Reduction and Elimination in Philosophy and the Sciences

More documents

Recommendations

Info

The Elimination of Meaning in Computational Theories of Mind Paul Schweizer, Edinburgh, Scotland, UK 1. The Computational Paradigm According to the traditional conception of the mind, semantical content is perhaps the most important feature distinguishing mental from non-mental systems. For example, in the scholastic tradition revived by Brentano (1874), the essential feature of mental states is their ‘aboutness’ or intrinsic representational aspect. And this traditional conception has been incorporated into the foundations of contemporary scientific approaches to the mind, insofar as the notion of ‘mental representation’ is adopted as a primary theoretical device. For example, in classical (e.g. Fodorian) cognitive science, Brentano’s legacy is preserved in the view that the properly cognitive level is distinguished precisely by appeal to representational content. There are many different levels of description and explanation in the natural world, from quarks all the way to quasars, and according to Fodor (1975), it is only when the states of a system are treated as representations that we are dealing with the genuinely cognitive level. The classical paradigm in cognitive science derives from Turing’s basic model of computation as rule governed transformations on a set of syntactical elements, and it has taken perhaps its most literal form of expression in terms of Fodor’s Language of Thought hypothesis (LOT), wherein mental processes are explicitly viewed as formal operations on a linguistically structured system of internal symbols. But a fundamental tension is already built into the classical picture: a central purpose of the symbolic structures is to carry content, and yet, to the extent that they are formal elements of computation, their alleged content is completely gratuitous. Computation is essentially a series of manipulations performed on uninterpreted syntax, and formal structure alone is sufficient for all effective procedures. The specification and operation of such procedures makes no reference whatever to the intended meaning of the symbols involved. Indeed, it is precisely this limitation to syntactic form that has enabled computation to emerge as a mathematically rigorous discipline. If syntax alone is not sufficient, and additional understanding or interpretation is required, then the procedure in question is, by definition, not an effective one. But then the purported content of mental ‘representations’ is rendered superfluous to the computations that comprise the ‘cognitive’ processes of cognitive science. The intended interpretation of internal syntax makes absolutely no difference to the formal mechanics of mind. For a number of years now there has been a high profile struggle between opposing camps within the computational approach to the mind. In contrast to the classical paradigm derived from Turing, connectionist systems are based on networks of large numbers of simple but highly interconnected units that are brain-like in inspiration. But according to Fodor, the brain-like architecture of connectionist networks tells us nothing about their suitability as models of cognitive processing, since it still leaves open the question of whether the mind is such a network at the representational level. So a number of connectionists have taken up the challenge and seek out ways of projecting representational content onto artificial neural networks. One comparatively recent such attempt (Churchland, P.M.1998, Laakso, A. and G. Cottrell 2000, O’Brien, G. and J. Opie 2001) uses cluster analysis to locate ‘vehicles’ of representational content within artificial neural networks, where such clusters serve as surrogates for the classical notion of internal syntax. However, I would contend that such attempts suffer from exactly the same built-in tension that afflicts the LOT model; namely, the purported content for which the clusters serve as vehicles does no work in the processing path leading from inputs to outputs. Just as in the classical case, the postulation of content within the connectionist framework is gratuitous, because it plays no role