodology, which doesn't - 70 issue areas in the 7 maps similar to the logic gates of a Effectors contiguous to each other. But such a model does not pay Other 111 Humberto Maturana, 1970, as articulated by presuppose the existence of representations in its explanations of social practices. According to disputed Action upon approaches intelligence. digital computer. Their all-or- by sufficient attention to the role of meaning in learning. Learning Terry Winograd and Fernando Flores, 1986 ethnomethodology: Possession of a body is none firing potential gives Environment Redrawn from Newell and Simon (1976). occurs when a mind that reasons dialectically and predicationally to AI learning 62 John Strom and Lindley Darden, 1996 essential to human Biological action is situated. Cognition and • Practices sometimes may be explained merely by materials in the local environment of the them a discrete character and interprets what it perceives in terms of the meaning of what language need to be interpreted biologically. The notion culture. Deep heuristic search has been used to achieve reasoning, pattern allows them to be combined into it perceives. Because computers don't work with meanings, expert performance. Deep Thought (the precursor to 65 Hubert Dreyfus, 1972 recognition, and of representation is inadequate for this task. Cognition is • A difference in practices does not necessarily mean a difference in plan or internal representation. • The structure of systems of practice are not essentially ordered by rules or norms that can be - 32 sidebars history and further the AND gates, OR gates, and interpretation and the relevant kind of learning is impossible Humans see the chess board as a Gestalt and thinking are best described as a history of "structural other logic gates that underlie Pyramidal cell Purkinje cell Retinal bipolar cell for them. Deep Blue) plays chess at the grandmaster level by interaction. Understand- 98 Zenon Pylyshyn, 1974 disputed considering more moves than any previous system. Its whole. Humans see a chess board as an coupling" between an organism and its environmental ing what a chair is, for The body is not essential to by digital computation. Note: Rychlak's further arguments about artificial intelligence niche. can be found in the "Can physical symbol systems think success demonstrates the effectiveness of deep heuristic organized pattern or Gestalt. Any move is part example, presupposes intelligence. The body is important eth • no • meth • od • ol • o • gy: An approach to social science, search, and shows that AI is not, as Dreyfus thinks it is, of the unfolding Gestalt pattern. Past experiences knowledge of how the to the development or genesis of deriving from Garfinkel (1957), that focuses on everyday cultural 9 John von Neumann, 1958, dialectically?" arguments on this map. is supported by and the history of the current game work together is 10 Zenon Pylyshyn, 1974 is supported by a degenerating research program. In fact, given the body sits, bends, fatigues, Hubert Dreyfus intelligence (as Jean Piaget has practices and activities in the social world. Ethnomethodology as articulated by Hubert Dreyfus, 1972 background Analogue systems cannot represent general concepts. Analogue devices only capture particular success of AI, it may be Dreyfus's criticisms that are to build up an integrated awareness of "lines of disputed provides useful models for how plans are enacted and designed. is supported by shown), but not to its ultimate form. is The brain is an analogue device. Even sensory patterns. They cannot (by themselves) be used to recognize and process universal concepts. For degenerating. force, the loci of strength and weaknesses, as by By the time a human reaches is supported by disputed if neuron firings are all-or-none, the message example, an analogue retinal image of a chair is not by itself adequate to represent the universal concept well as specific positions" (p. 105). In this way, adulthood, the body is no longer by pulses that carry neural information are of a chair. The retinal image must be recognized as a typical chair pattern and must be associated with a chess masters zero in on unprotected pieces and is supported by essential. If the body were essential analogue. They involve complex graded and is promising areas for attack or defense. 119 Alonso Vera and Herbert Simon, 1993a verbal label by some sort of discrete digital mechanism. to intelligence, Dreyfus would have The mechanisms described by situated action are symbol systems. All of the mechanisms nonlinear factors. So the brain seems to be disputed Note: Pylyshyn allows that some analogue computation may be important in practice, and he thinks it is is Note: Also, see the "Can computers recognize Do humans use rules as to claim that an adult quadriplegic is suggested by the situated action program can be interpreted as physical symbol systems. The claims of an analogue device. by likely that practical AI systems will use hybrid analogue–digital mechanisms. He is arguing against the disputed Gestalts?" arguments on Map 5. is unintelligent. Because it is possible Note: Von Neumann subjects the relationship disputed situated action have been achieved in existing symbolic models. claim that all mental computation is analogue. by to model adult intelligence without 6 Warren McCulloch and