Evolving Artificial Minds And Brains Pete Mandik-Books Pdf

Evolving artificial minds and brains Pete Mandik
08 Apr 2020 | 15 views | 0 downloads | 20 Pages | 310.74 KB

Share Pdf : Evolving Artificial Minds And Brains Pete Mandik

Download and Preview : Evolving Artificial Minds And Brains Pete Mandik


Report CopyRight/DMCA Form For : Evolving Artificial Minds And Brains Pete Mandik



Transcription

Mental States, behavior and to sketch an account of what sorts of things representations must be if. they are to explain intelligent behavior, Several opponents of representational explanations have built their cases by. starting with the simplest examples of intelligent behavior and attempting to dem. onstrate that in such examples no representations are to be found and thus no. representations need be referred to in order to explain the behaviors at hand This. is the strategy followed for example by roboticists and artificial intelligence re. searchers such as Brooks 1991 and Beer 1990 in their arguments for the pos. sibility of intelligence without representation We will employ a similar strategy. but we will be drawing different conclusions We will examine some of the simplest. cases of intelligent behavior and demonstrate that in these cases the behavior at. hand is best explained in terms of representations Further our account of repre. sentations will be fully realist and reductive To say that the account is realist is to. say that the attributions aren t purely instrumental ways of speaking as if the crea. tures had representations It is instead to pick out states of creatures that would be. there independently of our speaking of them To say of our account that it is reduc. tive we will be identifying representational states in ways that are straightforward. ly explicable in terms of states of creatures nervous systems and relations between. their neural states and environmental states, One way to examine the simplest examples of intelligent behavior is to examine. the simplest examples of organisms that behave intelligently This strategy confers. the following advantage The simpler the creature the easier it will be to keep track. of the creature s internal structures the structures of the creature s environment and. the relations between the two kinds of structure in virtue of which the former count. as representations of the latter Further dealing with extremely simple cases will al. low for tractable computer simulations of creature behavior as well as simulations of. the evolutionary forces that contribute to the emergence of such behaviors. Our motive for caring about the evolutionary background of the simplest cog. nitive behaviors emerges from the following presumptions We presume and are. unlikely alone in doing so that the simplest forms of intelligent behaviors are. adaptive That is intelligent behaviors at least of the simplest varieties provide. biological benefits to the organisms that perform them We presume also that just. as there was a time in the history of the universe that there were no biological or. ganisms there was a time in the history of the universe that there were no organ. isms performing intelligent behaviors Since abiogenesis is the term referring to the. hypothesized emergence of life from non living matter we coin the term apsycho. genesis to refer to the hypothesized emergence of intelligence from non intelligent. systems When in the history of the universe did abiogenesis and apsychogenesis. occur No one knows but we doubt that apsychogenesis preceded abiogenesis. They either coincided or abiogeneis occurred first However the latter option. Chapter 5 Evolving artificial minds and brains, strikes us as the more plausible of the two Adding to our growing list of presump. tions we further presume that the problem of understanding apsychogenesis is. best understood in the context of an evolutionary framework Thus we are led to. ask What pressures applied to non intelligent organisms yielded the earliest and. simplest forms of intelligence If mental representations are to underwrite intelli. gent behavior then questions of the evolvability of intelligence will be closely re. lated to questions of the evolvability of mental representations. We will tackle the topics of intelligence representation and evolution by ex. amining computer simulations of evolved adaptive behaviors The simulated or. ganisms behaviors and environments will be simple enough to make tractable. questions concerning the relations that constitute representation and the roles rep. resentations play in adaptive intelligent behaviors. The structure of the rest of the paper is as follows First we will briefly examine. a few cases in which representations are invoked to explain the intelligent behav. iors of humans and non human animals The goal here will not be to extract a. definition of representation from these examples but instead to only note a few key. features of the roles representations play in such explanations Formulating a defi. nition of representation is a goal to be achieved or at least approximated toward. the end of the paper and not a presupposition to be made at its outset Following. the examination of these sample explanations we will describe the basic intelligent. behavior of positive chemotaxis and highlight the ways in which the problem that. chemotaxis poses for organisms can be solved in a variety of ways involving repre. sentations Next we describe mathematical and computer models of positive. chemotaxis The models are informed by neuroanatomical and neurophysiological. data from real animals Finally we discuss what account of representation seems. best supported by the models, 2 Mental representations in explanations of intelligent behavior.
