Search

External representation is the use of the physical world for cognitive ends, the enlargement of the mechanisms of representation to include the action-perception cycle. It has recently been observed that such representation is pervasive in human activity in both pragmatic and more abstract tasks. It is argued here that by forcing an artificial learning system to off-load all of its representation onto a (simulated) external world, we may obtain a model that is biased in a very natural way to represent functional relations in ways similar to those used by people. After learning a function from examples, such a model should therefore generalize to unseen instances in ways that we would consider correct. These ideas are tested by developing two machine learning systems, in which representation relies on the sensorimotor control of simulated robotic agents. These systems are able to represent a variety of functional relations by means of their action and perception, and they learn to spontaneously do so from examples. Moreover, they generalize extremely well to unseen problems even after a small number of examples, including on functions such as n-parity that are notoriously difficult to generalize for machine learning algorithms. It is argued that despite these systems' simplicity, the external representations that they evolve are similar to those used by people on similar tasks. [ABSTRACT FROM AUTHOR]

Whatever approach regarding internal representations, the idea was always that of a mapping of an outside world, more or less successfully performed by our cognitive apparatus. In the following we want to develop a principally different approach where representation is no more considered any kind of mapping of a predefined external reality, but simply as stabilities in the coupling between organisms and their local environment. Since this kind of representation evolves in the dynamic interaction with the environment it is a fundamentally active process of construction and not a passive mapping. Neuroscientific and psychological evidence favor this concept over old concepts on purely bottom-up mapping of the environment. [ABSTRACT FROM AUTHOR]

Consciousness has many different connotations, some of which are amenable to treatment within neurobiological description systems while others are not. It is possible to define in neurobiological terms the brain states associated with consciousness. It is also conceivable that neurobiology will ultimately provide a reductionistic explanation of mechanisms which enables the brain (1) to construct from the sparse and diverse signals of its sensors coherent models of its environment, including the organism itself, and to generate abstract descriptions, (2) to iterate the same strategy to monitor its own states, thereby generating meta descriptions, (3) to weigh the combined results of these analyses in order to reach decisions and to generate adapted behavioural responses, and (4) to communicate through various channels at different levels of abstraction the results of these cognitive processes to other brains. Since it became clear that the concept of the Cartesian theatre is untenable, that processes in the brain are highly distributed and that there is no single convergence center where the results of the numerous parallel operations are brought together for joint interpretation and decision making, analysis of processes that are in principle amenable to neurobiological explanation is in itself a major challenge. Problems of different nature are encountered if one attempts a reductionistic explanation of the subjective connotations of consciousness associated with self-awareness, attributes that are assessed by introspection and by extrapolation from one's own awareness of mental states to that of others. I shall defend the position that these aspects of consciousness cannot be understood as emergent properties of individual brains alone but come into existence only through communication among brains whose cognitive abilities must be sufficiently developed to generate a theory of mind, i.e. to generate models of presumed states of the respective other brain. Thus, self-awareness and the ability to experience sensations as subjective reality would have to be considered as cultural achievements or, and this is equivalent, as the result of experiencing dialogues of the kind: "I know that you know that I know". Hence, these aspects of consciousness come into existence only through a social learning process in which brains experience a class of mental phenomena that emerge only from mutual reflection. These phenomena are ontologically different from those qualified above as amenable to direct neurobiological investigation because unlike the latter they are the result of a dialogue among brains that got increasingly refined during cultural evolution. This is probably the reason why these phenomena appear as not deducible from analysis of individual brains in the same way as one can analyse the neuronal substrate of pattern recognition, memory or motor control. My proposal is that the phenomena that give rise to the so called "hard problems" in the philosophy of consciousness, problems resulting from the ability to be aware of one's own brain functions can be understood as emergent properties of brains without having to take a dualistic position; however, because these phenomena have a social or cultural origin and hence both a historical and interpersonal dimension, they cannot be understood as an emergent property of an isolated brain alone and hence transcend the reach of conventional neurobiological approaches. [ABSTRACT FROM AUTHOR]