between brain and computer to extensive by simulating its development, it is Walter Pitts, 1943 analysis, and his classic lectures on the subject possible to put intelligence in a is is 112 J. Y. Lettvin, Humberto Maturana, physical symbol The logical calculus of neural are still relevant today. John von Neumann 61 computer without a body. disputed disputed 97 Maurice Merleau-Ponty, 1962 The argumentation maps: 1 11 Zenon Pylyshyn, 1974 Computers play Warren McCulloch, and Walter Pitts, 1959 by by is supported by activity. Because of their all-or- Purely analogue machines lack the flexibility of digital machines. A purely analogue device 48 Alan Turing, 1950 The body is a synergic system. Frog retinas provide information none threshold, the activity of expert-level chess cannot make contingent if-then branches. That is, analogue devices cannot do "one thing under one set Being regulated by natural laws implies using heuristic All parts of the body are interrelated in without processing representations. neurons can be completely described is a "synergic system." Hands, feet, arms, 3 of circumstances and a completely different thing under a discretely different set of circumstances" (p. 68). being a rule-governed machine. The rules search. Chess 99 Doug Lenat, 1992, Instead of sending a detailed representation of 13 David Rumelhart, James McClelland, disputed 63 Hubert Dreyfus, 1996 systems do? in terms of logical operations. The For that reason, purely analogue machines inherently lack the flexibility of universal digital machines. This objection confuses rules of conduct with laws of and so forth are not juxtaposed in a the environment to its brain, the frog retina firing of a neuron is like the assertion and FARG, 1986 by programs play expert- Brute-force search is not how humans play chess. I never predicted as articulated by limitation can be overcome by adding a discrete threshold element, but that still does not make an analogue behavior. It is true that we cannot formulate a level chess using coordinate space, but are enveloped in only sends that information that is most 2 - arrange debate so that the cur- of a proposition, and relations Processing in the brain is distributed. is supported by failure for brute-force techniques in chess playing. All I argue is that brute- is supported by Hubert Dreyfus, 1992 device the best way to represent cognitive processes. complete set of rules of conduct for human advanced search a frontier of interrelations: sights can be relevant to the frog's needs. 120 Phillip E. Agre, 1993 122 William Clancey, 1993 between neural firings are like Processing in the brain is not mediated by force techniques such as heuristic search are not psychologically realistic. heard, sounds can be seen, and motor Madeleine reasons The symbol systems approach is a worldview, not performance. But human behavior is still governed techniques and without a body. Manipulation of symbols doesn't logical relations between some central control. Neural processing is Strom and Darden blur the distinction between AI as a form of psychology abilities can pass from one limb a theory. What is central to the symbol systems approach encompass all of human thought. Every by natural laws. And because these laws of behavior heuristics. and AI as any sort of technique that uses symbolic representation. Madeleine, a patient propositions. distributed—in other words, many regions 47 Anticipated by can, in principle, be given a mechanical description, to another. is a set of metaphors, such as "inside" and "outside." These interaction with the environment is an instance of contribute to the performance of any particular Alan Turing, 1950 discussed by Oliver Sacks, metaphors constitute a worldview rather than a theory, That's the 15 Nick Chater and Mike Oaksford, 1990 it is also possible to build a machine to fit this used a wheelchair, was The girl learning. We do not experience the world in terms rent stopping point of each task. Spatial distribution is inadequate. Impossible to write is description. Thus, humans are a kind of machine because they provide a framework for explaining the facts of preestablished categories; we create categories wrong kind of blind, and was unable to ran down distribution. Fodor and Pylyshyn claim that symbolic every rule. It is impossible disputed governed by rules. but don't provide explanations themselves. This worldview as we go along through a dialectical process of 12 to provide rules for every by read Braille. In effect, she the street. can't explain interactional entities that don't, strictly speaking, Hearing representations can be physically is lacked a body. Yet she still coadaptation between cognitive and perceptual The brain has a von Neumann architecture. The 16 David Rumelhart, James McClelland, and FARG, 1986 distributed in memory. But that kind of disputed eventuality that a computer 64 Hubert Dreyfus, 1972 belong to the external world or to the internal mind. systems. Thus, representations do not have a formal following features of the von Neumann architecture also Graceful degradation. Because processing in the brain is distributed, its might face. is managed to acquire is supported by distribution does not afford the right kind by Heuristic search is inconsistent with human phenomenology. structure independent of their