Citation Context

...mation whether the thing contacting it will provide sustenance. If it ‘judges’ in the affirmative, it will release paralyzing darts from its snout and then ingest its prey, usually another protozoan (=-=Alberts et al., 1989-=-, p. 22). Multicellular organisms have evolved more # Blackwell Publishing Ltd. 2004SINBAD Neurosemantics 213 elaborate mechanisms to accomplish even better ‘predictive’ tricks. In the mammalian nerv...

Citation Context

...tionally poor Hebbian signaldoesnotassignblameatall;itisnotanerror signal. This rule would place severe limitations on what functions dendrites could learn (Anderson, 1995; Shouval and Perrone, 1995; =-=Elman et al., 1996-=-, p. 57; Phillips and Singer, 1997; Hanson, 1999). With simple error-correction learning, e.g. the delta rule or the perceptron rule, some of these limitations would be overcome. If the learning algor...

Citation Context

...ons to which I attribute representational powers are pyramidal cells (see figure 1), the most common cell type in the cerebral cortex (70% to 80% of the neurons in the cortex fall into this class—see =-=Abeles, 1991-=-; Douglas and Martin, 1998). In this section and the next, I will describe a ‘predictive’ trick that may be performed by networks of these cells, and which takes advantage of patterns of regularities ...

Citation Context

... cell—call it ‘Sinbad’—that has acquired its ‘predictive’ abilities through exposure to water will not be equivocal between water (H2O) and the superficially identical twin-water (XYZ) on twin-Earth (=-=Putnam, 1975-=-), because twin-water is not an explanatory source for it. Similarly, Twinbad (on twin-Earth) refers to twin-water, not water. These representational facts will hold even if Sinbad and Twinbad are ins...

Citation Context

...m glossing over many subtleties here. In fact, there is no agreed-upon formulation of the etiological idea, originally due to Wright (1973), or of its evolutionary implementation; for discussion see (=-=Millikan, 1984-=-; Allen et al., 1998; Ariew et al., 2002). Since the SINBAD mechanism does not stray near any hotly disputed territory, I think I am safe in assuming that the correct etiological theory will dictate t...

Citation Context

...find a correlation (for a review of dendritic growth, which occurs throughout life, see Quartz and Sejnowski, 1997). If it is still unsuccessful, at some point the cell will ‘give up’ and degenerate (=-=Edelman, 1987-=-). # Blackwell Publishing Ltd. 2004222 D. Ryder more and more robust dendritic matching, the correlational seed ends up producing a complex dendritic tree, which realizes complex functions relating t...

Citation Context

...cross time as it learns more about the source of correlation it is supposed to correspond to. This is a good thing, if SINBAD neurosemantics is ultimately to ground propositional attitude ascription (=-=Fodor, 1998-=-; Millikan, 1998). However, sometimes we need to draw attention to two different ways we have of representing the same source of correlation, e.g. Superman. In a SINBAD network, this would occur when ...

Citation Context

...., 2003). 27 Some advocates include (Burge, 1979; Millikan, 1984; Lycan, 1988; McGinn, 1989; Stalnaker, 1989; Devitt, 1990; Bilgrami, 1992; Dretske, 1993; Baker, 1995; Perry, 2001). For dissent, see (=-=Searle, 1983-=-; Segal, 2000). # Blackwell Publishing Ltd. 2004SINBAD Neurosemantics 233 mental content. Our thoughts of horses are not somehow contaminated with thoughts of rods and cones. The first thing to note ...

Citation Context

... have the capacity to misrepresent. Along with the representation of the non-existent, this is a hallmark of intentionality, and a crucial test for any theory of mental representation (Dretske, 1986; =-=Fodor, 1987-=-). There are actually several forms of representational error, but the usual place to start is with perceptual misrepresentation, i.e. the perceptual application of a representation to some object or ...

Citation Context

...he cell’s activity will be low, and it will elaborate its dendrites in search of new inputs until it is able to find a correlation (for a review of dendritic growth, which occurs throughout life, see =-=Quartz and Sejnowski, 1997-=-). If it is still unsuccessful, at some point the cell will ‘give up’ and degenerate (Edelman, 1987). # Blackwell Publishing Ltd. 2004222 D. Ryder more and more robust dendritic matching, the correla...

