DIVERSITY IN COGNITIVE MODELS
Andrew Basden, Maria Kutar
Informatics Research Institute at Salford, University of Salford, UK.
This final chapter looks back and forward. It looks back over the other chapters, to trace a 'thread' that runs through the work as a whole, which can suggest new and interesting avenues for future work. A very obvious theme argued in several chapters is that cognition is developed by interaction with the environment, especially of an embodied kind. Rather taken for granted, but no less important, is the diversity of these (embodied) interactions between the agent and its domain of application. It is diversity as such which this chapter traces as the thread that runs through the entire work, to point a way forward.
Diversity may be understood by reference to Dooyeweerd's philosophical exploration of distinct spheres of meaning and law which, as aspects, enable us to differentiate types of interaction and embodiment. Plotting the chapters by these aspects reveals interesting patterns, which allow a number of suggestions to be made for future research, including focusing on under-represented aspects, providing agents with good quality meta-rules that express the distinct rationality of each aspect, and contributing to meeting a number of challenges. It concludes by suggesting that, in addition to contributions this work makes to the field of cognitive systems and the paradigm of embodied interaction, it can also contribute an understanding of diversity itself.
The chapters in this book exhibit a very interesting diversity. How can the reader benefit from this diversity? The reader could treat it as a tasty buffet, taking away what they like, and leaving the rest for others, but that does not do justice to the coherence of this book. The title of this work is Advances in Cognitive Systems. As well as each chapter making a separate little advance, we want the whole work to be an advance and we want to understand what that advance is and how all its chapters contribute to that advance.
In this chapter the thread will be traced that runs through all the others and stitches them together, which makes what is, perhaps, the greatest contribution of this work. The chapters fall into the general area of cognition - but that is not the thread we seek; cognition is the fabric as a whole. Some of the reported research uses neural nets, others use semantic nets - but the thread of neural versus semantic has rather worn thin from exhaustive debate over the last 30 years. The thread this chapter traces is, rather, that of diversity itself.
Rolf Pfeifer, who contributes to this volume, has published the following diagram in Pfeifer & Bongard [2007; used with permission] (it is in Pfeifer & Gomez' chapter in this volume):
Figure 1. Two ways of approaching artificial intelligence.
This contrasts what the authors call traditional and "modern, embodied" ways of trying to understand (artificial) intelligence. In the traditional approach the focus is on the brain and central processing, while in the modern approach, it is on engagement and interaction with the environment. Cognition, they suggest, emerges from this interaction. Cognition is therefore not the instantiation of some detached reasoning or pre-designed representational model on the world, but is itself engaged with that world. Its very processes and model are modified by it.
Not only does the diagram give a message of interaction, but it also speaks another message very loudly: the things that we interact with are highly diverse in a meaningful way. We see: food, painting, clothing, telephoning, muscular exercise, reading, typing, sport and leisure. Presumably whoever drew this diagram did not intend to exhaust all the types of interaction that might occur, but could, if they wished, have also included, for example, reprimanding or rewarding children, caring for others, religious worship, tending household crops or chickens, and many more, and most likely would have done so had they been from a majority-world rural culture. (These are the types of interaction that the first author found to be important during a working visit to rural Uganda.) Issues like this, which are important in such cultures, will be added to those found in Fig. 1 when it is useful to emphasise a diversity that extends beyond that of Western culture.
Three threads run through the chapters in this book. One is, as its title suggests, cognition. As already mentioned, however, this cannot be a thread of major interest here because the debate between neural and semantic net approaches to modelling been well-rehearsed over the last thirty years. Moreover, to see cognition as the main thread would be taking the traditional attitude expressed in the left-hand side of Fig. 1 and would ignore the diverse interactions depicted in the right-hand side.
The next thread is that stressed by Pfeifer & Bongard , that cognition itself is constructed by the interaction with the world. To make that point is important because it is still not universally accepted, and how it is to be actualized in information technology is still being explored. Many of the chapters contribute to that exploration.
A final chapter of a collection like this, however, should not just review the others but should, if possible, suggest new and interesting ways forward that are, perhaps unknowingly, already implied in the other chapters. This chapter does both by reference to the third thread, which is meaningful diversity. It finds that the book as a whole, and all chapters individually, contribute something to understanding the diversity of ways in which we interact meaningfully with the world.
If cognition is indeed built up by interaction with the world and if these interactions are highly diverse, as the diagram shows, then it will become necessary to understand the nature of this diversity. This is because, if the types of interaction that contribute to cognition are diverse, then the processes and rules by which cognition is built up might also be diverse - at least it is reasonable to explore whether this is so. We are not concerned with surface diversity, such as the difference between two sports like football and rugby, but with diversity of a more fundamental kind, the difference between, for example, playing sport, feeding oneself, socialising, and reprimanding or rewarding children. So what this chapter focuses on is the thread of 'meaningful diversity'.