application. characterize processing in the brain. performance diminishes in proportion to the degree of neuronal damage or noisy of damage tolerance. Representations laws of be • hav • ior: "Laws of nature as is supported by disputed Human experts play chess by "zeroing in" on relevant moves in fringe commonsense knowledge debate thread is easily seen • Processing is sequential. Seeing input. The performance of the brain "gracefully degrades" in problematic applied to man's body, such as 'If you pinch him by from books that were read • Symbol strings are stored and accessed at specific memory circumstances. In von Neumann machines, by contrast, a single disruption or glitch must be internally distributed in a rules of con • duct: "Precepts, such as 'Stop if he will squeak'" (Turing, 1950, p. 452). consciousness, rather than by iterating though a list of possibilities. to her. Her experience is is connectionist activation pattern rather you see red lights,' on which one can act, and of Note: Dreyfus uses similar considerations to dispute machine translation, shows that having a body is disputed addresses. is will generally have catastrophic consequences for the system as a whole. than merely spatially distributed in natural language understanding, and pattern recognition, which also require is supported by disputed • There is a central processing unit that controls processing. disputed memory. which one can be conscious" (Turing, 1950, p. 452). fringe consciousness and zeroing in. not essential to human by by Note: For more on the assumption of a central control, see is reasoning. by Note: For more on connectionist - identify original arguments by "Searle Assumes a Central Locus of Control," Map 4, Box Thinking disputed representations, see the "Can connectionist 75. by networks exhibit systematicity?" arguments What do is I need Implemented Model is disputed on Map 5. 49 Hubert Dreyfus, 1972 disputed Humans behave in an orderly manner without recourse to rules. rules for? 67 Feng-hsiung Hsu, Thomas Anantharaman, Murray Campbell, and Andreas Nowatzyk, 1990; Memory by by 114 James Gibson, 1977, 1979 121 Alonso Vera and Herbert Simon, 123 Alonso Vera and Herbert Simon, is supported by Human activity may be described by rules, but these rules are not necessarily IBM, 1998 1993c Affordances are features of the environment. Affordances are directly perceived 1993b over 380 protagonists world- followed in producing the activity. For example, if I wave my hand in the air, Deep Blue. A chess-playing computer that beat world champion Andrei Kasparov in 2 out 66 William James, 1890, as articulated by Hubert Dreyfus, 1972 Clancey's account of symbols is too is touch something accidentally, and move my hand back to that spot, I am of 6 games at a match in 1996, Deep Blue is an IBM parallel computer running 256 chess Humans zero in on information in fringe consciousness. higher-order properties of things in the environment. For example, water affords That the symbol systems approach limited. Clancey errs in his criticism of Central Processing Unit 14 Jerry Fodor and Zenon Pylyshyn, 1988 is swimming and floating; hills afford climbing; plains afford walking and running; foods is a worldview is not crucial. We do disputed performing a complex series of movements which can be described geometrically, processors, which allow it to consider 50–100 billion moves in the 3 minutes allotted to a player. The fringes of consciousness provide marginal awareness of the symbol system hypothesis and in his Symbol structures can be distributed. A classical DOG by disputed afford eating; and so on. Affordances aren't representations used as guides. They are present the symbol systems approach as a by but the only principle I follow is, "Do that again." Similarly, the planets are In addition to deep search of possible moves, Deep Blue consults a database of opening "background" information. For example, the experience of the claims for situated action. symbol processor can be physically distributed in memory, not solving differential equations as they revolve around the sun, even if their games and endgames played by chess masters over the last 100 years. Deep Blue's success is front of a house is fringed by an awareness of the back of the house. 100 Hubert Dreyfus, 1992 environmental factors that an organism directly "picks up" in order to maintain itself. worldview. But our criticisms of situated • The symbol systems hypothesis does not and can thereby exhibit graceful degradation. So distributed movements can be described by differential equations. based on engineering rather than on emulation. Even if doesn't think like a human, it is In chess, "cues from all over the board, while remaining on the action are specific. We address several limit itself to linguistic symbols: anything wide over 40 years Input Processing Control systems like connectionist networks don't have any principled D Madeleine has bodily and imaginative skills. Although Madeleine is blind and uses a climb-on-able principal theses propounded by the situated O relevant to business applications that require rapid management of large amounts of data. Deep fringes of consciousness, draw attention to certain sectors by wheelchair, she has a body with an inside and an outside and can be moved around in the world. is a symbol that is both patterned and