Citation Context

...ter of perceptual application elsewhere; 22 here we can get by with the crude (but usual) assumption that a perceptual representation is applied to the object or event that occasions it (Fodor, 1987; =-=Dretske, 1988-=-). Models are composed of elements that enter into relations. In the solar system model, the elements are balls, and they enter into spatial and dynamic relations; in a SINBAD network, the elements ar...

Citation Context

...ct these other, non-pyramidal cells (whose effects are purely local) are ancillary to the basic SINBAD mechanism (see Favorov and Ryder, forthcoming; Favorov et al., 2003). 27 Some advocates include (=-=Burge, 1979-=-; Millikan, 1984; Lycan, 1988; McGinn, 1989; Stalnaker, 1989; Devitt, 1990; Bilgrami, 1992; Dretske, 1993; Baker, 1995; Perry, 2001). For dissent, see (Searle, 1983; Segal, 2000). # Blackwell Publishi...

Citation Context

...ne. However, the theory I present is likely to be consistent with alternative accounts, if they prove viable. In particular, some may be attracted to a non-historical account of SINBAD teleofunction (=-=Cummins, 1975-=-; Bigelow and Pargetter, 1987; Schroeder, 1998; McLaughlin, 2001), since this would allow Swampman (Boorse, 1976; Davidson, 1987) to have full-fledged representations. If any teleofunction has a chanc...

Citation Context

...s of correlation our brains need to have some cells tune to, consider the fact that in the densely interconnected human cerebral cortex, there are somewhere between 11 and 25 billion pyramidal cells (=-=Pakkenberg and Gundersen, 1997-=-). Compare this to a good adult vocabulary of 50,000 words. (There is also a mechanism to prevent too many cells from tuning to the same source of correlation—see Ryder and Favorov, 2001; Favorov and ...

Citation Context

...ion responsible for regularities in raw input data. These cells’ outputs will be the following layer’s inputs, so cells in this next layer will tune to sources of correlation among edges, e.g. depth (=-=Becker and Hinton, 1992-=-) and types of translational motion. The subsequent layer may begin to tap into movable shapes (which are sources of correlations involving motion and edge depths), and so on (Favorov and Ryder, forth...

Citation Context

...rrelations in cell SB lies in kind K—K is what I call the ‘explanatory source’ for this post-learning cell. Now, evolution 22 With reference to Ramsey’s vision of beliefs as ‘maps by which we steer’ (=-=Ramsey, 1931-=-), the subject of perceptual application belongs in an analysis of steering (Ryder, 2002). Here I am giving an account of the maps. 23 Such hardwired representations will rarely be mental, in my view....

Citation Context

... effects are purely local) are ancillary to the basic SINBAD mechanism (see Favorov and Ryder, forthcoming; Favorov et al., 2003). 27 Some advocates include (Burge, 1979; Millikan, 1984; Lycan, 1988; =-=McGinn, 1989-=-; Stalnaker, 1989; Devitt, 1990; Bilgrami, 1992; Dretske, 1993; Baker, 1995; Perry, 2001). For dissent, see (Searle, 1983; Segal, 2000). # Blackwell Publishing Ltd. 2004SINBAD Neurosemantics 233 ment...

Citation Context

...me may be attracted to a non-historical account of SINBAD teleofunction (Cummins, 1975; Bigelow and Pargetter, 1987; Schroeder, 1998; McLaughlin, 2001), since this would allow Swampman (Boorse, 1976; =-=Davidson, 1987-=-) to have full-fledged representations. If any teleofunction has a chance of being explicated non-historically, those of SINBAD cells do. This is because a SINBAD network is intuitively a very ‘genera...

Citation Context

... explains the cell’s achievement of equilibrium. (This is not a purely causal explanation, but of course not all explanation is purely causal.) Combined with an adequate structuralism in mathematics (=-=Benacerraf, 1965-=-; Resnik, 2000), this suggests a way in which the representation of numbers may be accommodated. Millikan’s (1998; 2000) ‘substances’ are very similar to what I am calling sources of correlation. She ...

Citation Context

... Ryder And what determines the nature of the influence they come to exert? This influence can be quite nuanced, and it has been shown that dendrites are capable of computing highly complex functions (=-=Mel, 1994-=-; Mel, 1999; Gurney, 2001; Poirazi and Mel, 2001). The SINBAD idea is that learning in cortical pyramidal cells is governed by the following principle: each principal dendrite will adjust its connecti...