Let us take a brief look at a folk psychological explanation of a piece of intelligent. behavior Consider George George is opening a refrigerator Why What explana. tion is available for this action A folk psychological explanation will advert to a. collection of psychological states that jointly constitute a cause of George s behav. ior An example collection of such states would include a desire a perception and. a memory One explanation of Georges behavior then would advert to George s. desire to drink some beer George s visual perception that there is a refrigerator in. front of him and George s memory that he put some beer in the refrigerator the. day before,Mental States, There are a few useful points to note about this explanation First the psycho. logical states are not individually sufficient to cause a behavior but must act in. concert A belief that there is beer in front of you will contribute to causing you to. move toward it if combined with a desire for beer and will contribute to causing. you to move away from it if combined with a fear of beer Similarly a desire for. beer will contribute to causing you to move forward if combined with a belief that. beer lies ahead and cause you to move in some other direction if combined with. some other belief In summary psychological states contribute to the causes of. behavior by acting in concert, A second useful point to note about this sort of explanation is that the psycho. logical states are identified in part by their representational content and in part by. what attitude the person is taking toward that content In the case of George s. memory that he put some beer in the refrigerator the representational content of. the memory is that George put some beer in the refrigerator and the attitude is one. of remembering Different types of attitude can be taken toward one and the same. content e g remembering buying beer planning on buying beer and one and the. same attitude type can be taken toward different contents e g perceiving that. there is a beer in front of me perceiving that there is a slice of pizza in front of me. In summary psychological states that are causes of intelligent behaviors admit of a. distinction between their representational contents and the attitude that is taken. toward those representational contents, A third useful point to note about these sorts of explanation is that we can. make attributions of content without explicit knowledge of what in general rep. resentational content is We construct such explanations on the fly without know. ing for example what the right theory of content is or even having a theory of. content in mind We plan to exploit this in what follows We will present relatively. clear cases of synthetic organisms that behave in ways explainable in terms of rep. resentational states and we will do so before offering a definition of what represen. tations are or what representational content is This leaves open to empirical inves. tigation what the best accounts of representation and content are as opposed to a. matter that must be settled a priori before such investigations take place. It is worth noting that the power of representational explanation is not simply. some story we tell ourselves and each other sustained by our own possibly mis. taken views of ourselves One way to appreciate the power of such explanations is. to appreciate them in the context of explaining the behaviors of non human ani. mals The literature is filled with such examples We briefly mention just a few. Consider the impressive feats of maze learning exhibited by rats A Morris water. maze is filled with water rendered opaque to obscure a platform that will offer a rat. a chance to rest without having to tread water When placed in the maze for a first. time a rat will explore the area and eventually find the platform When the rat is. Chapter 5 Evolving artificial minds and brains, returned to the starting position the rat does not repeat the exploratory strategy. but instead swims straight to the remembered location of the platform Appar. ently the perceptual inputs gained during the exploration were utilized to com. pute the straight line path to the platform The rat s behavior is thus explicable in. terms of psychological states such as perceptions and memories and computations. that operate over them Much more detail can be given to be sure but for now our. main concern is only to call these sorts of explanation to the reader s attention. Much more detail concerning for instance the neural underpinnings of percep. tion memory and computation will be supplied later Gallistel 1990 2 describes. another such example, Every day two naturalists go out to a pond where some ducks are overwintering.
and station themselves about 30 yards apart Each carries a sack of bread chunks. Each day a randomly chosen one of the naturalists throws a chunk every 5 sec. onds the other throws every 10 seconds After a few days experience with this. drill the ducks divide themselves in proportion to the throwing rates within 1. minute after the onset of throwing there are twice as many ducks in front of the. naturalist that throws at twice the rate of the other One day however the slower. thrower throws chunks twice as big At first the ducks distribute themselves two. to one in favor of the faster thrower but within 5 minutes they are divided fifty. fifty between the two foraging patches Ducks and other foraging animals can. represent rates of return the number of items per unit time multiplied by the average. size of an item emphasis ours, In both the cases of the rats and the ducks the ultimate explanation called for is. going to require mention of some relatively subtle mechanisms inside of the ani. mals that are sensitive to properties of the environment