advantage over physical symbol systems. G Blue is the descendant of earlier work by graduate students at Carnegie Mellon University. making them appear promising, dangerous, or simply worth looking action school. denotative. She can also communicate with others and imagine how they encounter the world. The claim that Unmapped Territory into" ( p. 104). Madeleine acquired common sense solely from books ignores these bodily and imaginative factors. • Clancey's ideas about the creation of new 50 Hubert Dreyfus, Either Or Unmapped Territory Note: In applying James's theory of the fringe to chess, Dreyfus Additional categories are too radical and are not Output 17 Keith Butler, 1993a 46 1972 is drawing on the work of Michael Polanyi (1962). Gibson AI supported by the research. Some - make the current frontier of 18 David Rumelhart, James McClelland, Classical machines implemented in connectionist The rule-following Programmed Additional relatively constant categories are 19 Jerry Fodor and networks can gracefully degrade. If a classical symbol assumption. Humans, like The machine must treat the The machine must take a arguments and FARG, 1986 behavior is either chess and other is necessary and their existence has been The brain processes information in Zenon Pylyshyn, 1988 D system were implemented in a connectionist network, then machines, behave intelligently by strictly rule-like new usage of language as a "blind stab" at interpretation game-playing demonstrated in animals. disputed parallel. Von Neumann machines process is Symbol processing can take the symbol structures could have the kind of internal following rules, which can, in or arbitrary. In case that falls under existing and then update its rule base, arguments by 102 Hubert Dreyfus, 1996 • The situated action paradigm is is information sequentially, one bit at a time. disputed place in parallel. It is possible to distribution that Chater and Oaksford require. In such a case, principle, be spelled out as explicit confronting a new rules, sit-on-able untestable, whereas there is much disputed is by implement a classical system in a D the symbol system could exhibit graceful degradation. if-then statements. in which case Collins's understanding of drinkable debate easily identifiable The brain receives and manipulates usage of language, Madeleine is science fiction. experimental evidence in support of the by disputed parallel architecture—for example, by Notes: in which case massive amounts of information at the O O machines face a symbol systems hypothesis. by same time, in parallel. executing multiple symbolic processes • Also, see the "Do connectionist dilemma. the machine is behaving in Madeleine was nothing like an at the same time. So parallel processing G networks follow rules?" rules covering all cases must an arbitrary, and hence immobile box. In crawling, kicking, is balancing, overcoming obstacles, is systems like connectionist networks, don't arguments on Map 5. be built in beforehand (see nonhuman, fashion. • Because heuristic searches are disputed 101 Harry Collins, 1996 finding optimal distances for listening, disputed is have any principled advantage over G 20 Herbert Simon, 1995 "The Infinite Regress of - provide summaries of eleven by and so forth, the baby Madeleine had by disputed 21 George Hinton, James McClelland, and David Rumelhart, classical symbol systems. Thought is serial despite being implemented in a parallel architecture. At the symbolic level, human sometimes described in terms of "heuristic rules," the "Does mental Rules," Box 51). If Madeleine can learn common sense, then so can a computer. enough of a body structure to allow by Legend Other physical symbol 1986 is supported by thinking is a serial process, despite the parallelism of its processing rely on heuristic If someone with as nonstandard a body her to be socialized into our human The brain accesses information by content rather neural implementation. For example, to multiply 5 by 12 search?" arguments on this map as Madeleine can acquire world. than by memory address. Humans rapidly access requires taking several serial steps, in which 5 is multiplied are relevant to this region. In Either Case 73 commonsense social knowledge, then The arguments on these maps are organized by links that carry a range of meanings: is AI programs are brittle. Because symbolic AI a computer with its own nonstandard major philosophical camps of memories by way of their contents. For example, memories disputed by 2, then by 10, and then the results are added together. about the president are accessed by information about the programs use rigid rules and data structures, they cannot body—a fixed metal box—could also The machine is not like a human. A native speaker, by systems arguments by adapt fluidly to changing environments and ambiguous 115 Alonso Vera and president (his or her name, face, etc.), not by way of an explicit contrast, is embedded in a context of human life, which acquire commonsense knowledge. If is Arguments that uphold or defend another claim. Examples include: circumstances. Symbol structures are brittle—they Herbert Simon, 1993a 118 Jerry Fodor and Zenon Pylyshyn, address. Connectionist networks and the brain have "content- addressable" memories of this type. If allows him or her to make sense of utterances in a non-rule- is