Citation Context

...ldoesnotassignblameatall;itisnotanerror signal. This rule would place severe limitations on what functions dendrites could learn (Anderson, 1995; Shouval and Perrone, 1995; Elman et al., 1996, p. 57; =-=Phillips and Singer, 1997-=-; Hanson, 1999). With simple error-correction learning, e.g. the delta rule or the perceptron rule, some of these limitations would be overcome. If the learning algorithm amounts to full-scale errorba...

Citation Context

... which pattern permits SINBAD networks to ‘predict’. This clustering pattern has been used as the basis for several closely related, metaphysically innocuous accounts of natural kinds (see especially =-=Boyd, 1991-=-; Kornblith, 1993; Millikan, 1999). We will be going beyond such kinds, but they are a natural place to start. The ‘unified property cluster’ account says that a natural kind is characterized by a set...

Citation Context

...in the full confidence of discovering new properties which were by no means implied in those we previously knew. Mill is describing the phenomenon that biological kinds tend to be ‘inductively rich’ (=-=Gelman and Coley, 1991-=-), i.e. they tend to be sources of large amounts of predictively useful correlation. In the case of biological kinds, these correlations are not due to an underlying chemical structure, but rather due...

Citation Context

... cells (whose effects are purely local) are ancillary to the basic SINBAD mechanism (see Favorov and Ryder, forthcoming; Favorov et al., 2003). 27 Some advocates include (Burge, 1979; Millikan, 1984; =-=Lycan, 1988-=-; McGinn, 1989; Stalnaker, 1989; Devitt, 1990; Bilgrami, 1992; Dretske, 1993; Baker, 1995; Perry, 2001). For dissent, see (Searle, 1983; Segal, 2000). # Blackwell Publishing Ltd. 2004SINBAD Neurosema...

Citation Context

... ‘confusing’ the two, and doing its job less than ideally. (It would not be right to say the cell has a disjunctive content, because there is no representational separation of the disjuncts; see also =-=Millikan, 1998-=-.) So not only are SINBAD cells supposed to correspond to sources of correlation in the context of an isomorphism, the details of the SINBAD mechanism allow us to determine exactly which source(s) of ...

Citation Context

...e theory I present is likely to be consistent with alternative accounts, if they prove viable. In particular, some may be attracted to a non-historical account of SINBAD teleofunction (Cummins, 1975; =-=Bigelow and Pargetter, 1987-=-; Schroeder, 1998; McLaughlin, 2001), since this would allow Swampman (Boorse, 1976; Davidson, 1987) to have full-fledged representations. If any teleofunction has a chance of being explicated non-his...

Citation Context

...particular, some may be attracted to a non-historical account of SINBAD teleofunction (Cummins, 1975; Bigelow and Pargetter, 1987; Schroeder, 1998; McLaughlin, 2001), since this would allow Swampman (=-=Boorse, 1976-=-; Davidson, 1987) to have full-fledged representations. If any teleofunction has a chance of being explicated non-historically, those of SINBAD cells do. This is because a SINBAD network is intuitivel...

Citation Context

... they must also have the capacity to misrepresent. Along with the representation of the non-existent, this is a hallmark of intentionality, and a crucial test for any theory of mental representation (=-=Dretske, 1986-=-; Fodor, 1987). There are actually several forms of representational error, but the usual place to start is with perceptual misrepresentation, i.e. the perceptual application of a representation to so...

Citation Context

...urrently exhibiting (e.g. it would freeze at 0 C). This pattern of regularity organized around a ‘source of correlation’ is not restricted to chemical natural kinds. As J. S. Mill remarked (quoted in =-=Hacking, 1991-=-): . . .a hundred generations have not exhausted the common properties of animals or plants. . .nor do we suppose them to be exhaustible, but proceed to new observations and experiments, in the full c...

Citation Context

...what determines the nature of the influence they come to exert? This influence can be quite nuanced, and it has been shown that dendrites are capable of computing highly complex functions (Mel, 1994; =-=Mel, 1999-=-; Gurney, 2001; Poirazi and Mel, 2001). The SINBAD idea is that learning in cortical pyramidal cells is governed by the following principle: each principal dendrite will adjust its connections so that...