The diversity that characterizes the chapters in this collection lies not so much in the cognitive modelling techniques or technologies employed, as in the ways in which interaction with the world is meaningful. That is, the diversity of interest here is the diversity of applications of cognitive modelling techniques. This diversity is summarised in Table 1, where chapters are identified by a number plus initial of name of first author (the number is an arbitrary identification number, which is not the chapter number).
|1-M. Montebelli et al.||Understanding bodily anticipation||How the biological affects the behavioural|
|2-M. Montesano et al.||Robots imitating humans||Affordance|
|3-D. DurÃ¡n||Modelling behaviour of infants||Wrong choices|
|4-V. Vernon et al.||-||Cognitivism v. Emergentism|
|5-C. Coutinho & Cangelosi||Effect of music on emotions||Modelling how music affects emotions|
|6-N. Nefti et al.||Numerical optimization||Psychology can improve handling of risks and constraints|
|7-R. Ruini et al.||Organism motivations||Strategic v. tactical|
|8-G. Gemrot, et al.||Creator of bots for games||Realistic behaviours in games|
|9-S. Simko & Cummins||Understanding speech production||How organic structure affects this|
|10-A. Alissandrakis et al.||Robots imitating humans||Choosing what to imitate from gestures|
|11-K. Kutar & Basden||Interacting with symbolic systems||User satisfaction, effectiveness|
|12-G. Grinsberg, et al.||Prisoner Dilemma games||How cooperation affects success|
|13-O. Oussalah & Strigini||Several||Adjudication between multiple agents|
|14-T. Tsagarakis et al.||Robot actuators||How to achieve human organic safety|
|15-L. Lovric, et al.||Understanding how investors behave||Modelling creative financial decisions|
|16-P. Pfeifer & Gomez||Sensori-motor activity of robots||How morphology can assist|
This chapter will open with a general discussion of meaningful diversity and introduce a way of getting to grips with it, drawn from the philosophy of the mid-twentieth century Dutch thinker, Herman Dooyeweerd (1894-1977). This is applied to the chapters of this collection, allowing some patterns to emerge that reveal the thread of diversity. Finally, suggestions of strategic direction for future 'advances in cognitive systems' will be briefly outlined.
2. MEANINGFUL DIVERSITY
One major integrating thread running through this work is meaningful diversity. The word 'meaningful' is used advisedly, and to refer to two things. One is that the diversity is meaningful to those interested in cognitive modelling, in that diversity has an impact on the way we model things and on the way cognition operates and develops. Modelling is not just of a passive world, but is engaged with that world, insofar as different types of things and activities being modelled are reflected not only by differences in the content of the model itself but also in the types of modelling formalisms and technologies used. This type of meaningfulness is returned to later.
The other way the diversity is meaningful is that we live in a meaningful world. The diversity of the world itself is not mere accidental variety that makes it impossible to obtain coherence, but works in such a way that the diverse ways in which things are meaningful all work with each other, support each other and refer to each other. All the activities shown in the right hand side of Fig. 1 contribute to the life and very being of that person; all the chapters in this work contribute to its meaning as a whole.
This challenges us to understand diversity. We want to be able to orient ourselves within it, not just so that we know where we are at the moment, but so that we know in which direction we should head and know what we might expect on the way that we have not yet encountered. In relation to this work, these mean that we want to know
- whether and how all the chapters relate to each other,
- how we might usefully continue research in the future, of the type encountered in this book, and
- how to cope with the next application of cognitive modelling that we come across: to what extent should we try to squeeze it into the moulds offered by the chapters of this book?
The importance of the third point becomes clear when considering two examples. Music is important in Uganda. If we wished to model the effect of music on emotions there we could apply the methods of Coutinho & Cangelosi in this volume almost directly even if the music were different.
But suppose we wished to model the rewarding and reprimanding of children in a family setting, so that in future they will learn to behave in a socially responsible way. No chapter in this collection seems to offer much help. Grinsberg, Lalev & Hristova's chapter makes a useful contribution to understanding the link between attitude and gamesmanship. It might seem relevant if we were to replace prisoners by children because it speaks of cooperation and defection, but it makes a number of assumptions that reduce its relevance. It could only relate to groups, not single children; it discusses the interaction among the children whereas reward and reprimand is an interaction between children and parents; the chapter is about immediately measurable gain whereas social responsibility is a long-term character trait; use of the chapter would presuppose that children (as prisoners) are pitted against their parents (as prison warders) rather than being in a relationship of care, nurture and respect. Though that chapter might be used to model the situation in dysfunctional families where these conditions apply, doing so would make a very doubtful contribution, not only because this would be a minority of cases, leaving the majority of cases unmodelled, but because it still overlooks the kernel issue in reward and reprimand: justice, fairness and appropriateness. The model might in fact make a harmful contribution to the field by perpetrating a replacement of this kernel issue with those of gamesmanship and attitude.
Thus, if we are to benefit from the diversity in this collection, we need to understand the diversity of such fundamental, kernel issues, and know how to differentiate the more fundamental diversity from that exhibited by differences between types of music or sport.