To get a feel for what. might be called for contrast the way in which we would explain on the one hand. the movements of the rat toward the platform or the duck toward the bread and. on the other hand a rock falling toward the earth The rock s movement is ex. plained by a direct appeal to a fundamental force of nature that constitutes the at. traction between the respective masses of the earth and the rock Such a direct. appeal to a fundamental force will not explain the rat s movement to the platform. This is not to say of course that something non physical is transpiring between. the rat and the platform There is of course energy flowing between the two that. impacts the rat in ways that ultimately explain its behavior But unlike the case of. the rock the transference of energy from platform to rat will only have an impact. on the rat s behavior insofar as the rat is able to transduce the information carried. by that energy into a code that can be utilized by information processing mecha. nisms in its central nervous system Such mechanisms will be able to store infor. mation in the form of encoded memories and make comparisons between en. coded memories and current sensory input to compute a course of action toward. Mental States, a goal state Going into further detail of how the nervous system of an animal. might encode such information and perform such computations can get quite. complicated Before proceeding it will be useful to turn our attention toward nerv. ous systems much simpler than those of vertebrates. 3 Modeling the simplest forms of intelligence, Chemotaxis directed movement in response to a chemical stimulus is one of the. simplest forms of organism behavior It is an adaptive behavior as when for exam. ple positive chemotaxis is used to move toward a food source or negative chemo. taxis is used to move away from a toxin The underlying mechanisms of chemo. taxis are relatively well understood and amenable to modeling and simulation. Mandik 2002 2003 2005 Chemotaxis is appropriate to regard as cognitive As we. will argue below it constitutes what Clark and Toribio 1994 call a representation. hungry problem To appreciate the informational demands that chemotaxis places. upon an organism it is useful to consider the problem in the abstract The central. problem that must be solved in chemotaxis is the navigation of a stimulus gradient. and the most abstract characterization would be the same for other taxes such as. thermotaxis or phototaxis To focus on a simplified abstract case of positive photo. taxis imagine a creature traversing a plane and utilizing a pair of light sensors one. on the left and one on the right Activity in each sensor is a function of how much. light is falling on in it in such a way that the sensor closer to the source of light will. have a greater degree of activation Thus the difference in the activity between the. two sensors encodes the location of the light source in a two dimensional egocen. tric space Information encoded by the sensors can be relayed to and decoded by. motor systems responsible for steering the creature For example left and right op. posing muscles might have their activity be directly modulated by contralateral. sensors so that the greater contraction corresponds to the side with the greatest. sensor activity thus steering the creature toward the light. Consider now the problem of phototaxis as confronted by a creature with only. a single sensor The one sensor creature will not be in a position to directly perceive. the direction of the light since activity in a single sensor does not differentiate from. say light being three feet to the left or three feet to the right Of course the creature. might try to exploit the fact that the sensor is moving and make note of changes in. sensor activity over time but such a strategy will be available only to creatures that. have some form of memory Exploiting the change of sensor activity will require a. means of comparing the current sensor activity to some past sensor activity. Note the folk psychological explanation of how a human would solve the. problem of one sensor taxis To imagine that you are in a gradient it will do to. Chapter 5 Evolving artificial minds and brains, imagine that you are literally in a fog so dense that while you can ascertain how. dense it is where you are you cannot ascertain in which direction the fog gets. more dense and in which direction it gets less dense However after walking for a. while you notice that the fog is much less dense than it was previously By compar. ing your current perception of a less dense fog to your memory of a more dense. fog against the background of your knowledge that you have been walking it is. reasonable for you to infer that you are moving out of the area of greatest concen. tration Conversely if your current perception reveals a greater concentration of. fog than remembered it is reasonable for you to infer that you should turn around. if you want to get out of the fog, There are several points we should get from the above discussion The first is that.