Citation Context

...he basic SINBAD mechanism (see Favorov and Ryder, forthcoming; Favorov et al., 2003). 27 Some advocates include (Burge, 1979; Millikan, 1984; Lycan, 1988; McGinn, 1989; Stalnaker, 1989; Devitt, 1990; =-=Bilgrami, 1992-=-; Dretske, 1993; Baker, 1995; Perry, 2001). For dissent, see (Searle, 1983; Segal, 2000). # Blackwell Publishing Ltd. 2004SINBAD Neurosemantics 233 mental content. Our thoughts of horses are not some...

Citation Context

... attribute representational powers are pyramidal cells (see figure 1), the most common cell type in the cerebral cortex (70% to 80% of the neurons in the cortex fall into this class—see Abeles, 1991; =-=Douglas and Martin, 1998-=-). In this section and the next, I will describe a ‘predictive’ trick that may be performed by networks of these cells, and which takes advantage of patterns of regularities due to sources of correlat...

Citation Context

...re matching dendritic functions—is it a horse stand-in or an animal stand-in? The things that explain its achievement of equilibrium are members of both kinds. This is an instance of the qua problem (=-=Devitt, 1981-=-): when one kind subsumes another kind, what determines which kind is represented? There will normally be a fact of the matter which kind explains how a cell has acquired its predictive abilities. The...

Citation Context

...Favorov and Ryder, forthcoming; Favorov et al., 2003). 27 Some advocates include (Burge, 1979; Millikan, 1984; Lycan, 1988; McGinn, 1989; Stalnaker, 1989; Devitt, 1990; Bilgrami, 1992; Dretske, 1993; =-=Baker, 1995-=-; Perry, 2001). For dissent, see (Searle, 1983; Segal, 2000). # Blackwell Publishing Ltd. 2004SINBAD Neurosemantics 233 mental content. Our thoughts of horses are not somehow contaminated with though...

Citation Context

...napses. Each principal dendrite—an entire tree-like structure attached to the cell body (see figure 1)—produces an activity determined by all of the excitatory and inhibitory inputs that it receives (=-=Spruston et al., 1999-=-). This activity is that dendrite’s output, which it passes onto the cell body. The output of the whole cell (delivered elsewhere via the cell’s axon) is determined in turn by the outputs of its five ...

Citation Context

...ite’s output, which it passes onto the cell body. The output of the whole cell (delivered elsewhere via the cell’s axon) is determined in turn by the outputs of its five to eight principal dendrites (=-=Feldman, 1984-=-). The input/output profile of a dendrite, and thus its contribution to the whole cell’s output, can be modified by adjusting the strengths of both excitatory and inhibitory synaptic connections, and ...

Citation Context

...mechanism (see Favorov and Ryder, forthcoming; Favorov et al., 2003). 27 Some advocates include (Burge, 1979; Millikan, 1984; Lycan, 1988; McGinn, 1989; Stalnaker, 1989; Devitt, 1990; Bilgrami, 1992; =-=Dretske, 1993-=-; Baker, 1995; Perry, 2001). For dissent, see (Searle, 1983; Segal, 2000). # Blackwell Publishing Ltd. 2004SINBAD Neurosemantics 233 mental content. Our thoughts of horses are not somehow contaminate...

Citation Context

...rning algorithm they implement. The informationally poor Hebbian signaldoesnotassignblameatall;itisnotanerror signal. This rule would place severe limitations on what functions dendrites could learn (=-=Anderson, 1995-=-; Shouval and Perrone, 1995; Elman et al., 1996, p. 57; Phillips and Singer, 1997; Hanson, 1999). With simple error-correction learning, e.g. the delta rule or the perceptron rule, some of these limit...

Citation Context

...cell body, which is due to the arrangement of principal dendrites (one extending upwards, the others radiating horizontally and obliquely downwards). dendrite as well (e.g. structural properties: see =-=Woolley, 1999-=-; McAllister, 2000). In this way, each dendrite comes to have a causal profile that implements some mathematical function relating synaptic inputs to dendritic output. The cell as a whole also comes t...