supported by break apart under the pressure of a novel or ambiguous we can figure out what process Madeleine went through to become a disputed Affordances are just 1981 is supported by supporting evidence, further argumentation, thought experiments, like yet nonarbitrary way. by symbolic Affordances are trivial. Affordances 24 Nick Chater and Then situation. socialized human, then we might apply extensions or qualifications, and implemented models. the protagonists (or schools of 23 Jerry Fodor and Zenon Pylyshyn, 1988 Mike Oaksford, 1990 is Note: Versions of this claim are widely discussed in that process to a computer as well. representations. are just another name for whatever it is in The 100-step constraint is directed at the The 100-step constraint disputed the literature and on these maps. For example, Note: For more on Collins's views Affordances are nothing but the environment that makes an organism implementation level. All the 100-step constraint is relevant to the cognitive by brittleness is discussed by Hofstadter (see "The Front- about socialization, see the "Can internal representations respond as it does. But such a notion can't 52005 22 Jerome Feldman, 1985 demonstrates is the obvious fact that symbolic thought processes Table The 100-step constraint. Algorithms level. It is not true that the End Assumption is Dubious," Map 1, Box 74), Brooks computers reason scientifically?" whose symbolic character is provide a substantial explanation of A charge made against another claim. Examples include: are implemented differently in the brain than they are on a 100-step constraint is an irrelevant (see sidebar, "Postulates of Subsumption Architecture," arguments on Map 1. concealed from conscious- perception. For example, to say we is that model cognitive processes must meet a digital computer. The 100-step constraint has to do with 53 Jerry Fodor and Zenon Pylyshyn, ness. Acquiring affordances recognize a shoe by perceiving its property logical negations, counterexamples, attacks on an argument's thought). 100-step constraint imposed by the implementation detail. An on this map), Dreyfus (see sidebar, "Postulates of disputed timescale of the brain, which performs is implementation details rather than with real cognitive processes. algorithm that classical systems 52 David Rumelhart, 1988 68 Jerry Fodor, 1975 Dreideggereanism," on this map), and the connectionists is a matter of encoding of being a shoe doesn't explain anything. by emphasis, potential dangers an argument might raise, thought is The language of thought. The Affordances add nothing new to our complex tasks in about 100 time-steps. disputed disputed execute in millions of time-steps James McClelland, and Classicists are not committed 51 Ludwig Wittgenstein, 1953 language of thought (also known as (see Map 5). It is also raised in the context of fuzzy sensory stimuli as symbolic experiments, and implemented models. knowledge of the mechanisms behind How do I get a Classical sequential algorithms currently run by The 100-step constraint is may not be executable in 100 FARG, 1986 to explicit rules. The possibility The infinite regress of logic. representations. by mentalese) is a formal language that perception. in millions of time-steps, so it seems a mere implementation time-steps. The 100-step constraint Explicit rules are of implicit rules does not argue against rules. Assuming that all the classical symbolic framework, nonarbitrary human behavior mental processes operate on. Like spoken unlikely that they can meet this 100-step 99 detail. poses a nontrivial challenge to unnecessary. Connectionist language, the language of thought has a constraint. classical AI, because it "severely networks exhibit lawful behavior because there is a wide body of work is governed by rules, then rules 74 James Lighthill, 1973 is interpreted as 100 within the classicist camp that shows must be specified in order to combinatorial syntax and semantics. Just 75 John McCarthy, 1990a A distinctive reconfiguration of an earlier claim. [limits] the class of cognitively without following explicit rules. as complex sentences are generated from Unmapped Territory The Lighthill Report. There are 3 kinds of work in AI. is plausible algorithms" (p. 95). how implicit rules can be modeled. apply the original rules, and Lighthill's categories 117 Alonso Vera and Regularities emerge from the combinations of words, complex mental are irrelevant to AI disputed In fact, most classicists agree that at further rules must be specified is supported by Additional Herbert Simon, 1993c interactions of low-level representations are generated from research. The Lighthill by is processing units, rather than from least some rules must be implicit. for those rules, ad infinitum. language of Affordances need to be However, the possibility of explicit combinations of simpler representations. thought A: Advanced automation. C: Computer-based study of the Report implies that AI redefined. Explaining the disputed the application of high-level rules. Although no formal theory of the B: Bridge activity. The use research is only useful by Although it may be possible to is rule-based systems does argue against arguments The use of computers to central nervous system. The use mechanisms of perception disputed connectionism, because connectionist requires language of