This paper identifies the theory of the General Comunication System with its view of information as stable, storable and transferrable entities and the machine translation inspired search for the language of thought as the main trends in past cognitive scientifically relevant research in translation. The present reorientation with its remarkably different set of research questions is introduced. The contextuality of cognition, the twofold process of interpretation by meaning and sense construction and the development of cultural and expert competences are depicted as the foundation of a cognitive scientifically coherent picture on translation. Finally, it is shown how the introduction of the concept of compatibility revolutionizes the epistemological foundation of translation: the theory of linguistic reproduction is abandoned, and translation is seen as active construction of new meanings and situations. [ABSTRACT FROM AUTHOR]

Studying the underlying neuronal substrate of language processing with electrophysiological techniques, we could provide evidence for a physiological reality of linguistic categories. The processing of abstract concepts (Nouns) activates a less complicated network in the brain, thus showing different functional representation than concrete concepts. As we have demonstrated earlier this is only true with respect to those frequency bands of the EEG, which reflect higher cognitive processes. In other frequency bands the processing of both abstract and concrete nouns activates similar networks. This can be explained by the fact that mere acoustical and visual word perception does not differ between concrete and abstract nouns. The comprehension of complex sentences requires analysis of whole scenarios depending on phonological, syntactic and semantic entities across time. This can also be monitored by EEG-analysis. Even working memory demands can be observed in frontal cortical regions during sentence processing. EEG coherence analysis seems to be an important tool for the investigation of the physiology of language representation and supports theoretical findings in linguistics and philosophy of language. [ABSTRACT FROM AUTHOR]

Adaptive behaviour research has been heralded as a new paradigm in the cognitive sciences, whose main virtue is in eschewing the notion of representation. But the assumption that this new paradigm is immune from Heideggerian critiques which have been seen to be relevant to traditional representational AI is false. Behaviour-Based ideas are in accordance with an aspect of the critique but omit to confront the nature of language. The problem is the elusiveness of the traditionary aspect of language which ultimately renders any computational cognitive science impotent. It in not only the traditional notion of representation which needs despatching, but also a related notion of reality as purely physical reality. This paper identifies common structures in contemporary thought which serve to raise the traditionary aspect of language to the fore. [ABSTRACT FROM AUTHOR]

Several areas of research within social psychology depend on assumptions about underlying cognitive processes. These assumptions have not typically been examined from the perspective of cognitive science. Two theoretical constructs, the self and attitudes, are specifically considered. In contrast to earlier models of symbolic representation and rationalistic decision-making, a cognitive science perspective allows these to be viewed as habits or patterns of thought and behaviour that depend on learnt associations between multiple features of people's experience. The observation that individuals can display different characteristics and attitudes in different stituations suggests the importance of context-dependent learning, resulting in multiple attractors. Such findings are located within the intellectual tradition of empiricism, and argue for the dependence of social cognitions on social reality. [ABSTRACT FROM AUTHOR]

The article focuses on four questions for a theory of semantics: the ontological, semantic, learnability and communicative questions. It is shown how different realist, cognitivist and constructivist semantic theories answer the questions. [ABSTRACT FROM AUTHOR]

We explore a number of pragmatic principles of communication in a series of computer simulations. These principles characterize both the environment and the behavior of the interacting agents. We investigate how a common language can emerge, and when it will be useful to communicate rather than to try the task without communication. When we include the cost of communicating, it becomes favorable to communicate only when expectations are not met. [ABSTRACT FROM AUTHOR]