2.1 Towards a Coherent View of Diversity
An understanding of diversity cannot come from any one science, not even psychology, because each science has its main focus on one aspect of reality. Clouser  argues that when a science tries to explain other aspects in its own terms the result is often a reduction that makes important aspects invisible (such as attempting to explain life in terms of physics alone). A clear example of focus on a single aspect may be found in Coutinho & Cangelosi's chapter on the impact of musical emotions. Music, as an everyday experience, has aesthetic, historical, social and emotional aspects, among others. The chapter looks at music through the eyes of psychology, studying the emotional aspect in detail; it does not claim to address the aesthetic issue of what makes great music, so it does not fall into the trap of reductionism. With Clouser, Dooyeweerd [1955/1984] and Hart  argue that science is fundamentally unable to span across the diversity of aspects, but that it is the role of philosophy to attempt this. Therefore, it is to philosophy that we look to understand the nature of diversity. Three approaches will be examined before a proposal is made.
The philosopher Heidegger [1927/1962] explored the nature of existential interaction with the world. An entity gains its 'being' by virtue of its being situated in a world. From Descartes, Western thinking had assumed the notion of subject and object to be one of an active, thinking agent separate from world of objects. This caused many problems, not least of which was the elevation of the aspect of thinking above all others, as depicted in the left-hand diagram above. Heidegger tried to dissolve the distinction between subject and object and investigated the nature of our interaction with, or our being in, the world. Unfortunately, his extensive studies focused only on the nature of this being-in, and provides very little help in understanding the nature of its diversity.
Evolution theory concerns itself with diversity, insofar as it offers an explanation for how diversity comes about, usually over a long period. Evolutionary biological science has been especially powerful in explaining the origin of types of living thing (species). Evolutionary philosophy has been harnessed to explain the origin of diverse aspects of human functioning, such as language, social activity, ethics and religion. Unfortunately, it is not as much help here as we might have expected, because evolution theory explains only the process of emergence of diversity, and not what diversity we face today. Indeed, it must presuppose the latter in that, before it can begin to construct an explanation of how, for example, language or ethics emerged, we must present it with a prior understanding of what language and ethics mean. All we need here is this understanding of what diversity there is, which is prior to evolution theory.
An approach that claims to provide such prior understanding of what diversity there is, is to look at ontologies of the world offered by various thinkers. For example, Bunge  distinguished five 'systems genera' in the world: physical, chemical, biological, technical, social. As soon as we try to apply this, its limitations become clear. We are interested in diversity of interactions; these must first be rendered as systems (example from Fig. 1: the system might be person, frying pan, egg and tossing), but two problems emerge as we try to do this. One is that drawing a boundary round the system can prove difficult or even meaninless (e.g. for football in Fig. 1, because football is not just a team playing with a ball but is a whole culture). The other is that an interaction can be of two different systems, for example the cooking of the egg might be for food (biological system) or to show how it is done (educational system), or both. A third problem is with Bunge's ontology: it is too limited for our use, dissolving the difference between, for example, phoning, playing football, raising one's hat, which are all social in nature.
These problems occur because Bunge's ontology, and others like it such as Hartmann's  'strata', Habermas'  types of social action and Maslow's  needs, are ontologies of types of thing or action - the 'what' of the cosmos - rather than ways or modes in which things can be or occur - the 'how' of the cosmos. What is needed is a comprehensive delineation of different 'how's, or ways in which things can be and occur. (Strictly, this should not be called an ontology.)
2.2 Dooyeweerd's Aspects
Dooyeweerd [1955/1984] is one philosopher who might offer the view we need. He drew up a set of fifteen aspects or law-spheres, which delineate the 'how', the various meaningful ways in which things can be, occur and interact. His set of aspects has a wider range than the ontologies above (see Basden [2008, p.66] for a comparison). Moreover, Dooyeweerd held that things or activities exhibit many aspects, in a way that will be useful here. Each aspect has a distinct kernel meaning that defines ways in which things may be meaningful to us. Table 2 shows these, and also how the interactions listed in the Introduction might link with some aspects.
|Aspect||Meaning kernel||How it relates to Figure 1 etc.|
|Quantitative||to do with quantity, amount|
|Spatial||to do with continuous extension, space|
|Kinematic||to do with movement; flowing movement||Yoyo, football and egg-tossing involve flowing movement.|
|Physical||to do with fields, forces, energy + mass||The yoyo works because of gravity and momentum.|
|Biotic / Organic||to do with life functions||The egg is meaningful as food
Dumbbells are meaningful to build up muscles
|Psychic/ Sensitive||to do with sense, feeling, emotion|
|Analytical||to do with distinguishing|
|Formative||to do with history, culture, technology: shaping, achievement, goals, ends, means, techniques, structures||The keyboard is meaningful for achieving things.|
|Lingual||to do with symbolic signification, language||Phoning is meaningful as communication. |
The book is meaningful as information.