the informational demands of one sensor chemotaxis can be readily appreciated. from the point of view of folk psychological explanation The same point of view. allows us to construct possible solutions to the problem of one sensor chemotaxis A. creature that is both able to perceive the current concentration and remember the. past concentration is thus in the position to make an inference about whether to. keep moving ahead or turn in order to reach a desired location in the gradient. One sensor chemotaxis is accomplished by natural organisms One particu. larly well studied example is the nematode worm Caenorhabditis Elegans C Ele. gans Despite having four chemosensors a pair in the head and a pair in the tail. there are good reasons to believe that the worm effects one sensor not four sen. sor chemotaxis Ferr e Lockery 1999 First off the worms are able to effect. chemotaxis even when their tail sensors are removed Second the two sensors in. the head are too close together for there to be an appreciable difference between. the activity in each of them in response to local concentration of attractant Third. when navigating chemical gradients on the effectively two dimensional surface of. a Petri dish the worms are positioned on their sides with the pair of head sensors. orthogonal to the gradient Fourth artificial neural network controllers inspired. by the neurophysiology of C Elegans with only a single sensor input are able to. approximate real chemotaxis behaviors in simulated worms These simulations are. especially interesting to examine in some detail, We next briefly review work done in simulating C Elegans chemotaxis in. Shawn Lockery s lab at the University of Oregon Institute of Neuroscience In par. ticular we focus here on work reported in Ferr e and Lockery 1999 263 277 and. Dunn et al 2004 Ferr e and Lockery construct a mathematical model of the con. trol of C Elegans whereby the time derivative of the chemical concentration is. computed and used to modulate the turning rate of the worm in the gradient One. of our purposes in reviewing this work is to point out how it at best supplies only. a partial explanation of how the actual nervous systems of C Elegans regulates. chemotaxis Ferr e and Lockery begin by constructing a model network that makes. Mental States, many simplifying assumptions about the neuroanatomy and neurophysiology of. the relevant circuits in C Elegans They hypothesize that the worm must assess the. gradient by computing the temporal derivative of concentration as it moves through. the chemical environment and that the behavioral upshot of this assessment is that. the worm attempts to keep its head pointed up the gradient Their model network. consists of five neurons whose various states of activation model voltage The single. sensory input has a state of activation that reflects the local concentration of the. chemical attractant Two output neurons model the voltages of dorsal and ventral. motor neurons whose relative voltages determine the worm s neck angle The re. maining three neurons are interneurons Each of the five neurons is connected to. every other neuron by both feed forward and feedback connections thus making a. recurrent network Ferr e and Lockery optimized network parameters by using a. simple simulated annealing training algorithm to maximize a fitness function de. fined in terms of the change of chemical concentration The optimized network. resulted in simulated worm behavior similar to that of real worms oriented move. ment up the gradient and persistent dwelling at the peak However Ferr e and. Lockery point out that it is not obvious how the networks are effecting these behav. iors Simple inspection of the parameters does not necessarily lead to an intui. tive understanding of how the network functions however because the neural ar. chitecture and optimization procedure often favor a distributed representation of. the control algorithm To derive an intuitive mathematical expression for this al. gorithm they manipulated the analytic solution to the linear system of equations. that comprise their mathematical model The analytic solution for the linear recur. rent network is an equation wherein the rate turning is equal to the sum of a turn. ing bias and the cumulative effect of chemosensory input on the rate of turning. This equation produces exactly the same response to chemosensory input as the. original optimized network In order to further improve our intuition about. chemotaxis control in this model Ferr e and Lockery produce a Taylor expansion. of the equation in time derivatives of the input The extracted rule for chemotaxis. control equates rate of turning with a sum whose first term is a turning bias the. second term is the zeroth time derivative of chemical concentration the third term. is the first order time derivative of chemical concentration the fourth term is the. second order time derivative of chemical concentration and so on Next they com. pared simulated behavior wherein only some of the terms are kept With just the. turning bias and the zeroth order term the resultant behavior was not chemotaxis. but instead just a circular motion around the starting position Adding the first. order term resulted in chemotaxis as did adding the first and second order terms. Likewise adding the first order but omitting the zeroth order term. Ferr e and Lockery describe their accomplishment as follows Using analyti. cal techniques from linear systems theory we extracted computational rules that. Chapter 5 Evolving artificial minds and brains, describe how these linear networks control chemotaxis Ferr e Lockery 1999. 276 However we find the resultant mathematical descriptions unsatisfying inso. far as they do not constitute explanations of how the networks effect chemotaxis. And they do not constitute explanations because too little has yet been said about. what the underlying mechanisms are and how it is that they are functioning When. we say that they do not supply a complete account of the mechanism by mecha. nism we intend it in the sense of Craver 2001 58 Mechanisms are collections. of entities and activities organized in the production of regular changes from start. or set up conditions to finish or termination conditions See also Craver Darden. 