Symbol grounding has been put forward as a candidate solution to the problem of associating intrinsic meaning obtained from sensorimotor data, to the arbitrary symbols that are so common in the cognitive domain. This paper focuses on the notion of how intrinsic meaning may be acquired and represented within an artificial cognitive system and considers how this task is influenced by varying the initial representations of sensory data and also the internal mechanics of the learning mechanism employed. The task of ‘colour naming' which involves grounding sensory representations of words (linguistic input) using sensory representations of colour (visual input), is described and it is shown, using a number of connectionist models, how the initial representation of colour effects the acquisition of the ‘colour naming' ability. Results from the use of both psychologically based and other purely arbitrary representations lead to the conclusion that the representations used within cognitive architectures should possess structure that is lawfully related to the objects being represented, or alternatively representation must pay close attention to reality. [ABSTRACT FROM AUTHOR]

The grounding problem is, generally speaking, the problem of how to embed an artificial agent into its environment such that its behaviour, as well as the mechanisms, representations, etc. underlying it, can be intrinsic and meaningful to the agent itself, rather than dependent on an external designer or observer. This paper briefly reviews Searle's and Harnad's analyses of the grounding problem, and then evaluates cognitivist and enactive approaches to overcoming it. It is argued that, although these two categories of approaches differ in their nature and the problems they have to face, both, so far, fall short of solving the grounding problem for similar reasons. Further it is concluded that the reason the problem is still somewhat underestimated lies in the fact that modern situated and embodied AI, despite its emphasis of agent-environment interaction, still fails to fully acknowledge the historically rooted integrated nature of living organisms and their environmental embedding. [ABSTRACT FROM AUTHOR]

The ‘information processing metaphor' as the traditional approach to visual perception suffers from a number of conceptual problems which result from the existence of purely sensory representations and the separation between perception and the generation of behaviour. Based on this criticism and a discussion of alternative approaches, a contrary theory of perception is presented. This approach of ‘perception through anticipation' tries to avoid the problems of the information processing metaphor by replacing sensory with sensorimotor representations, and by considering perception to be an active and generative process rather than a pure projection. Perception of space and shape is assumed to be a process of anticipating the sensory consequences of actions; appropriate actions are selected within the same neural process. [ABSTRACT FROM AUTHOR]

Cognitive structures are organized systems of information that embody the knowledge used to construct an individual's reality. Having phylogenetic and ontogenetic determinants, they reside latently in the connectional organization of the cerebral cortex, both within and between areas. Different cortical areas, containing separate classes of knowledge in their local associative memories, operate in conjunction with one another to instantiate cognitive structures in perceptuo-motor behavior. As multiple areas recursively interact in largescale networks, their mutual constraint leads to the emergence of coordinated large-scale activity patterns. These patterns constitute a consistent construction of reality that fits the constraints imposed by the structures of the internal and external environments. This construction is the dynamic manifestation of cognitive structures in the cerebral cortex. [ABSTRACT FROM AUTHOR]

A model of human cognition is presented in which peripheral sensory images are selectively enhanced by corticofugal pathways. The same processes also generate quasi-sensory mental images even in the absence of sensory stimulation. This inversion of forward sensory processing can come about through hillclimbing algorithms or a type of Hebbian synaptic facilitation I call inverse learning. The neural mechanisms that enable us to simulate actions and evaluate alternative scenarios, are described as internal sketchpads. It is further argued that early artistic expression and language origin can be understood as natural extensions of cognitive processes in which central symbolic neural activity is projected outward to generate images and symbols. [ABSTRACT FROM AUTHOR]

We discuss a series of problems facing referential theories of mental representations and we propose a scheme based on neurophysiological principles that avoids previous limitations. According to this scheme, mental representations are brain traces linked to stimuli in the environment, via a causal but probabilistic process of categorical perception, and fluctuations in activity reflect fluctuations in the confidence-level of perceptual and cognitive hypotheses. The scheme provides an explanation for cases of misrepresentation and is consistent with the abundance of recurrent connections in the cortex, which play an important role in mediating a process of interpretation and of binding of relational properties via temporal synchronisation. [ABSTRACT FROM AUTHOR]