The keyboard is meaningful for writing.
Cooking an egg may be a demonstration.
|Social||to do with social interaction and institutions||Raising one's hat is meaningful as social etiquette.|
|Economic||to do with frugality, resources, management||The keyboard is meaningful for work.|
|Aesthetic||to do with harmony, surprise, fun, beauty||Painting is meaningful as the aesthetic activity of art. |
The football and yoyo are meaninful as play.
|Juridical||to do with what is due; rights and responsibilities, justice, appropriateness||reprimanding or rewarding children|
|Ethical||to do with self-giving love, e.g. generosity||caring for others|
|Faith / Pistic||to do with belief, vision, aspiration, commitment, loyalty, creed, religion||Religious worship|
Dooyeweerd's fifteen may be seen as a superset of the other categorizations mentioned above and is able to differentiate most of the activities in the right-hand diagram, as shown in the third column of that figure in ways that allow activities to be meaningful in several ways.
Though Dooyeweerd emphasised that all things and activities do in fact exhibit all aspects (so cooking an egg might be primarily biotic in function, but it might also have a social function with guests, an aesthetic function in the enjoyment of food, a lingual function in education, a formative function as a skill, and so on), he proposed that there is usually one aspectual function that 'leads' the activity. The leading aspect is the one that most defines the way it is meaningful to us, and the other aspects support it. Dooyeweerd saw aspects not only as categories but as the underlying cosmic mechanism of meaningful diversity, in that they are the transcendental conditions that enable everything to be, occur and interact.
Dooyeweerd did not claim that his suite of aspects is in any way a final, complete list; indeed he claimed the opposite [Dooyeweerd, 1955/1984,II, p.556]. But it is useful to us here because it is not only more comprehensive than others, but also more grounded in everyday experience, and more soundly based in philosophy, including critical philosophy [Basden & Wood-Harper, 2006]. We will adopt it here to gain insight into the diversity of contributions in this work.
3. UNDERSTANDING DIVERSITY IN THE CHAPTERS
Dooyeweerd's aspects will be used here as an analytical tool. The main aspects of the application areas mentioned in each chapter will be identified, with the links between them. Secondary aspects are ignored.
3.1 Aspectual Profiles of Chapters
Figure 2 depicts these main aspects of each chapter, identified by the number given in Table 1. For most chapters at least two aspects are shown, linked by an arrow, the meaning of which is explained in the text below.
Figure 2. Aspectual profiles of application areas
A brief explanation of why it is reasonable to identify certain aspects with each chapter follows; the reader might find it useful to refer to the aspect kernels in Table 2.
Chapter 1-M (Montebelli et al.) discusses how the "non-neural" bio-regulative states of an agent affects sensori-motor activity. This is naturally expressed in Fig. 2 by an 'affects' arrow from the biotic to the analytic aspect.
Chapter 2-M (Montesano et al.), using the notion of affordance, investigates how a robot can categorise object properties of a sensori-motor kind (visual and tactile), identify relationships between them so as to imitate a human performing simple actions. Categorization is an analytic functioning and sensori-motor functioning is psychic, and so this chapter's interest may be expressed by an arrow between these two aspects meaning 'allows us to infer'.
Chapter 3-D (DurÃ¡n) applies nonlinear dynamical systems theory to simulating the behaviour of an infant who looks in the wrong place for a toy, even though they have seen the toy put in a different place (the A-not-B error noted by Piaget). From the researcher's perspective, the stimulus given to the infant is movement and the error is one of wrong discrimination, so we may plot the interest of the chapter as how kinematic affects analytic functioning.
Chapter 4-V (Vernon et al.) is a different kind of chapter, which does not include any cognitive modelling approach as such. Rather, it contrasts two paradigms, cognitivist and emergentist. Here we do not take sides, but only try to express the contrast by an arrow that visually drives two aspects apart. Cognitivism stresses abstract conceptualising and separation from the environment (analytic aspect), while emergentism stresses the "embodied and embedded" nature of the agent, which is a strongly biotic theme.
Chapter 5-C (Coutinho & Cangelosi) investigates several models of how music affects emotions. In aspectual terms, music is primarily of the aesthetic aspect, though it involves many others, and emotions are of the psychic aspect. So we may express the link discussed in the chapter by an 'affects' arrow from the aesthetic to the psychic aspect.
Chapter 6-N (Nefti et al.) argues that the very promising PSO method of numerical optimization lacks the ability to handle risk and uncertainty, and proposes that insights from a psychological theory can provide this ability. So this chapter may be expressed by an 'improves' arrow from the psychic to quantitative aspect.