2001 Machamer Darden Craver 2000 Bechtel Richardson 1993. To get a feel for what we think is still missing recall the earlier discussion be. tween the difference between two sensor chemotaxis and one sensor chemotaxis. In the case of two sensor chemotaxis the difference in activity between the left. and right sensors can be straightforwardly exploited by a steering mechanism that. would guide the animal right up the gradient For example left and right steering. muscles could be connected to the sensors in such a way that the greater activity in. the right sensor will result in a greater contraction in the right steering muscle thus. turning the head of the worm toward the direction of the greatest concentration. If the worm s head is pointed directly in the direction of the greatest concentration. then the activity in the left and right sensors will be approximately equal as will be. the amount of contraction in the left and right steering muscles thus keeping the. worm on course In this description of the two sensor case we have at least a. sketch of what the mechanisms underlying chemotaxis are We are not in a com. parable position yet with Ferr e and Lockery s mathematical description The. computation rule tells us that the time derivative of the concentration is being. computed but we are not yet in a position to see how it is being computed We. know enough about the underlying mechanisms to know that there is sufficient. information present upon which to compute the time derivative because we know. that the chemical concentration detected by the sensor is changing over time as. the worm moves through the environment However we need to know more than. that the information is there We need to know how the information is encoded and. subsequently used by the organism As Akins 2001 381 puts a similar point. Information that is carried by but not encoded in a signal is information that is. available only in theory To say that the information is present is to say only that. there exists a computable function which if used would yield the correct result. It is present as it were from the point of view of the universe But no creature has. ever acted upon information that is available only in principle. Lockery and his colleagues are not blind to this sort of shortcoming In a subse. quent publication Dunn et al 2004 138 write The chemosensory neurons re. Mental States, sponsible for the input representation are known as are the premotor interneu. rons for turning behavior Much less is known about the interneurons that link. chemosensory input to behavioral output To get a further handle on what the in. terneurons might be doing Dunn et al run simulations of networks optimized for. chemotaxis The networks in these simulations have a single input neuron one. output neuron and eight interneurons All of the neurons in each network are con. nected to each other and have self connections as well After optimization and test. ing the networks that performed successful chemotaxis were subjected to a prun. ing procedure whereby unused neurons and connections were eliminated Dunn et. al report that the pruned yet still successful networks have only one or two in. terneurons and they all have inhibitory feedback among all of the neurons Dunn. et al proposed that the main function of this feedback is to regulate the latency. between sensory input and behavior but we note that while this latency regulation. may indeed be occurring it certainly does not explain how successful chemotaxis. is accomplished The mere introduction of a delay between input and response. surely cannot suffice for successful chemotaxis We hypothesize that the crucial yet. underappreciated mechanism in the successful networks is the existence of recur. rent connections Recurrence has been noted by many authors e g Mandik 2002. Churchland 2002 Lloyd 2003 as a mechanism whereby a network may instantiate. a form of short term or working memory since activity in the network will not. simply reflect the information currently coming into the sensory inputs but also. reflect information feeding back and thus representing past information that came. into the sensory inputs We hypothesize then that the recurrence is implementing. a form of memory that allows the network to compute the time derivative of the. concentration in virtue of both encoding information about the current concentra. tion in the state of the sensor and encoding information about past concentration. in the signal propagated along the recurrent connections. To test this hyopothesis we conducted our own simulations of C Elegans sin. gle sensor chemotaxis For our simulations we utilized the Framsticks 3 D Artifi. cial Life software Komosinski 2000 that allowed for the construction and testing. of worms in a simulated physics and the optimization of the worms using simu. lated Darwinian selection The morphologies of our synthetic worms are depicted. in Figure 1 and their neural network topologies are depicted in Figure 2 Networks. are modular One module constitutes a central pattern generator that regulates. forward motion by sending a sinusoidal signal to the chain of muscles that control. flagellation Another module regulates steering with a single sensory neuron three. interneurons and one output neurons This five neuron steering network is recur. rent with every neuron in it connected to every other. Chapter 5 Evolving artificial minds and brains, Figure 1 Synthetic C Elegans On the left front view On the right top view.