In this paper, I discuss the principle that bridges neural firing and perception, questioning some fundamental aspects of the neural correlates of conscious perception/cognition, which are central to the new trends in cognitive science. The assumption is that in order to understand perception, the state of neural firing in the brain is necessary and sufficient (the neuron doctrine in perception). The concept of response selectivity, currently the de facto central dogma in explaining the relation between neural firing and the mind, is found to be incompatible with the neuron doctrine. I put forward two new concepts, Mach's principle in perception and the principle of interaction simultaneity. The latter is concerned with the origin of the subjective time. The approach outlined in this paper has elements common with the constructivist approch in cognitive science. [ABSTRACT FROM AUTHOR]

The recent interest in Dynamical Systems Theory (DST) is part of a growing tendency within cognitive science to replace the structure-sensitive processing of representations by associative processes of a neurophysiological kind. More specifically, neurodynamical work on categorization can be located within an associationistic tradition because of its attempt to explain behavior on the basis of correlated processes in the brain. Although associationism is a tradition not favored by traditional cognitive science, a recognition of DST's associationistic tendency seems called for. This does not imply a weakening of DST's relevance for the study of cognition, but rather a strengthening of classic associationistic ideas. [ABSTRACT FROM AUTHOR]

A definition of the scope of a scientific theory of human cognition is proposed in which for any psychological state a corresponding physiological state can be identified, and causal relationships between psychological states have corresponding causal relationships between physiological states. The vital role of a simple functional architecture in functionally complex commercial electronic systems is described, and it is argued that selection pressures have resulted in simple functional architectures in biological brains. However, the functional architecture is qualitatively different from the architectures in electronic systems. Electronic systems have the instruction architecture in which functional components exchange unambiguous information. The only alternative is the recommendation architecture in which functional components exchange ambiguous information. Systems with the recommendation architecture demonstrate phenomena with a striking similarity to psychological experiences such as learning, object recognition, associative memory, dream sleep without recall, constant sensory independent sequences of mental images, and individual differences between the experience of the same conditions. All of these phenomena can be described in a consistent fashion on both psychological and physiological levels. It is therefore argued that biological brains have the recommendation architecture, and that this architecture makes possible a scientific theory of cognition. The nature of representation in such an architecture is discussed. [ABSTRACT FROM AUTHOR]

I argue against a widespread assumption of many current models of cognition—namely, that the process of creating representations of reality can be separated from the process of manipulating these representations. I hope to show that any attempt to isolate these two processes will inevitably lead to programs that are either basically guaranteed to succeed ahead of time due to the (usually carefully hand-crafted) representations given to the program or that that would experience combinatorial explosion if they were scaled up. I suggest that the way out of this dilemma is a process of incremental representational refinement achieved by means of a continual interaction between the representation of the situation at hand and the processing that will make use of that representation. [ABSTRACT FROM AUTHOR]

Analysis of the relation between representation and reality depends on the kind of representation under investigation. Computing systems taken by themselves can always be interpreted as computing more than just one function. The most popular response to the problem of multiple inter-pretability attempts to anchor a system's content in its environment. While this approach would reinforce a number of pretheoretic intuitions about the nature of mental representation, it presupposes that the computed function is defined over a system's environment. The main tenant of this paper is to argue that this assumption is false. Instead, the computed function may turn out to be defined over states of the computing system itself. [ABSTRACT FROM AUTHOR]

Representations, it is argued, cannot have effects as representations; yet, even so, they are essential to some explanations in cognitive science. These apparently incompatible claims can be reconciled if we distinguish between explaining cognitive abilities and explaining cognition. Certain cognitive abilities can be seen to require representations for their explanation; but explanations of this type cannot explain how the required representations come to be available, or come to be appropriately connected. A clear understanding of the distinction between explaining cognitive abilities and explaining cognition suggests a certain approach to the latter project, and throws light on some recent controversies. [ABSTRACT FROM AUTHOR]