Chapter 7-R (Ruini et al.) stresses the distinction between strategic and tactical levels of behaviour. But what is meant by 'strategic' and 'tactical'? The specific application concerns (artificial) organisms moving towards food or water or away from predators under 'motivations' of hunger etc., and thus the two levels might seem to refer to the psychic and biotic aspects. The authors' interest seems wider, however: the link between goals that direct behaviour (formative aspect) and what it is that affects choice between them when they conflict. Choice among goals presupposes a belief, often tacit, by the organism about what is of ultimate importance (such as expressed by "I must stay alive in order to reproduce"). Such beliefs are pistic in nature. If this does indeed reflect the authors' interest then it may be expressed in Fig. 2 by an 'affects' arrow from the pistic to formative aspect.
Chapter 8-G (Gemrot, et al.) presents an approach to building bots for games, with the aim that they are as realistic as possible. This kind of realism, which enhances the quality of games, is constituted in giving due respect to reality, which is the juridical aspect. Bot-building is a formative activity. So a link is shown ('affects') from the juridical to the formative aspect.
Chapter 9-S (Simko & Cummins) looks at how the bodily characteristics of organs affect how speech is actually carried out, with reference to similar work related to gestures. To speak of bodily characteristics is meaningful in the biotic aspect, with a few features like mass being meaningful in the physical aspect, while speech and gestures, as language, are concepts that are meaningful in the lingual aspect. So the theme of this chapter may be represented aspectually by an 'affects' arrow from biotic to lingual aspect.
Chapter 10-A (Alissandrakis et al.) discusses how robots choose what, of humans, to imitate, by a cyclical communication process that consists of human gestures to demonstrate what to do and feedback from the robot about what they have learned. This may be expressed in Fig. 2 by an 'affects' arrow from the lingual to the analytic aspect.
Chapter 11-K (Kutar & Basden) presents the Cognitive Dimensions Framework as a way of evaluating the degree of satisfaction users experience when interacting with information or symbols. This may be expressed by an 'affects' arrow from the lingual aspect to that of satisfaction, the psychic.
Chapter 12-G (Grinsberg, et al.) discusses how the Prisoners' Dilemma type of games may be approached with a view to successful outcome. Success is of the formative aspect. Central to the Prisoners' Dilemma is the ethical distinction between self-giving (cooperation) and self-interest (defection). So the interest of this chapter may be expressed by an 'affects' arrow from ethical to formative.
Chapter 13-O (Oussalah & Strigini) addresses the general problem of adjudication between solutions offered by different agents. Such adjudication is largely of the analytic aspect here, despite the juridical overtones of the word used. Unlike other chapters, this one seeks to generalise across types of application (physical, sensory, social, etc.) so this is expressed in Fig. 2 by multiple 'can apply to' arrows radiating from the analytic aspect.
Chapter 14-T (Tsagarakis et al.) is concerned that the hard, rigid actuators of conventional robots threaten the safety of the human body, and proposes using soft, compliant ones instead. It discusses both material properties and control mechanisms for this. This may be expressed in Fig. 2 by two 'affects' arrows converging on the biotic aspect from the physical and formative.
Chapter 15-L (Lovric, et al.) aims to identify the types of conceptual structures that arise from aggregate financial decisions. We may express this as an 'allows us to infer' arrow from the social and economic aspects (aggregate and financial) to the analytic and formative aspects (concepts and structures).
Chapter 16-P (Pfeifer & Gomez) shows by numerous examples drawn from the natural world how appropriate morphology can greatly reduce the complexity of computation needed for good sensory or motor functioning. The theme is that the biotic aspect directly improves the psychic functioning, as well as indirectly by providing a substratum for it.
3.2 Patterns Emerging
This aspectual picture provides various ways of discerning characteristics of the book as a whole, some of which merely support what we might already suspect, while some disclose new and surprising things.
The first thing that is obvious is the rich diversity of types of application of cognitive modelling. That they are diverse is obvious from a cursory reading of the chapters, but an aspectual analysis reveals how rich the diversity is. Thirteen of Dooyeweerd's fifteen aspects are represented. The wide range of aspects represented in the collection means that it offers a usefully wide range of exemplars. In some multi-aspectual edited works of this kind, one aspect predominates, and the others feel like outliers, just bolted on for no real reason. Though certain aspects - the biotic, psychic, analytic and formative - occur more frequently, there is no sense that others are 'bolted on'. Rather, these central aspects are almost always linked with one of the others.
A second thing we might notice, related to the above, is that the majority of inter-aspect links in the chapters are those that involve impact ('affects') or the coming-into-being of something different ('allows us to infer'), which are two expressions of embodied interaction with the world. The whole work is thus revealed to be no merely static collection but has a dynamic character, almost alive with potential.