Figure 2 Neural network for the synthetic C Elegans Neurons include one sensor s and. several motor neurons m and interneurons i Single headed arrows indicate flow of in. formation from one neuron to the next A double headed arrow between two neurons in. dicates both a feed forward and a feedback connection between them. In our simulations the initial morphologies and network topologies were set by. hand The connection weights however were optimized through a Darwinian. process whereby mutations are allowed only for connection weights and not to. morphologies or network topologies Fitness is defined in terms of overall lifetime. distance This forced the worms both to maintain a high velocity and also to ex. tend their lives by replenishing their energy store with found food We compared. the performance of three kinds of orientation networks fully recurrent networks. with sensory input recurrent networks with no sensory input blind networks. Mental States, and strictly feed forward networks with sensory input Four populations of each. of the three kinds of orientation networks were subjected to the evolutionary sim. ulation for 240 million steps of the running program. Results are shown in Figure 3 of the lifetime distances averaged over the four. populations for each of the three kinds of orientation networks The performance. of the blind networks involved the maximal distance accomplished by worms with. maximally optimized velocities but no extension of lifespan through finding food. beyond whatever food they collided with accidentally Worms with sensory inputs. and recurrent connections were able to maximize their lifespan through food find. ing by chemotaxis Further their swimming behaviors were similar to those exhib. ited by real C Elegans directed movement up the gradient and dwelling at the. peak Worms without recurrent connections were conferred no advantage by sen. sory input Our explanation of this is that without the recurrent connections to. constitute a memory the worms are missing a crucial representation for the com. putation of the change of the local concentration over time We turn now to exam. ine the nature of these underlying representations. Figure 3 Results of the experiment comparing recurrent feed forward and blind net. works in an evolutionary simulation of chemotaxis,4 What the representations are in the models. We admittedly do not yet have a complete explanation of C Elegans chemotaxis. but we do have a pretty good sketch of what is going on Heading in the gradient. Chapter 5 Evolving artificial minds and brains, is determined by a computation that takes as inputs both a sensory representation. that encodes information about the current local concentration and a memory. representation that encodes information about the past local concentration The. existence of a memory mechanism was predicted by the folk psychological expla. nation and supported by the simulation experiments Further we are in a position. with respect to these models to make some remarks about what the representa. tions are and what relations obtain that determine the representational contents. In the orientation networks we may discern three types of representations sensory. representations memory representations and motor representations The sensory. representations are states of activations in the chemo sensory input neuron the. memory representations are signals conveyed along recurrent connections and. the motor representations are states of activation in neurons that output to mus. cles In each case the contents of the representations are the things that are repre. sented In the sensory case what is represented is current local concentration In. the memory case what is represented is past local concentration In the motor. case the representation is a command signal and what is represented is a level of. muscular contraction, The question arises of what the relation is between representation and the rep. resented is such that the former is a representation of the latter Two major sorts of. suggestion common in the philosophical literature on representational content. seem initially applicable to the case of the chemotaxis networks informational ap. proaches and isomorphism based approaches The first sort of suggestion is that. the relations that underwrite representation are causal informational On such a. suggestion it is in virtue of being causally correlated with a particular external. state that a particular internal state comes to represent it In the chemotaxis exam. ples there are indeed relations of causal correlation between the representations. and what they represent In the case of the sensory representation there is a reli. able causal correlation between the sensor state and the current local concentra. tion and in the memory case there is a reliable causal correlation between the re. current signal and the past local chemical concentration The informational view. must give a slightly different treatment of motor representations since commands. are the casual antecedents of their representational targets Mandik 1999 2005. The isomorphism suggestion seems applicable as well though before discuss. ing its application we need to spell out the relevant notion of isomorphism An. isomorphism is a structure preserving one to one mapping A structure is a set of. elements plus a set of relations defined over those elements So for example a set of. temperatures plus the hotter than relation constitutes a structure as does a set of. heights of a mercury column in a thermometer and the taller than relation A one. to one mapping exists between a set of temperatures and a set of heights just in case. Mental States, for any height and the next higher one they are mapped respectively to a tempera.
