Cognitive Psychology - Essay Example

Cognitive Psychology Introduction: Cognitive psychology considers that the human mind functions in the forms of perception, memory, attention, language, thinking, and decision making. Although debatable, this envisions the mind as a conglomeration of neurons that are connected to each other much like the computer circuitry that can process information with similar input, storage, and retrieval functions (Edwards, 1997, 73-85). Cognitive states and processes were considered by the theorists as theoretical entities that cause behavior. These concepts distinguish cognitive psychology in that in contrast to behavior or neural connections, cognitive states and processes cannot be observed directly. Thus cognitive states and processes can be defined as theoretical entities that provide a functional characterization of operations in the central nervous system (Estes, 1994, 19-29). Therefore hypothesis about cognitive processes can be evaluated only by testing their predictions regarding the effects of various environmental conditions. Cognition can thus be termed as acts of coming to know and states of knowing, as well as for states of wanting and for decisions insofar as they are guided by knowledge. Therefore, the logical foundation of cognition could guide the readers to either of the many approaches to explain the process of cognition (Feldman & Ballard, 1982, 205-254). Cognitive psychologists have adopted either the psychometric approach or the experimental one. The psychometric approach is to devise tests of mental traits or abilities - comprehension, sensitivity to others, intelligence and so on - and to study correlations among test scores. The experimental approach is more guided by theory. The main strategy is to devise experiments that test theories about mental operations. Today, the experiment is sometimes accompanied by computer modeling since computer modeling is sufficiently vigorous and autonomous to deserve a name of its own: the modeling strategy. Experimental cognitive psychology is built on the concept that cognitive processes can be tested using scientific methods because they can be inferred from behavior obtained under controlled conditions. The crux of this interpretation of science, particularly the development of hypotheses about and models of hypothetical generative mechanisms of cognition is that, although unobserved by investigators, such mechanisms are necessary for the production of the observed phenomena (Fodor & Pylyshyn, 1988, 3-71). To find out the logical basis of these hypotheses about unobserved generative mechanisms, scientists have taken the help of the physical sciences. In physical sciences, the hypotheses are not the result of blind guesswork or the unfettered imagination. They are created by the invention of models or hypothetical representations of what such mechanisms may plausibly be in reality. The invention of adequate and plausible models is constrained by the requirement that the nature of what is proposed should conform to the basic type hierarchy that expresses the beliefs people have about the nature of the world. Hypotheses about cognition can be evaluated only by testing their predictions regarding the effects of various environmental manipulations on behavior (McCloskey, 1991, 387-395). These theoretical entities are said to provide a functional characterization of the central nervous system. This is built on the assumption that the same cognitive process could be implemented or instantiated in a variety of different neuroanatomical structures or neurophysiological processes. In other words, this characterization of cognition is the materialist, but it does not assume a simple one-to-one mapping between cognitive and neural states and processes. Thus the definition of the cognitive processes can further be modified by a process that may or may not be objects of conscious awareness, and it receives inputs from other cognitive states and processes and from perception, and have outputs to other cognitive states and processes and to behavior. In a nutshell, the theories of cognitive science use computational or computer models to explain and understand cognition. It allows specification of a theory to predict behavior. Like computational models, flow charts are used to construct theories and to provide a plan for which input can be investigated as well as the nature of the storage and decision process can be examined. All the collected information can then be utilized to devise a computer program. Although this can be argued that the interaction of psychological theory with computer programming is difficult and the relationship between the performance of a program and human participation may be different, it indeed provides the closed possible alternative available today to study the psychological processes involved in cognition (Pylyshyn, 1973, 26-83). With the computer-modeling component of cognitive science, the scientists take serious responsibility to go beyond standard experimental strategies in offering a unified theory of a significant portion of cognition. While dealing with this, it must be taken into account that computers, however, do not interpret their symbols into a reality external to the computer. While the computer scientists attempt to model cognitive states, it has been criticised that they require a stance that may be termed as methodological solipsism, indicating psychology from the skin in, where the world disappears, so