in the cognitive manipulation of inputs to yield the salient outputs. Indeed, if content weren’t gratuitous, then computational versions of cognitive processing would be lamentably deficient in terms of their specification of the inputs. These are characterized solely in formal or syntactical terms, and content is entirely absent from the external stimuli recognized by the operations that can be defined within the model. If representational content were at all relevant, then cognitive systems would have to process content itself. But according to computational methods, content is not specified with the input, nor does it play any efficacious role in internal processing. So, from a perspective that takes computation as the theoretical foundation for cognition, it seems quite retrograde to posit content on top of the factors that do the actual work. Surely this is an exemplary occasion for invoking Ockham’s razor. 2. Searle’s Objection Of course, John Searle’s (1980) celebrated Chinese Room Argument (henceforward CRA) runs the dialectic in exactly the reverse direction: rather than taking the formal, syntactic nature of computation as a reason for eschewing content in a properly naturalistic approach to the mind, Searle instead takes it as a reason for rejecting computation as the appropriate theory of the mental. So, from the perspective of the present discussion, it is instructive to explicitly cast Searle’s argument in terms of the separability of syntactical structure from its intended meaning. In what follows I will abstract away from the somewhat picturesque details of Searle’s original version and express the logical core of the CRA via two premises and a conclusion: (1) semantical content is an essential feature of the mind, (2) syntactical manipulations cannot capture this content, therefore (3) the mind cannot be reduced to a system of syntactical manipulations. Preimse (1) is an expression of the traditional conception of the mind, and is accepted by both Searle and by his opponents in orthodox cognitive science and AI. Classical cognitive science and AI view the mind according to the model of rule governed symbol manipulation, and premise (1) is embraced insofar as the manipulated symbols are supposed to possess representational content. Searle’s dispute with cognitive science and AI centers on his rejection of the idea that internal computation can shed any real light on mental content, which leads to his conclusion (3), 313
and a concomitant dismissal of the research paradigm central to cognitive science and AI. In response, a standard line for defenders of this paradigm is to try and defuse the CRA by arguing against premise (2), and claiming that the manipulated symbols really do possess some canonical meaning or privileged interpretation. However, I would urge that this is a serious strategic error for those who wish to defend the computational approach. As stated above, a distinguishing mathematical virtue of computational systems is precisely the fact that the formal calculus can be executed without any appeal to meaning. Not only is an interpretation intrinsically unnecessary to the operation of computational procedures, but furthermore, there is no unique interpretation determined by the computational syntax, and in general there are arbitrarily many distinct models for any given formal system. Computational formalisms are syntactically closed systems, and in this regard it is fitting to view them in narrow or solipsistic terms. They are, by their very nature, independent of the ‘external world’ of their intended meaning and, as mentioned above, they are incapable of capturing a unique interpretation, since they cannot distinguish between any number of alternative models. This can be encapsulated in the observation that the relation between syntax and semantics is fundamentally one-to-many; any given formal system will have arbitrarily many different interpretations. And this intrinsically one-tomany character obviates the possibility of deriving or even attributing a unique semantical content merely on the basis of computational structure. The inherent limitations of syntactical methods would seem to cast a rather deflationary light