A third observation discloses types of interaction with, and embodiment in, the world. The central aspects, which are mentioned in most chapters, are aspects of our individuality as human beings. Aspects of individuality are those whose laws concern activity or behaviour as an individual, and include the biotic aspect (laws of enabling an organism to remain distinct from its environment), the sensitive aspect (laws of sentience and response), the analytic aspect (laws concern conceptualisation, distinction and categorisation) and the formative aspect (laws of intentional formation, design, achievement). The laws of the physical aspect, by contrast, concern fields etc. which pervade individuals, and those of post-formative aspects largely presuppose social or inter-individual functioning. This suggests that there are at least four kinds of interaction or embodiment that an individual can experience with the world: biotic, sensory, conceptual and intentional, and that they are fundamentally different from each other such that each may be researched in its own right even when in combination. It may be helpful, therefore, to see the 'cognitive body' that is build up by interaction as not just biological, but also pychological, analytical and intentional. Another interesting proposal, which deserves research attention, comes from Dooyeweerd's contention that we function in all aspects: that all of these aspects of individuality work in harmony.
The fourth observation is that in most cases the aspect of individuality is linked to an aspect of application or world. Each of the four may in principle interact with any of the aspects of the world. The environment with which the 'body' interacts exhibits many more aspects, not only the biotic to formative, but also the quantitative, spatial, kinemantic, physical, lingual, social aspects, and aspects of resources, beauty, harmony and fun, responsibilities, and it is ethical and spiritual in nature. This picture of interactivity and embodiment not only displays the diversity of types of interaction with which this chapter began, but provides a basis for addressing (understanding, discussing, studying) the diversity of embeddedness in the world, of enaction, of embodiment.
3.3 Pointers to Future Work
Some application aspects, and so some types of interaction or embodiment, are not as well represented in this collection as others. Though the quantitative aspect pervades discussion of modelling techniques, only one chapter discusses our interaction with the world of numbers. Though it has been argued that one chapter is interested in the pistic functioning of beliefs that lead to choice of goals, this is rather indirect; interaction with, and embodiment in, the world of beliefs and commitments needs much fuller and more explicit research. Likewise, the ethical aspect is present only in the very limited form of collaboration versus defection, and the richness of generosity and self-giving (or their opposites) experienced in everyday life deserves much fuller treatment. A similar thing could be said in relation to the juridical, aesthetic, economic and social aspects. Thus there is considerable potential for future research in how cognition is embodied in all these aspects of life.
A second, and perhaps more fundamental, suggestion is that the nature of cognition and how it is built up might differ in fundamental ways for each aspect of the world that is meaningful when interaction occurs. To put it another way, the different types of application should not be seen as merely requiring different models to be built up, but as requiring different ways of building models. This is perhaps more fundamental than most chapters in this work realise, and will be discussed at some length.
In Fig. 1 different rules apply to each type of interaction (football, cooking, etc.). These are not just different rules but different kinds of rule, because each kind of interaction is meaningful in different ways. This gives two levels of rules. If a cognitive agent is to enactively learn and construct a set of rules for interacting, then it already presupposes meta-rules that guide it in the construction of its set, which encapsulate kinds of rationality that lie behind the rules. Not only are the rules diverse, but so might be the meta-rules, as is already reflected in this collection in the difference between neural and semantic net approaches.
In a classic article, Winch  argued that there are different kinds of rationality (specifically those directing the physical sciences, social science and religion). Dooyeweerd extended this idea and gave it concrete form in his theory of aspects. Table 3 summarises the rationality of each aspect, and with what it is concerned; the third column is explained below. The force of these aspectual rationalities may be understood if we reflect on situations in everyday life and see the variety of ways in which things 'make sense' or in which we justify our actions.
|Aspect||Rationality concerned with||Modelling approach|
|Quantitative||Arithmetic, statistics, calculus|
|Physical||Causality over fields|
|Biotic / Organic||Health, growth, reproduction and speciation|
|Psychic/ Sensitive||Responses to stimuli and mental states||Connectionism|
|Analytical||Discrimination, categorization, logic||Connectionism |
|Formative||Plans and actions to achieve goals and intentional change||Object-orientation|
|Lingual||Representation, cross reference, Communicative rationality [Habermas, 1986]|
|Social||Respect and role|
|Economic||Frugal management of resources|
|Aesthetic||Harmonising, playing and humour|
|Juridical||Ensuring due; legal argument|
|Faith / Pistic||Commitment, relationship with the Absolute|
The suggestion here is that when a cognitive agent constructs its set of rules for interacting with the world in a way that is meaningful in a certain aspect, the meta-rules for such construction should be appropriate to that aspect's rationality, and that meta-rules of other aspects are inappropriate. For example, when constructing a model of physical interaction, expect to include causal relationships and fields, but when contructing a model of interaction with symbols, expect relationships of cross-reference, expect that some of them will be implicit rather than explicit, but do not expect to include causal relationships. Some sets of aspectual meta-rules might prove less relevant to specific types of interaction than others, but in the general case every set should be made available within the agent. This has much to do with affordance and a small example of it in operation can be found in Montesano et al.'s chapter in this work. They provided meta-rules, in the simplified form of vectors of properties, for a robot to follow in making up its rules for interaction. The robot was guided to consider both physico-spatial and psychic properties and, as expected, found that the psychic property (colour) was irrelevant for physical tasks and did not contribute to the set of rules it built up.