to say, from the reference, although this is not to deny role for computers in modelling different areas of psychology, such as perception; nor does it deny the mind's computational abilities (Pylyshyn, 1980, 111-169). The connectionist network theory suggests that networks, much identical to neural networks, have layered structures that are interconnected. Concepts are stored within the network and are activated by patterns that are generated by the simple association between inputs and outputs. These have a feedback mechanism of back propagation that uses a comparison between actual and correct responses. Ideally, such machines are modeled in such a way that it not only would output representations of what human beings would interpret as correct answers to the questions that had been input but is expected to also work in ways similar to the way a human being accomplished a problem-solving task, be it practical or cognitive. Millar and co-workers in their very subtle and forward-looking publication of the 1960s, Plans and the Structure of Behaviour, had set out a program for a cognitive science (Miller et al., 1967: 27–36) where they used sophisticated ideas about how to create a model of the relevant abstract cognitive process and a material mechanism that might be able to perform it. Although this theory was nullifying the metaphysics of behaviorism, this actually was founded on the paradigm of Thorndike’s law of effect, that the probability of the emission of a particular behavior was a function of the effect of emitting the same behavior in the past. This law was based on the metaphysical assumption that there were neural structures and reflex arcs underlying the observed correlations. It was assumed that the neural structures that underlay behavioral correlations were either inborn or established as reflex arcs in the nervous system by conditioning, a conditioning made possible by the phenomenon described in the law of effect. But, this theory did not take into account the feedback loops that were a really activated on all occasions of a cognitive process. When this concept was incorporated, a novel imaginary mechanism resulted that had the potential to self-adjust itself to the reality in the environment, and this was called TOTE unit. The overall pattern of this unit is Test/Operate/Test/Exit, or TOTE (Miller et al., 1967: 39-48). One can see that it also incorporates norms of correctness. As can be envisaged, this paradigm can be interpreted in at least three ways. First, it could simulate the real mechanism, where the appropriate scheme of energy flow through the system can be represented much like the neural circuitry or computer circuits. It could also be a schematic diagram of the neural structures involved in a particular act of cognition. Second, more abstractly, this could represent the information system through which cognitive messages are transmitted along channels (Button et al., 1995, 33-47). This information would suggestively take the form of representations of correlations in the appropriate medium, such as incongruity/operate and congruity/exit. Third, a single TOTE unit or a hierarchy of many such units could assume control over the cognitive process that might affect a certain sequence of actions which might involve after exit from a TOTE unit, the transfer of control to another unit that automatically starts up after one (Miller et al., 1967: 49-91). This gives a startling analogy to computational models, where a program is known to be a sequence of instructions, each of which executes an operation in turn. However, a system such as a TOTE hierarchy is a sequence of dedicated operations, each of which executes a specific operation, in turn, dictated by logic. Instead of a sequence of instructions for one mechanism to do different things, a system is a sequence of different mechanisms, each of which does something different at a given point in time. Running a computational program on the data in a single central processing unit and activating a system through which a stream of data flows are functionally equivalent to this model. A program operating sequentially on the initial state of one device would perform the same task as a system in which each instruction was realized in its own dedicated module, activated sequentially. Programs include grouped instructions, while systems include groups of modules, and in each case, there is a basic level of ‘unit' computational instructions and a basic level of unit modules (Boden, 1988, 7-51). To find out evidence for this analogy, the most pertinent question is, does this really simulate the human brain and its circuit mechanism. The anatomy of the human brain and neural responses has been well elucidated in the present days. The concept of cell assembly is a widely agreed upon theory on neural representation of concepts. This is circuitry of a large number of neurons that have strong connections. It is interesting to note that this circuit is activated by external input that causes some of the neurons to generate an action potential. The external input can be sourced from sensory stimuli or another cell assembly previously existing or acquired (Anderson, 1977, 27-90). With a sufficiently strong external input, the neurons in the assembly would excite similar events in the other neurons in the assembly leading to ignition of the assembly. Concepts are stored in the brain by these assemblies that are made up of tightly connected neurons that may be