on the project of explicating mental content within a computational framework. Indeed, they would seem to render hopeless such goals as providing a computational account of natural language semantics or propositional attitude states. Non-standard models exit even for such rigorously defined domains as first-order arithmetic and fully axiomatized geometry. And if the precise, artificial system of first-order arithmetic cannot even impose isomorphism on its various models, how then could a program, designed to process a specific natural language, say Chinese, supply a basis for the claim that the units of Chinese syntax posses a unique meaning? So I think that the advocates of computation make the wrong move by accepting Searle’s bait and taking on board the attendant ‘symbol grounding problem’ endemic to computational theories of mind. Instead I would accept Searle’s negative premise (2) and agree that computation is too weak to underwrite any interesting version of (1). Hence I would concur with Searle’s reasoning to the extent of accepting the salient conditional claim that if (1) is true then (3) is true as well. So the real crux of the issue lies in the truth-value of (1), without which the consequent of the if-then statement cannot be detached as a free-standing conclusion. Only by accepting the traditional, a priori notion of mentality assumed in premise (1), does (3) follow from the truth of (2). And it’s here that I diverge from the views of both Searle and orthodox cognitive science. 3. Representation as Heuristics There have been a number of prominent positions advanced in negative reaction to ‘classical’ cognitive science that take anti-representationalism as one their hallmarks, including dynamical systems theory (e.g Van Gelder 314 The Elimination of Meaning in Computational Theories of Mind — Paul Schweizer 1996), behaviour based robotics (e.g. Brooks 1991), approaches utilizing sensory-motor affordances (e.g. Noe 2004), who campaign on the platform of ‘intelligence without representation’. In order to locate my position on the philosophical landscape, it is salient to note that it is not anti-representational in this sense. On my view, there could well be internal structures that play many of the roles that people would ordinarily expect of representations, and this is especially true at the level of perception, sensorymotor control and navigation – things like spatial encodings, somatic emulators, internal mirrorings of salient aspects of the external environment. So, unlike the antirepresentationalists, I do not deny that there may be internal structures and stand-ins that various people would be tempted to call ‘representations’. But I would argue that this label should be construed in a weak, operational sense, and should not be conflated with the more robust traditional conception. To the extent that internal structures can encode, mirror or model external objects and states of affairs, they do so via their own causal and/or syntactic properties. And again, to the extent that they influence behaviour or the internal processing of inputs to yield outputs, they do this solely in virtue of their causal and/or syntactic properties. There is nothing about these internal structures that could support Searle’s or Brentano’s notion of original intentionality, and there is no independent or objective fact of the matter regarding their ‘real’ content or meaning. The crucial point to notice is that these internal ‘representations’ do all their scientifically tangible cognitive work solely in virtue of their physical/formal/mathematical structure. There is nothing about them, qua efficacious elements of internal processing, that is ‘about’ anything else. Content is not an explicit component of the input, nor is it acted upon or transformed via cognitive computations. All that is explicitly present and causally relevant are computational structure plus supporting physical mechanisms, which is exactly what one would expect from a naturalistic account. In order for cognitive structures to do their job, there is no need to posit some additional ‘content’, ‘semantical value’, or ‘external referent’. Such representation talk may serve a useful heuristic