In addressing the question "Do representations need reality?", this paper attempts to show that a principled understanding of representations requires that they have objective, systematic content. It is claimed that there is an interesting form of nonconceptual, intentionality which is processed by non-systematic connectionist networks and has its correctness conditions provided by a modest biosemantics; but this type of content is not properly representational. Finally, I consider the consequences that such a verdict has on eliminativist views that look to connectionism as a means of radically reconceiving our understanding of cognition. [ABSTRACT FROM AUTHOR]

This article discusses the possible representational nature of two brain cognitive functions: perceptual and executive. Assuming the Newellian definition of representational processes as those that establish an isomorphic relation between two structures, I claim that perceptual processes generate only a partial correspondence (between stimuli properties and brain states) and therefore should not be properly conceived as representational. On the other hand, executive processes encompass the combination of copies (i.e., representations) of perceptual patterns, generating new patterns that subserve behavior. In summary, I criticize the notion of perceptual representations, and propose that brain representational processes are related to executive functions, having a pragmatic dimension. [ABSTRACT FROM AUTHOR]

Anti-representationalism is in the air. In the last few years, many philosophers and cognitive scientists have considered or even embraced the claim that cognition is not representational, often without giving explicit consideration to what exactly this means. The point of this essay is to try to make some sense of claims that cognitive science can do without representations by proposing a taxonomy for them. In what follows, I will make a distinction between two different varieties of anti-representationalism. And, with this distinction in hand, I will consider some actual scientific work that has led to claims that cognitive science can do, at least in part, without representations. [ABSTRACT FROM AUTHOR]

The goal of this paper is to argue that the ontological status of representations can only be evaluated within a theory. In other words, what counts as representation, or whether a certain representation is better than another one, depends solely on the (level of) description of the phenomenon under scrutiny. It is shown how "representation", being a semantic notion, can be defined in terms of the notion "meaning". For cognitive science, in particular, it follows that representations, functioning as mere descriptive devices to facilitate one's goal of explaining and modeling brain/thought processes, cannot in and by themselves give rise to ontological or epistemological claims. [ABSTRACT FROM AUTHOR]

In this paper I demonstrate that connectionism is, or can be, largely in line with most recent trends in cognitive science. The core of my argument is a distinction of several types or uses of representation in cognitive science. I demonstrate how connectionism helps in abandoning one of these types—the notion that representation is a mirror of an objectively existent world—while maintaining the important other two types—causal correlates of physical states, and internal mental states standing in for past perceptions. With this distinction the question "Does representation need reality?" can be answered. I further depict a connectionist route to embodied and situated cognitive models, as they are put forward by recent cognitive theories. After a short discussion of connectionism's role for dynamicist theories of cognition, I conclude that much of current connectionist research is highly relevant to modern cognitive science, even if the models are apparently remote from truly embodied or situated ones. [ABSTRACT FROM AUTHOR]

The first papers address the epistemological status of representation and possible categorizations. These rather general considerations about representation are followed by contributions that focus on computational approaches, especially on how meaning can be simulated in artificial devices, and how meaning can be grounded. Solutions are offered on various levels, among them symbol grounding and system-theoretical considerations to representation. Various disciplines independently arrive at emphasizing the importance of actions for representations and the necessity to close the sensory-motor loop; examples from the perspective of computational approaches ("embodiment" into an environment), the neuroscientific perspective ("top-down" processes), the psychological perspective and others are presented. Several papers address the role of social interactions—in particular language—as a means of stabilizing systems, and in the emergence of meaning. [ABSTRACT FROM AUTHOR]

This paper discusses the notion of representation and outlines the ideas and questions which led to the organization of this volume. We argue for a distinction between the classical view of referential representation, and the alternative concept of system-relative representation. The latter refers to situated cognitive processes whose dynamics are merely modulated by their environment rather than being instructed and determined by it. [ABSTRACT FROM AUTHOR]