activated by a stimulus. The connectionist model attempts to incorporate important features of the brain architecture, it does not offer a detailed account of the neural processes and instead proposes a model of the quality of processing involved in cognition. Thus, this suggests a deep idea of thinking as computation. According to this principle, any cognitive process can be represented by a computable function (Bechtel and Abrahamsen, 1991, 12-21). The result of computation using such a function represents the outcome of the cognitive process the function represents. This type of modeling also presupposes that there is or will be some hardware with which these computations can be performed in vitro. This hardware can be likened to the human brain and central nervous system in which these cognitive processes will take place. It is at this point that the fundamental question arises whether any inorganic machine and the way it runs would be an adequate and appropriate model of the brain and its cognitive functions (Coltheart, 1987, 1-25). It may be impossible to use human-like learning methods for connectionist systems or for any computer-based intelligent systems. Computers are literally capable keeping into memory the contents of large text files or entire dictionaries, while they lack perceptual and reasoning capabilities in an entirely new situation. The issue, then, is how connectionist models relate to natural and artificial intelligence. It is a fact that human brain, although structurally similar, does not compute like a conventional computer (Bruner, 1991, 1-21). The neural computing elements in the human brain, although slower form similar but very much more complex parallel connections form a structure that is dramatically different from a predominantly serial computer. Current research in neuroscience is concerned with tracing out these connections and interconnections in the human nervous system in order to discover how they transfer information. Thus extending these ideas into the connectionism, one might state that connectionist models are large networks of simple parallel computing, each of which would carry a numerical activation value that it computes from the simple numerical formula out of the neighboring elements in the network. Like computing, these network elements or units influence each other’s values through connections that carry a numerical strength or weight (Berwick, 1983, 383-416). There are a number of different ways that a connectionist system can be set up depending on the way a variety of parameters set. The units those were spoken about may be set on the basis of permissible activation levels of the units, either in a small number of discrete states such as on or off or may be allowed to vary over a specified range such as 0 and 1, much like the binomial algorithm of computer programming lexicon. A decay function can be included so that without new activation, the activation of a unit may be dropped, and a threshold can be set such that unless the input exceeds a certain quantity, the unit is not activated. The next parameter is that of output that could be proportional to its own activation or it could be governed by the threshold. There are a variety of learning rules that can govern the functioning of this connection pathway (Estes, 1988, 196-212). Two of the most widely employed are Hebbian rule that strengthens or weakens a pathway depending on whether the two units are alike or different in their levels of activation and the delta rule that decreases or increases the strength of each input pathway in order to enhance the fit between the produced and the target values of the unit. The problem arises when the detailed models of cognitive science are built to suit a computational or dynamic framework. In attempting to describe or understand such systems, it is a standard practice to provide a concrete model. The connectionists, however, add another aspect to this modeling, the concept of simulation (Fennel & Lesser, 1977, 98-111). The connectionist model in the cognitive science is made up of units that are connected to each other, and they are capable to adjust their activation as a function of their total input given by the summated weighted activation of the other units. Therefore, the cognitive connectionist modeling is a high-dimensional, homogeneous, and neural dynamic system that by nature is overwhelmingly abstract in the sense that they are simulated by the computational systems. Computational systems have been characterized as those whose states consist at least in part of the configuration of symbols (Newell & Simon, 1972, 1-23). Therefore, these symbols can represent or stand for some other aspect of the world, both inside or outside the system. These symbols in the computational systems can stand for numerical states for some dynamic systems that are abstract actually, or even they may represent the changing quantities of some concrete dynamical system. This may be stated in a different form, the computational system can simulate the dynamical system (Cottrell & Small, 1983, 89-120). In reality, cognitive processes are based on representations of some kind in the relevant domain that includes representations of the current state of the domain and representations of the long-term knowledge. The most powerful forms of representation are generally symbolic, and so it can be derived that computational models in connectionist theories can provide a natural medium within which