role, but it remains a conventional, observer-relative ascription, and accordingly there’s no independent fact of the matter, and so there isn’t a sense in which it’s possible to go wrong or be mistaken about what an internal configuration is ‘really’ about. Instead, representational content is projected onto an internal structure when this plays an opportune role in characterizing the overall processing activities which govern the system’s interactions with its environment, and hence in predicting its salient input/output patterns. But it is simply a matter of convenience, convention and choice. From the point of view of the system, these internal structures are manipulated directly, and the notion that they are ‘directed towards’ something else plays no role in the pathways leading from cognitive inputs to intelligent outputs. Hence the symbol grounding problem is a red herring – it isn’t necessary to quest after some elusive and mysterious layer of content, for which these internal structures serve as the syntactic ‘vehicle’. Syntactical and physical processes are all we have, and their efficacy is not affected by the presence or absence of meaning. I would argue that the computational paradigm is thematically inconsistent with the search for content or its supposed vehicles. Instead, computational models of cognition should be concerned only with the processing structures that yield the right kinds of input/output profiles, and with how such structures can be implemented in the brain. These are the factors that do the work and are
Page 1 and 2:
31 Alexander Hieke Hannes Leitgeb H
Page 3 and 4:
Reduktion und Elimination in Philos
Page 5 and 6:
Distributors Die Österreichische L
Page 7 and 8:
6 Inhalt / Contents Wittgenstein me
Page 9 and 10:
8 Inhalt / Contents The Evolution o
Page 12 and 13:
Formal Mechanisms for Reduction in
Page 14 and 15:
4. Property language and theoretica
Page 16 and 17:
imperatives, or commands in the for
Page 18 and 19:
Referential Practice and the Lure o
Page 20 and 21:
The Augustinian account confuses mo
Page 22 and 23:
hurried to fix it on the page. The
Page 24 and 25:
The Essence (?) of Color, According
Page 26 and 27:
Literature Austin, James 1980 “Wi
Page 28 and 29:
Wittgenstein’s Externalism - Gett
Page 30 and 31:
Informal Reduction E.P. Brandon, Ca
Page 32 and 33:
An Anti-Reductionist Argument Based
Page 34 and 35:
An Anti-Reductionist Argument Based
Page 36 and 37:
Did I Do It? -Yeah, You Did! Wittge
Page 38 and 39:
Literature Did I Do It? -Yeah, You
Page 40 and 41:
tween the two approaches. The task
Page 42 and 43:
On Two Recent Defenses of The Simpl
Page 44 and 45:
On Two Recent Defenses of The Simpl
Page 46 and 47:
Queen Victoria’s Dying Thoughts T
Page 48 and 49:
Diagonalization. The Liar Paradox,
Page 50 and 51:
Proof: Given a function ƒ from con
Page 52 and 53:
that I am sure that it is true that
Page 54 and 55:
experiencing something as a red pen
Page 56 and 57:
There can be Causal without Ontolog
Page 58 and 59:
There can be Causal without Ontolog
Page 60 and 61:
objects. But they also leave unansw
Page 62 and 63:
The Knower Paradox and the Quantifi
Page 64 and 65:
The Knower Paradox and the Quantifi
Page 66 and 67:
lent. It also implies that an inven
Page 68 and 69:
The Scapegoat Theory of Causality M
Page 70 and 71:
We may have evolved a specialized c
Page 72 and 73:
easoning have their source in the l
Page 74 and 75:
Wittgenstein on Frazer and Explanat
Page 76 and 77:
genstein 1993, p. 143). In order to
Page 78 and 79:
an adequate syntactic analysis of o
Page 80 and 81:
Wittgenstein meets ÖGS: Wovon man
Page 82 and 83:
Wittgenstein meets ÖGS: Wovon man
Page 84 and 85:
3. Der Elementarsatz wird als Relat
Page 86 and 87:
Literatur Glock, Hans Johann 2006
Page 88 and 89:
Explaining the Brain: Ruthless Redu
Page 90 and 91:
Occam’s Razor in the Theory of Th
Page 92 and 93:
Literature Campbell, Donald T. 1987
Page 94 and 95:
Die Nichtreduzierbarkeit der klassi
Page 96 and 97:
Literatur Die Nichtreduzierbarkeit
Page 98 and 99:
Let T I be the minimal theory of tr
Page 100 and 101:
Does Bradley’s Regress Support No
Page 102 and 103:
(3) Mutatis mutandis, realism has t
Page 104 and 105:
Zeitliche Ontologie und zeitliche R