ture and the next hottest one, Information based theories of representational content make it a necessary. condition on a representation r of a thing c that r carry information about causally. correlate with c Isomorphism based theories of representational content make it. a necessary condition on a representation r of a thing c that r and c be elements in. structures wherein an isomorphism obtains that maps r to c Can we adjudicate. between the informational and isomorphism suggestions More specifically can. the way in which attributions of representation in the explanations of the network. control of chemotaxis be used to favor information based theories over isomor. phism based theories or vice versa We see the respective roles of the notions of. representation information and isomorphism in this context as follows The sen. sory and memory states are able to drive successful chemotaxis in virtue of the in. formational relationships that they enter into with current and past levels of local. chemical concentration but they are able to enter into those informational rela. tions because of their participation in isomorphsims between structures defined by. ensembles of neural states and structures defined by ensembles of environmental. states In brief in order to have the representational contents that they have they. must carry the information that they do and in order to carry the information that. they do they must enter into the isomorphisms that they do To spell this out a bit. further will require spelling out two things First why it is that representation re. quires information and second why information requires isomorphism. We begin with the reason why representation requires information A large. part of the reason representation requires information in the example of the chem. otaxis networks is because of the sorts of representation that we are talking about. namely sensory and memory representations It is part of the nature of sensory. states that they carry information about the current local situation of an organism. and part of the nature of memory states that they carry information about the past. Another way to appreciate the carrying of information is to realize that if the net. works didn t encode information about the current and past chemical concentra. tions then they would not be able to give rise to the successful chemotaxis behav. ior Consider the blind worms they were deprived of the means of encoding. information about the present chemical concentration Consider also the worms. with strictly feed forward networks Without recurrent connections they were. deprived of the means of encoding the relevant information about the past It. seems that the crucial aspect of attributing sensory and memory representations. in explaining successful one sensor chemotaxis is that such attributions track the. information bearing properties of the states, To see why isomorphism is important it helps to begin by considering how. hard it would be to not have isomorphism First off note that as Gallistel 1990. Chapter 5 Evolving artificial minds and brains, has pointed out a one to one mapping can be considered as structure preserving. even if the structures involved are defined only in terms of sets of elements and the. identity relation On such schemes the resultant representations are what Gallistel. calls nominal representations For example the set of numbers assigned to play. ers on a sports team is a set of nominal representations in this sense There is a. one to one mapping between numbers and players and the only relation between. numbers that is mapped onto a relation between players is identity one and the. same number can only be mapped onto one and the same player Larger numbers. however need not indicate larger or heavier players Nonetheless they still satisfy. the requirements for isomorphism since the mapping is structure preserving. Similarly even if the information bearing states of a nervous system constitute a. set of nominal representations of environmental states they would nonetheless. satisfy the requirements for isomorphism, Setting aside identity based nominal representations as genuine isomor. phisms there is still a serious difficulty the informational theorist faces concerning. the alleged dispensability of isomorphism Even if there were a logically possible. scheme that had information without isomorphism it is incredibly difficult if not. impossible for such a scheme to be evolved or learned We can see the point con. cerning evolution in the context of the synthetic C Elegans in our artificial life. simulations Organisms bodies as well as the environments they are situated in. contain many physical systems that have states that fall into natural ordering rela. tions Consider for example that chemical solutions can be more or less concen. trated or that neural firings can have higher or lower rates or higher or lower. voltages It is hard if not impossible to see how there could be a counter example. to the following claim Any situation in which a particular level of neural activa. tion can be used to carry information about a particular level of chemical concen. tration is also going to be a situation in which a slightly higher level of neural acti. vation can be used to carry information about a slightly higher level of chemical. concentration In other words organisms and their environments are rich in struc. tures and it is hard to see how elements in those structures can be evolved to enter. into informational relationships without the structures themselves also entering. into isomorphism relationships, While our argument is to our knowledge unique it is worth mentioning cer.
tain similarities between our argument which is specifically about evolution and. some other arguments that focus on learning that have appeared in the literature. on isomorphism Cummins 1996 and Churchland 2001 both endorse isomor. phism based theories of representational content and both argue that a creature. can only be in a position to have states that carry information about external states. if the creature s internal states are embedded in a network of internal states that.