representational concepts can be implemented (Rosch, 1973, 328-350). Cognitive processes are neural computations in a dynamic form where the representations are manipulated, and transformation of one representational structure into another occurs through discrete sequential steps. Cognitive processes always begin with an input, are processed through internal manipulation, and result in an output. When applied to the cognitive science, the inputs are perceptual representations of the state of the environment, and the outputs are action representations. When this cycle comes into play in computational models, this ignores the issue of real time, and that cannot be done in the cognitive processes. Like a computer program and its implementation, the cognitive process can be broken down into a number of relatively independent modules with specific tasks assigned to them. These modules then interact by passing representations (Searle, 1980, 417-424). Cognitive connectionist modeling assumes that many modules can be simultaneously active. There is a startling similarity with present day computational models where this has been a natural mode of computation for widely interconnected computer networks of active elements. The generalization of these ideas to the connectionist view of the brain and behavior is that all important encodings in the brain are represented in terms of relative strengths in the synaptic connections. Connectionism can explain this by assuming that individual neurons do not transmit large amount of symbolic or representative information, instead, they compete by being appropriately connected to a large number of similar units, and the prevalent and conventional computer model fails to incorporate this in the present understanding of cognitive psychology (Smolensky, 1988, 1-74). Conclusion: However, this realization is important in that connectionist theories of cognitive psychology may with adequate research come out with a newer, modified, and more sophisticated model that explains all or the computer scientist may create a developed computer that can have a cognitive psychology of its own. There are, however, certain troubling questions that need to be answered before one venture into this area. Human cognition involves the management of meaningful signs according to standards of correctness (Massaro, 1988, 213-234). In developing a computer model according to connectionist theory, there is the probability that one might lose the two main features of human cognition, intentionality, the meaningfulness of signs, and normativity, conformity to standards. These properties are not exhaustively describable in terms of the material properties of the sign as a physical thing or event. References Anderson, J.A. (1977). Neural models with cognitive implications, In D. LaBerge & S. Samuels (Eds.), Basic processes in reading: Perception and comprehension (pp. 27-90). Hillsdale, NJ: Erlbaum. Bechtel, W. and Abrahamsen, A. (1991) Connectionism and the Mind, Oxford: Blackwell. 12-21. Berwick, R. C. (1983). Transformational grammar and artificial intelligence: A contemporary view. Cognition and Brain Theory, 6. 383-416. Boden, M.A. (1988). Artificial Intelligence in Psychology, Cambridge MA: MIT Press. 7-51. Bruner, J. S. (1991) ‘The narrative construction of reality’, Critical Inquiry autumn: 1–21. Button, G., Coulter, J., Lee, J.R.E. and Sharrock, W. (1995) Computers, Minds and Conduct, Cambridge: Polity Press. 33-47. Coltheart, M. (1987). Functional architecture of the language-processing system. In M. Coltheart, G. Sartori, & R. Job (Eds.), The cognitive neuropsychology of language (pp. 1-25). London: Erlbaum. Cottrell, G. W., & Small, S. L. (1983). A connectionist scheme for modeling word sense disambiguation. Cognition and Brain Theory. 6. 89-120. Edwards, D. (1997) Discourse and Cognition, London: Sage. 73-85. Estes, W.K. (1988). Toward a framework for combining connectionist and symbol processing models. Journal of Memory and Language. 27. 196-212. Estes, W.K. (1994) Classification, and Cognition, Oxford: Clarendon Press. 19-29. Feldman, J. A., & Ballard, D. H. (1982). Connectionist models and their properties. Cognitive Science. 6. 205-254. Fennel. R. D.. & Lesser, V. R. (1977). Parallelism in Al problem-solving: A case study of HEARSAY I I. IEEE Transaction on Computers C-26, 98-111. Fodor, J.A., & Pylyshyn, Z.W. (1988). Connectionism and cognitive architecture: A critical analysis. Cognition. 28. 3-71. Massaro, D.W. (1988). Some criticisms of connectionist models of human performance. Journal of Learning and Memory, 27, 213-234. McCloskey, M. (1991). Networks and theories: The place of connectionism in cognitive science. Psychological Science. 2. 387-395. Miller, G.A., Galanter G. and Pribram, K.H. (1967) Plans and the Structure of Behavior, New York: Holt Rinehart & Winston. 1-91. Newell, A.. & Simon, H.A. (1972). Human problem-solving. Englewood Cliffs, NJ: Prentice-Hall. 1-23. Pylyshyn, Z.W. (1973) Computation, and Cognition: Towards a Foundation for Cognitive Science, Cambridge MA: MIT Press. 26-83. Pylyshyn, Z. W. (1980). Computation and cognition: Issues in the foundations of cognitive science. The Behavioral and Brain Sciences. 3. 111 - 169. Rosch, E. (1973) ‘Natural categories’, Cognitive Psychology 4: 328–50. Searle, J.R. (1980) ‘Minds, brains and programs’, Behavioral and Brain Sciences 3:417–424. Smolensky, P. (1988). On the proper treatment of connectionism. Behavioral and Brain Sciences, II, 1-74. Read More