Page 106 and 107:
Betrachtung ontologischer Fragen, M
Page 108 and 109:
The decisive consequences of this a
Page 110 and 111:
Benacerraf and Bad Company (An Atta
Page 112 and 113:
offer a solution to the bad company
Page 114 and 115:
Literature Benacerraf, Paul 1965
Page 116 and 117:
"PA" by using "PA* ", at least at t
Page 118 and 119:
Hard Naturalism and its Puzzles Ren
Page 120 and 121:
The Mind-Body-Problem and Score-Kee
Page 122 and 123:
mental properties, for instance, or
Page 124 and 125:
kommt es zu einem gravierenden Miss
Page 126 and 127:
Reduction Revisited: The Ontologica
Page 128 and 129:
4. Conclusion Reduction Revisited:
Page 130 and 131:
All in all, we have seen that reduc
Page 132 and 133:
Physicalism Without the A Priori Pa
Page 134 and 135:
Literature Cartwright, Nancy 1999:
Page 136 and 137:
Wittgensteins Projektionsmethode al
Page 138 and 139:
Schlussbemerkungen Wittgensteins Pr
Page 140 and 141:
intentional states requires functio
Page 142 and 143:
Relating Theories. Models and Struc
Page 144 and 145:
Literature Relating Theories. Model
Page 146 and 147:
e invoked in response to the questi
Page 148 and 149:
Do Brains Think? Christopher Humphr
Page 150 and 151:
4. Conclusion So do brains think or
Page 152 and 153:
How Metaphors Alter the World-Pictu
Page 154 and 155:
The Modal Supervenience of the Conc
Page 156 and 157:
Literature van der Auwera, Johan an
Page 158 and 159:
3. The Determination of Form by Syn
Page 160 and 161:
Zwischen Humes Gesetz und „Sollen
Page 162 and 163:
Zwischen Humes Gesetz und „Sollen
Page 164 and 165:
Assessing Humean Supervenience Amir
Page 166 and 167:
determination thesis; it fails as a
Page 168 and 169:
Aus der obigen Definition folgt, da
Page 170 and 171:
Ding-Ontology of Aristotle vs. Sach
Page 172 and 173:
Ding-Ontology of Aristotle vs. Sach
Page 174 and 175:
How do Moral Principles Figure in M
Page 176 and 177:
“Downward Causation”: Emergent,
Page 178 and 179:
“Downward Causation”: Emergent,
Page 180 and 181:
A with distinguished subset Z of te
Page 182 and 183:
A Metaphysically Moderate Version o
Page 184 and 185:
Literature Balashov, Yuri 2002 ''Wh
Page 186 and 187:
“In der Frage liegt ein Fehler”
Page 188 and 189:
Problems with Psychophysical Identi
Page 190 and 191:
identity theory, or as a predecesso
Page 192 and 193:
of the world. It only matters that
Page 194 and 195:
Four Anti-reductionist Dogmas in th
Page 196 and 197:
Four Anti-reductionist Dogmas in th
Page 198 and 199:
notion of time and requires that so
Page 200 and 201:
Some Remarks on Wittgenstein and Ty
Page 202 and 203:
Some Remarks on Wittgenstein and Ty
Page 204 and 205:
concrete particulars can be subject
Page 206 and 207:
phenomenal concept of experiencing
Page 208 and 209:
Metaphorische Bedeutung als virtus
Page 210 and 211:
speaker is then attaching a special
Page 212 and 213:
„Vom Weißdorn und vom Propheten
Page 214 and 215:
„Die Einheit hören“ - Einige
Page 216 and 217:
„Die Einheit hören“ - Einige
Page 218 and 219:
S4) the necessity of arithmetic is
Page 220 and 221:
Getting out from Inside: Why the Cl
Page 222 and 223:
Dispensing with Particulars: Unders
Page 224 and 225:
Literature Dispensing with Particul
Page 226 and 227:
Reichenbach’s Concept of Logical
Page 228 and 229:
Defining Ontological Naturalism Mar
Page 230 and 231:
Literature Jackson, Frank 1986. “
Page 232 and 233:
properties are located. This proces
Page 234 and 235:
‘our conceptual scheme’ “is c
Page 236 and 237:
Properties and Reduction between Me
Page 238 and 239:
Does this mean that metaphysics sho
Page 240 and 241:
do not pick out pain in a system by
Page 242 and 243:
The Writing of Nietzsche and Wittge
Page 244 and 245:
his teachings. He coincides with hi
Page 246 and 247:
sense of a proposition without firs
Page 248 and 249:
Naturalistic Ethics: A Logical Posi
Page 250 and 251:
Literature Kant, Immanuel 2006 Crit
Page 252 and 253:
What we now require is a bridge the
Page 254 and 255:
Species, Variability, and Integrati
Page 256 and 257:
could mention the many sorts of iso
Page 258 and 259:
to work this way. Classic examples
Page 260 and 261:
The Key Problems of KC Matteo Pleba
Page 262 and 263:
have embraced the myth of prose: it