thought has yet been fully replace human beings in of computers to study phenomena of computers in the study of the insofar as it contributes to and cognition requires Focus Box: The lowest-numbered box in each issue area is an introductory focus box. characterize a network's behavior successful, the search for one is the goal that fall between categories A and C—in industrial applications according to high-level rules, none by networks cannot encode such rules. various military, industrial, brain, as a way of testing hypotheses understanding how The focus box introduces and summarizes the core dispute of each issue area, sometimes Note: For an example of an argument Rule of computational psychology. and scientific tasks. particular, the building of robots as a way of about the cerebellum, visual cortex, (category A) and to information is encoded in the printed copy? are involved in its underlying Note: For more arguments about this studying general intelligence. neuroscience (category C). 116 James Greeno and as an assumption and sometimes as a general claim with no particular author. mechanisms. (cited by Fodor and Pylyshyn) that requires and so forth. brain. Because affordances aspect of AI, see "The Representationalist is But AI has goals of its Joyce Moore, 1993 says that explicit rules can't be Affordances cannot be are not representations, they Assumption," Box 103, and the "Can Work in categories A and C is legitimate. But work in category B is plagued by a variety of problems (including combinatorial disputed own. AI researchers study provide no help in encoded by connectionist networks, Rule connectionist networks exhibit is explosion, past failures, limited worlds, and poor performance) and therefore seems unlikely to succeed. by structures of information redefined as symbols. Arguments with No Authors: Arguments that are not attributable to a particular see "The Past-Tense Model Does Not disputed Affordances aren't like understanding that coding source (e.g., general philosophical positions, broad concepts, common tests in artificial systematicity?" arguments on Map 5. Notes: and problem solving Can physical symbol systems think dialectically? • Also, see "Past Disappointments," Box 90. Argue Against Rule-Based by symbols. They don't mediate is process. On the other hand, independently of how intelligence) are listed with no accompanying author. • The Lighthill Report was commissioned by the Science Research Council in Britain to help the council make funding Explanation," Map 5, Box 24. between the organism and its disputed those models that have been is these structures are helpful in explaining the 69 John McCarthy, 1979 decisions for work in AI. Lighthill was chosen as a member of the scientific community who could make an unbiased realized in humans and environment. Affordances by disputed coding process have been by Thermostats can have assessment of the field. His report is said to have had devastating effects on AI funding in Britain during the '70s. animals. are directly "picked up" by symbolic. So it is best to Citations: Complete bibliographic citations can be found in the booklet that accompanies beliefs. Beliefs can Note: An early variant of the organism. So it is this map. Postulates of Ordinary Language I believe reasonably be ascribed to an inappropriate to redefine redefine affordances in terms 26 Joseph Rychlak, 1991 27 Arturo Rosenblueth, Norbert 28 Joseph Rychlak, 1991 54 Graham Button, Jeff Coulter, John R. E. Lee, and this argument was made of symbols. Negative feedback Wes Sharrock, 1995 1. The purpose of philosophy is the resolution of conceptual it's too hot entity when its actions are: What's the least by Professor D. Mitchie, affordances as symbols. Symbol systems can't Wiener, and Julian Bigelow, 1943 is supported by in here. consistent, the result of 76 Hubert Dreyfus, 1972, 1979, 1992 complex Methodology: A further discussion of argumentation analysis methodology can be Some machines behave can't explain AI rules cannot explain ordinary language. The confusion through the analysis of ordinary language. The critique of artificial reason. Artificial intelligence is the in an appendix to the exhibit agency. Symbol Postulates of the Dialectical Paradigm observation, and in accordance intellectual found in the booklet that accompanies this map. processors can never have teleologically. Machines with teleology. Goal- project of accounting for ordinary human language in terms culmination of a flawed tradition in philosophy that tries to explain Lighthill Report. directedness and is of explicit AI rules runs into the following problems. 2. Ordinary language is our primary medium of thought. with goals. Because the behavior that is human reason in terms of explicit rules, symbols, and calculating • Our linguistic practices are too open-ended to be captured agency. Agency requires servomechanisms are purposive. In is 1. Dialectical reasoning involves oppositional and predicational behavior of thermostats meet is you think disputed fact, any machine that can respond to feedback activity cannot disputed procedures. The assumptions of AI are implicitly critiqued by a range teleology, dialectical reasoning, disputed explain agency. An thinking, and is only possible for teleological beings who can by by a set of explicit rules. 