Related Books

California State University, San Bernardino General ...

California State University San Bernardino General

California State University, San Bernardino General Education Program for 2020 AY Area A English Language Communication and Critical Thinking. A1 Oral Communication: COMM 1006: Oral Communication A1: HON 1300: Communicating Orally A1: A2 Written Communication: ENG 1060A: Stretch Composition II A2: ENG 1060B : Stretch Composition II for Multilingual Students A2: ENG 1070A: First-Year ...

Fifth Edition THE NEW URBAN SOCIOLOGY - GBV

Fifth Edition THE NEW URBAN SOCIOLOGY GBV

Fifth Edition THE NEW URBAN SOCIOLOGY Mark Gottdiener University at Buffalo Ray Hutchison UniversityofWisconsin-Green Bay Michael T. Ryan Dodge City Community College -0 WESTVIEW PRESS A Member of the Perseus Books Group . CONTENTS Preface to the Fifth Edition 1 THE NEW URBAN SOCIOLOGY Urban Regions, 1 Defining the Metropolitan Region, 5 Megacities Around the World, 9 A New Approach to Urban ...

ICWA SOUTHEAST ASIA MZW-3

ICWA SOUTHEAST ASIA MZW 3

MZW-3 SOUTHEAST ASIA Institute of Current World Affairs The Crane-Rogers Foundation Four West Wheelock Street Hanover, New Hampshire 03755 U.S.A. Since 1925 the Institute of

1/05/2018 Marradi (Fi) - UISP

1 05 2018 Marradi Fi UISP

52 177 canti erik asd black road elite sport 13 01:41:33.60 14,77. 68. fantini massimiliano; team k-one asd m4; 11 01:41:40.90

BOOKLET 2 Mathematics Subtest - mtel.nesinc.com

BOOKLET 2 Mathematics Subtest mtel nesinc com

A. 1 2000 B. 1 200 C. 1 20 D. 1 2 Correct Response B: In a place value system, the value of a digit depends on the position it occupies. For the number 2010, the value of the position of each digit is shown below.

GRAVITY SEPARATOR - Cimbria

GRAVITY SEPARATOR Cimbria

CIMBRIA | SEED PROCESSING | GRAVITY SEPARATOR | 3 | GA 31 GA 110 LAB GA GA 310 MACHINE RANGE SERIES PRODUCTION Cimbria Heid produce machines in series production. During production each machine pass several specific quality controls to meet the requirements at the highest level. GA 71 GA 210. w | 4 | FEATURES The Gravity Separator should only be fed with clean products (i.e. after a Pre/Fine ...

plantilla dossier bodas paradores 2019

plantilla dossier bodas paradores 2019

Microsoft PowerPoint - plantilla_dossier_bodas_paradores_2019 Author: eventos.stoestevo Created Date: 12/20/2018 2:21:49 PM ...

2012 Chevrolet Colorado Owner Manual M - General Motors

2012 Chevrolet Colorado Owner Manual M General Motors

Chevrolet Colorado Owner Manual - 2012 Black plate (4,1) iv Introduction Using this Manual To quickly locate information about the vehicle, use the Index in the back of the manual. It is an alphabetical list of what is in the manual and the page number where it can be found. Danger, Warnings, and Cautions Warning messages found on vehicle

100 Kata Motivasi - BUKU MERUPAKAN GUDANGNYA ILMU MEMBACA ...

100 Kata Motivasi BUKU MERUPAKAN GUDANGNYA ILMU MEMBACA

Buku ini disusun dengan merujuk beberapa buku?buku/referensi lainnya yang ... 100 kata mutiara, bijak dan inspiring ini dikompilasi dari berbagai ...