Page 264 and 265: 5. Prior’s Problem As Patrick Bla
Page 266 and 267: Reductionism in Axiology: the Case
Page 268 and 269: developing of cancer tissue in huma
Page 270 and 271: (2) One-way Nagelian bridge princip
Page 272 and 273: Rethinking the Modal Argument again
Page 274 and 275: Rethinking the Modal Argument again
Page 276 and 277: that we are to find the features we
Page 278 and 279: Atypical Rational Agency Paul Raymo
Page 280 and 281: Literature Anscombe, G. E. M. 1957
Page 282 and 283: situation bieten sich keine Kriteri
Page 284 and 285: Two Reductions of ‘rule’ Dana R
Page 286 and 287: one that could take place, but does
Page 288 and 289: physics is not able to make, since
Page 290 and 291: Wittgenstein’s Attitudes Fabien S
Page 292 and 293: view of logic, in accordance with h
Page 294 and 295: Warum man auf transzendentalphiloso
Page 296 and 297: Making the Mind Higher-Level Elizab
Page 298 and 299: Literature Churchland, P. M. (1995)
Page 300 and 301: Zwischen Humes Gesetz und „Sollen
Page 302 and 303: Mental Causation: A Lesson from Act
Page 304 and 305: Again, this is mistaken. Non-reduct
Page 306 and 307: auf alle Arten von kontingenten Pro
Page 308 and 309: Reduction, Sets, and Properties Ben
Page 310 and 311: Literature Egan, Andy (2004): ‘Se
Page 312 and 313: there is a tension between evaluati
Page 316 and 317: sufficient to explain all of the em
Page 318 and 319: that 'A' and 'A' could never be tok
Page 320 and 321: the necktie sense) can differ in wh
Page 322 and 323: Supervenience and ‘Should’ Arto
Page 324 and 325: word 'norm' can be thought to expre
Page 326 and 327: Rule-following as Coordination: A G
Page 328 and 329: vast majority are bent: gruesome, g
Page 330 and 331: human life, the violent condition o
Page 332 and 333: A Division in Mind. The Misconceive
Page 334 and 335: A Division in Mind. The Misconceive
Page 336 and 337: could then deduce (1c) and conclude
Page 338 and 339: Neutral Monism. A Miraculous, Incoh
Page 340 and 341: Neutral Monism. A Miraculous, Incoh
Page 342 and 343: A somewhat Eliminativist Proposal a
Page 344 and 345: Impliziert der intentionale Redukti
Page 346 and 347: Impliziert der intentionale Redukti
Page 348 and 349: Structure of the Paradoxes, Structu
Page 350 and 351: Structure of the Paradoxes, Structu
Page 352 and 353: that he was working on a translatio
Page 354 and 355: Objects of Perception, Objects of S
Page 356 and 357: Objects of Perception, Objects of S
Page 358 and 359: Let us take the example of the hypo
Page 360 and 361: Reducing Sets to Modalities Rafał
Page 362 and 363: variables ai that (under an interpr
Page 364 and 365:
Are Lamarckian Explanations Fully R
Page 366 and 367:
A Note on Tractatus 5.521 Nuno Vent
Page 368 and 369:
And he goes on saying, referring to
Page 370 and 371:
The Place of Theory Reduction in th
Page 372 and 373:
Ethik als irreduzibles Supervenienz
Page 374 and 375:
unabhängig von denen ihrer Mitglie
Page 376 and 377:
Das ‘schwierige Problem’ des Be
Page 378 and 379:
The Supervenience Argument, Levels,
Page 380 and 381:
The Supervenience Argument, Levels,
Page 382 and 383:
their relations both to the physica
Page 384 and 385:
On the Characterization of Objects
Page 386 and 387:
On the Characterization of Objects
Page 388 and 389:
The Functional Unity of Special Sci
Page 390 and 391:
size, location, temporal characteri
Page 392 and 393:
Transcendental Philosophy and Mind-
Page 394 and 395:
From Topology to Logic. The Neural
Page 396 and 397:
From Topology to Logic. The Neural
Page 398 and 399:
The Calculus of Inductive Construct
Page 400 and 401:
The Four-Color Theorem, Testimony a
Page 402 and 403:
experience. Of course, a detailed a
Page 404 and 405:
Clearly x = y → x = ext y holds,
Page 406 and 407:
Intentional Fundamentalism Petri Yl
Page 408 and 409:
The second dimension to be consider
Page 410 and 411:
The very existence of the special s
Page 412 and 413:
Heil, J. (1999), “Multiple Realiz
Page 414 and 415:
independent of what is the fact, fa
Page 416 and 417:
415
Page 418:
Publications of the Austrian Ludwig
show all

Reduction and Elimination in Philosophy and the Sciences

You also want an ePaper? Increase the reach of your titles

Delete template?

Save as template?