3. Problems of philosophy arise through the misuse or incomplete those criteria, it is reasonable disputed humans can do Anticipated by Where this phrase appears in a box, it identifies a potential by of 20th-century thinkers (Maurice Merleau-Ponty, Martin Heidegger, • Human language is essentially embedded in a context of and free will. Computers lack negative feedback for guidance by "see" from another's point of view. to say they have beliefs. and computers these traits because they don't behaves teleologically. What explanation of agency understanding of ordinary language. by Michael Polanyi, Thomas Kuhn, etc.), who show how the nature of 77 John McCarthy, is attack on a previous argument that is raised by the author must address the use, and it continually adapts to this context of use. Note: These are the criteria for can't? (p. 149) 78 Hubert Dreyfus, 1996 human activity and being differ from that of calculating machines like disputed so that it can be disputed. • Ordinary language gets its meaning from practical understand meanings, the separates such machines from humans 2. Oppositional thinking proceeds by assertion, denial, and new belief ascription that McCarthy 1996 25 Joseph Rychlak, 1991 relations between them, and their is that humans have an added ability formation of goals and 4. The proper method for analyzing language is through the computers. by Get an AI system to understand, "Mary saw beliefs and the ability to synthesis. An assertion is made and is opposed by a denial involvement in "language games," not from its parsable explication of commonsense everyday use of language. mentions in connection with the What is the a dog in the window. She wanted it." When Symbol systems cannot think relation to the world. to make higher-order predictions about John McCarthy Note: This is a general statement of Dreyfus's critique. Specific easiest thing As articulated by Where this phrase appears in a box, it identifies a reform- dialectically. Computers can't reason is supported by change those goals and of that assertion. Out of this opposition, a new synthesis arises grammatical structures. thermostat example. He faced with the sentences, "Mary saw a dog in the Note: For similar arguments, see the behavior of other objects. instances of his critique are spread throughout this map. computers can't ulation of another author's argument. The reformulation You can order artist/researcher and transforms the original assertion by changing its context discusses further conditions dialectically because they can't synthesize the "Can computers have free beliefs. Current and/or scope. 5. Language can be studied in terms of "language games," in elsewhere. do? Dreyfus makes 80 John McCarthy, 1977 window. She wanted it," it is difficult for an AI system PETS opposing meanings into a new meaning. All will?" arguments on Map 1. machines only seek the which rules of usage are analyzed as if they were the rules some vague Formalized non- to know whether "it" refers to the dog or the window. is different enough from the original author's wording to they can do is shuffle symbols that are given goals they are of a game. monotonic logic. John McCarthy says this problem is "within the warrant the use of the tag. This phrase is also used when programmed to seek. 3. "Predication involves the cognitive act of affirming, denying, is arguments about capacity of some current parsers" (1996, p. 190). But meaning by a programmer. 55 Immanuel Kant, 1781 Formalized non-monotonic Note: Rychlack never claims that machines or qualifying broader patterns of meaning in relation to 6. Language can only be understood semantically from within Implemented Model disputed logic-based AI, but interpreting the sentence requires bodily know-how the original argument is impossible to locate other than in signed copies of all seven maps 30 Joseph Rychlak, 1991 narrower or targeted patterns of meaning" (Rychlak, 1991, p. All concepts are rules. 72 George Lakoff, 1987 he never issues a logic is a development of formal can't think. He just claims that they can't think Agency is due to predication and choice. Concepts are rules for combining ordinary language and language games. This means there is by precise, testable logic that allows for the and empathetic imagination. Figuring out what its articulation by a later author (e.g., word of mouth), or 7). no formal metalanguage relevant to philosophy. 70 John Laird, Allen Newell, and Paul Rosenbloom, 1987 Some conceptual organizations pronouns refer to in such sentences is the next problem dialectically or predicationally. He allows that Automatic decision-making processes, even if they is supported by the elements of perception into SOAR. SOAR is a general intelligence system that learns can't be translated into a universal conceptual framework. Different conceptual is challenge. introduction of new axioms to to denote a general philosophical position that is given a they have demonstrative reasoning. objective representations. This disputed invalidate old theorems. In this logic-based AI should try to deal with. is is possess delaying mechanisms, still cannot choose 4. Teleological beings can engage in various goal-directed 7. The rules of ordinary language are adapted to the various is supported by by "chunking," that is, by collapsing the work of satisfying systems have different