More generally, both connectionist and object-oriented approaches presuppose their own specific types of meta-rules, as indicated in the third column of Table 3. Connectionist (neural net) meta-rules assume rationality of the psychic aspect insofar as they operate by links between input and output, and the analytic aspect insofar is they presuppose vectors of distinct inputs and outputs. The meta-rules of object orientation (or semantic nets) assume rationality of the analytic aspect insofar as they presuppose distinct, categorized objects, and the formative aspect insofar as they operate by methods.
If we assume we can build rules using meta-rules of a different aspect then something important will be lost and the behaviour of the resultant agent will be unrealistic, unnatural or clumsy. Thus neither connectionist or object-oriented approaches are, on their own, appropriate for constructing models or sets of rules for the lingual or spatial aspects, for example. The meta-rules for the lingual aspect are concerned not only with very complex non-linear structures in the continuous stream of text or speech, but also with what Basden & Klein  call text-meaning (semantics) and especially with what they call life-meaning, which includes allusions, idioms and cultural connotations and illocutionary intention. The meta-rules for the spatial aspect are concerned with continuous extension, such as when a shape is expanded and, as Basden  shows, this can result in structures that were not allowed for. Neither connectionism nor object-orientation are appropriate for these and, if applied, can make the building up of the model (whether by enaction or design) cumbersome and error-prone. Instead, cognitive agents should have encapsulated within them meta-rules that recognise such lingual things as allusion, idiom, connotation and illocutionary intention, spatial things like continuous extension, and similar things germane to each and every aspect. That this is technically possible is evidenced by, for example, the powerful text-understanding mechanisms employed on the Internet and geometric reasoners of the type originally suggested so long ago by Funt .
So the second strategic suggestion for future research is that enactive cognitive agents should be furnished with the whole set of types of aspectual rationality, encapsulated as meta-rules for the building up of sets of rules (models) when they interact with the world. To furnish agents only with neural or semantic net formalisms, as most chapters here do, does them a disservice, and it will be an interesting challenge to expand to other aspects. This will be a long-term research programme. It is likely to require exploration of meta-rules for each aspect in turn, followed by research into how to integrate them into an overall system. Perhaps Oussalah & Stragini's chapter on adjudication might contribute to the integration?
The benefits of this approach accrue not only to those working under the enactive paradigm, however, but also to those working under the traditional paradigm, where the model in the artificial agent is designed and written by human knowledge representation activity. Basden  has discussed in some detail how the aspects may be used in a manner similar to the above to devise knowledge representation formalisms that are appropriate for each aspect. The benefits include faster, more natural modelling of real-world multi-aspectual complexity, which is less error prone and generates models with greater robustness and longevity. Nevertheless, in line with the prevailing attitude in this collection, the focus has been on enriching the enactive, 'embodied' paradigm.
A third suggestion for future research is that the aspectual approach outlined in this chapter might contribute to meeting the challenges posed in Vernon's chapter, which must be faced when adopting the embodied / enactive approach.
- Challenge 1 is that understanding how cognitive capabilities develop by building on sensory-motor capabilities is not straightforward. This challenge may be accounted for by the analytic aspect being irreducible to the psychic aspect. However Dooyeweerd maintained that each aspect has inherent connections of dependency and analogy with others so his discussions of how the psychic aspect 'anticipates' the analytic might contribute to meeting this challenge.
- Challenge 2 states that the speed of developing cognitive capabilities is limited to speed of interaction with environment. This argument, however, is at the level of rules rather than meta-rules; so it may be that providing the agent with high quality sets of meta-rules, as proposed above, can speed this up.
- Challenge 3 is to incorporate motivations that underpin goals, and the interplay between motivations of different types. Motivation makes goals meaningful and Vernon identifies two types, social and explorative. Dooyeweerd would suggest there are many more than two ways of being meaningful (viz. aspects), thus extending the challenge, but his extensive exploration of inter-aspect relationships might help meet it.
- Challenge 4 is to extend the self-construction of world models beyond sensory-motor perception-action, to what Vernon calls the 'more abstract knowledge' so important to real cognitive systems. Dooyeweerd not only recognised the 'more abstract' aspects, but has explored their types, ranging from analytic to faith types. Providing agents with meta-rules for all these might help meet this challenge.
- Challenge 5 is that new adaptive materials must be integrated into morphological models of cognition. This requires new ways of thinking because traditional physics has presupposed distinct entities but this is inappropriate when using compliant materials as discussed in Tsagarakis et al.'s chapter. Dooyeweerd's elucidation of kernel meanings of aspects would suggest that distinction is not to be found in the physical aspect itself but in the analytic aspect that has traditionally been uppermost in the act of thinking about physical reality. He thus clarifies the nature of the difficulty. Attending to the kernel meanings of aspects might open up not only the possibility of using adaptive materials but other new approaches to embodied cognitive systems.