Citation Context

... networks to ‘predict’. This clustering pattern has been used as the basis for several closely related, metaphysically innocuous accounts of natural kinds (see especially Boyd, 1991; Kornblith, 1993; =-=Millikan, 1999-=-). We will be going beyond such kinds, but they are a natural place to start. The ‘unified property cluster’ account says that a natural kind is characterized by a set of correlated properties, where ...

Citation Context

...they implement. The informationally poor Hebbian signaldoesnotassignblameatall;itisnotanerror signal. This rule would place severe limitations on what functions dendrites could learn (Anderson, 1995; =-=Shouval and Perrone, 1995-=-; Elman et al., 1996, p. 57; Phillips and Singer, 1997; Hanson, 1999). With simple error-correction learning, e.g. the delta rule or the perceptron rule, some of these limitations would be overcome. I...

Citation Context

...ancillary to the basic SINBAD mechanism (see Favorov and Ryder, forthcoming; Favorov et al., 2003). 27 Some advocates include (Burge, 1979; Millikan, 1984; Lycan, 1988; McGinn, 1989; Stalnaker, 1989; =-=Devitt, 1990-=-; Bilgrami, 1992; Dretske, 1993; Baker, 1995; Perry, 2001). For dissent, see (Searle, 1983; Segal, 2000). # Blackwell Publishing Ltd. 2004SINBAD Neurosemantics 233 mental content. Our thoughts of hor...

Citation Context

...Craik’s The Nature of Explanation (1943). See also Cummins (1989) and McGinn (1989, Ch. 3). 9 SINBAD models are to be distinguished from self-organizing feature maps, as achieved by Kohonen networks (=-=Ritter, 1995-=-) and competitive learning (Intrator, 1995). These merely produce a map that is topographically organized by stimulus similarity, with more space devoted to more common stimuli. They are not useful fo...

Citation Context

...ill always match. That is, the two dendrites will have to compute the same function of their inputs (or mathematically equivalent functions). However, dendrites do not get the same inputs, as a rule (=-=Favorov and Kelly, 1996-=-). Thus in the typical situation, the two dendrites’ matching task will not be trivial. The learning rule forces each dendrite to try to match the other’s outputs by modifying how it treats its inputs...

Citation Context

...isnotanerror signal. This rule would place severe limitations on what functions dendrites could learn (Anderson, 1995; Shouval and Perrone, 1995; Elman et al., 1996, p. 57; Phillips and Singer, 1997; =-=Hanson, 1999-=-). With simple error-correction learning, e.g. the delta rule or the perceptron rule, some of these limitations would be overcome. If the learning algorithm amounts to full-scale errorbackpropagation,...

Citation Context

...urely local) are ancillary to the basic SINBAD mechanism (see Favorov and Ryder, forthcoming; Favorov et al., 2003). 27 Some advocates include (Burge, 1979; Millikan, 1984; Lycan, 1988; McGinn, 1989; =-=Stalnaker, 1989-=-; Devitt, 1990; Bilgrami, 1992; Dretske, 1993; Baker, 1995; Perry, 2001). For dissent, see (Searle, 1983; Segal, 2000). # Blackwell Publishing Ltd. 2004SINBAD Neurosemantics 233 mental content. Our t...

Citation Context

.... See also Cummins (1989) and McGinn (1989, Ch. 3). 9 SINBAD models are to be distinguished from self-organizing feature maps, as achieved by Kohonen networks (Ritter, 1995) and competitive learning (=-=Intrator, 1995-=-). These merely produce a map that is topographically organized by stimulus similarity, with more space devoted to more common stimuli. They are not useful for ‘prediction’ since they do not mirror, i...

Citation Context

...rn permits SINBAD networks to ‘predict’. This clustering pattern has been used as the basis for several closely related, metaphysically innocuous accounts of natural kinds (see especially Boyd, 1991; =-=Kornblith, 1993-=-; Millikan, 1999). We will be going beyond such kinds, but they are a natural place to start. The ‘unified property cluster’ account says that a natural kind is characterized by a set of correlated pr...

Citation Context

...help of SINBAD neurosemantics, notably the relation between concepts and inferential roles, the explanatory role of content, propositional attitudes, and concept acquisition (Ryder and Favorov, 2001; =-=Ryder, 2002-=-; Favorov et al., 2003). 2. The organization of regularities around sources of correlation Perception and cognition, the processes that involve mental representations, are valuable because they afford...