conceptual organizations, and those differences are cognitively Note: The example, "Mary saw a dog in the window. special articulation by a particular author. disputed their own goals or reflect on the meaning of their 31 John Locke, 1690 "synthesis" of experience a subgoal into a single condition-action rule, or "production." significant. If the different organizations are translated into a universal conceptual by way non-monotonic logics activities that place them in a meaningful relation to the world. from www.macrovu.com for disputed presupposes rules of practical purposes of human conduct, and as such they must John McCarthy account for the human ability She wanted it," comes from Doug Lenat (quoted in by alternatives. The standards they use to choose It searches its explicit representations heuristically by means- framework, significant organizational differences are eliminated. Humans can shift New synthesis by between alternatives are not a result of their own Delaying action provides us with consciousness as well as a rule- be open to examination and modification. to revise assumptions in light Dreyfus, 1992 pp. xix–xx). Unmapped Territory This icon indicates areas of argument that lie on or near the free will. Free will stems from the 5. Computers reason mediationally. In other words, they work ends analysis using goals and subgoals. between frameworks without eliminating these differences. status as agents. governed world. Note: Lakoff cites Mixtec (a native Mexican language), as well as the work of Benjamin of new observations. boundaries of the central issue areas mapped on these maps. power to restrain from action. with symbols in abstraction from their relation to meanings 8. Rules for the use of language are like rules for playing games; Note: This definition is adapted Additional Delaying the impulse to act provides and the world. Immanuel Kant they are not like the parsing rules posited by AI researchers. Whorf (1956) in making this argument. arguments It marks regions of potential interest for future mapmakers from McDermott and Doyle $500.00 plus shipping and han- Ordinary language rules provide guides and signposts for the 71 Articulated by room for other actions to be 6. Computers reason demonstratively. They can draw conclusions George Lakoff, 1987 Yuu wa hiyaa 79 John McCarthy, 1996 (1980). and explorers. New synthesis Denial considered. This time in which use of language rather than a description of how language is is supported by Logic-based AI is making steady 29 Marvin Minsky, 1986, as articulated by Joseph Rychlak, 1991 based on valid patterns of inference between symbols, but 56 Noam Chomsky, 1965, 1980 Unmapped Territory generated. The universal cu-mesa. Unmapped Territory Delayed mediational processes in machines can generate agency. alternatives are weighed provides the progress. Dreyfus ignores slow but appearance of free will. The choices without working with the meanings of those symbols. Grammars are rule-based conceptual framework is (Stone the definite progress in logic-based AI. For By telling ourselves that we have made choices, we make ourselves agents. systems. The grammar of a Additional Proponents include Ludwig Wittgenstein; Graham Button, Jeff assumption. There is a neutral and disputed be-located dling. What we know is that free will occurs when decision-making processes are that are made during the delay need example, formalized non-monotonic Additional One of 7 in this Issue Mapping™ series—Get the rest! is supported by 7. Computers can not reason dialectically because they are is supported by language is a system of rules for Chomsky Coulter, John R. E. Lee, and Wes Sharrock; Charles Karelis completely general conceptual by belly-table)] logics have been used successfully to is supported by non-monotonic delayed to allow more alternatives to be tested. Systems capable of that kind not arise from some metaphysical The Issue Mapping™ series is published by MacroVU Press, a Assertion Denial freedom of the subject; free will is nonteleological and nonpredicational. They do not understand the production of sentences. These arguments (Map 2); Hans Obermeier (Map 4); and Gilbert Ryle (Maps 2 structure in which all knowledge can address the problem of understanding arguments © 1998 R. E. Horn. of delayed decision making have as much free will as humans do, and thus, meanings or the relations between them. rules are part of the unconscious and 6). Other notable proponents include John L. Austin, John be represented. For a machine to The remaining 6 maps in this Issue Mapping™ series can be ordered with MasterCard, VISA, check, division of MacroVU, Inc. MacroVU is a registered trademark any computer system with those processes can be said to have agency. just the consideration of alternatives. ambiguity. Dreyfus gives no compelling All rights reserved. "deep structure" of language. Wisdom, and Stanley Cavell. understand natural language, it must or money order. Order from MacroVU, Inc. by phone (206–780–9612), by fax (206–842–0296), of MacroVU, Inc. Issue Map and Issue Mapping are trademarks translate sentences into this universal reason to suppose that such Version 1.0 Mixtec speaker developments will end in failure. or through the mail (Box 366, 321 High School Rd. NE, Bainbridge Island, WA 98110). of Robert E. Horn. conceptual framework.