In moving away from a rationalistic view of cognition and AI towards a more 'embodied' view, we find that the important issue is not that body is opposed to mind, but rather the existential notion of interaction with the world. That this is so has been recognised increasingly over the past decade. But, accepting that it is so thrusts upon us the next challenge: the diversity of types of meaningful interaction. In our 'embodied' or existential state we interact in many types of ways, each of which can influence the types of cognitive modelling that are undertaken. So it is time to begin addressing seriously the nature of the diversity of ways in which we interact with the world.
In the field of cognitive modelling (including AI) these types of interaction are the types of application area, and they affect the very core of modelling itself. This collection of chapters provides a useful variety of types of application area and it can thereby provide a good study of meaningful diversity and the way diversity can affect modelling.
We have examined the potential of Dooyeweerd's [1955/1984] notion of modal aspects as a basis for understanding the nature of diversity. Aspectual thinking is common whenever thinkers wish to differentiate things that cannot be reduced to each other but must be considered separately during analysis, and many suites of aspects (or levels, strata, genera, action types, needs, etc.) have been proposed. Dooyeweerd's suite has certain advantages over most others, of being more comprehensive and more soundly based in both philosophy and everyday experience. His theory of aspects also allows things to function in more than one aspect in a coherent way.
We wanted to know whether and how all the chapters relate to each other. We can see that the chapters cover almost the entire range of Dooyeweerd's aspects. In almost all cases, at least two aspects were found, giving a richer picture than mere aspectual categorisation would have done. Each chapter is seen to encapsulate a dynamic, in which one aspect affects another, one being an aspect of individuality and the other an aspect of interaction with the world.
The aspects of our individuality include the biotic aspect of organic life, the psychic aspect of sentience, the analytic aspect of conceptualisation, the formative aspect of intentional structuring or processing. So this kind of analysis might enrich the notions of body and embodiment, to involve at least four aspects, all working in harmony as instruments do in an orchestra.
We wanted to know how to continue research in the future. One avenue is to investigate how these aspects of individuality or embodiment can work in harmony. Another is to recognise that each aspect of individuality interacts with aspects of the world that make the interaction meaningful, and to investigate the nature of embodied interaction in each such aspect; to increase generality, it might be particularly useful to study interactions found in majority-world situations. After this, it would be important to investigate the nature of multi-aspectual embodied interaction.
We wanted to know how to cope with possible new types of application and interaction not well represented here. The proposal is to furnish artificial cognitive agents with meta-rules that encapsulate the basic type of rationality and functioning of every aspect. No longer need researchers be restricted to neural or semantic net approaches, but a future can be envisaged in which a variety of types of meta-rules are integrated together, each encapsulating a different aspect of the world. This might contribute to addressing the five challenges for the embodied interaction paradigm that Vernon poses in his chapter.
This collection of chapters, entitled Advances in Cognitive Systems, not only deals with cognitive techniques and technologies, not only promotes an enactive-embodied paradigm, but, in the diversity of applications that it encompasses, it tentatively points a way towards the issue that is hastening to meet us very soon, the issue of diversity itself.
Basden, A. Philosophical Frameworks for Understanding Information Systems. 2008. IGI Global, Hershey, PA, USA.
Basden, A. & Klein, H.K. New Research Directions for Data and Knowledge Engineering: A Philosophy of Language Approach. Data & Knowledge Engineering, 2008, 67(2), p.260-285.
Basden, A. & Wood-Harper, A. T. A philosophical discussion of the root definition in soft systems thinking: An enrichment of CATWOE. Systems Research and Behavioral Science, 2006, 23, 61-87.
Bunge, M. Treatise on basic philosophy, Vol. 4: Ontology 2: A world of systems. 1979. Boston: Reidal.
Clouser, R. The Myth of Religious Neutrality; An Essay on the Hidden Role of Religious Belief in Theories. 2005. University of Notre Dame Press, Notre Dame, Indiana, USA.
Dooyeweerd, H. A New Critique of Theoretical Thought, Vol. I-IV. 1955/1984, 1975 edition. Paideia Press, Jordan Station, Ontario.
Funt, B.V. Problem-solving with diagrammatic representations. Artificial Intelligence 1980, 13(3):201-30.
Habermas, J. The theory of communicative action: Vol. 1. Reason and the rationalization of society. tr. T. McCarthy. 1986. Cambridge, England: Polity Press.
Hart, H. Understanding Our World: An Integral Ontology. 1984. Langham, USA: University Press of America.
Hartmann, N. The new ways of ontology. 1952. Chicago: Chicago University Press.
Heidegger, M. Being and Time. tr. J. Macquarrie, B. Robinson. 1927/1962. Blackwell Publishing, Oxford, UK.
Maslow, A. A theory of human motivation. Psychological Review, 1943, 50, 370-396.
Pfeifer R, Bongard, J. How the Body Shapes the Way We Think: A New View of Intelligence. 2007. Bradford Books, MIT Press, Cambridge, MA. USA.
Winch, P. The Idea of a Social Science and its Relation to Philosophy. 1958. London: Routledge and Kegan Paul.