Unifying the Mind

Unifying the Mind: Cognitive Representations as Graphical Models

David Danks
Copyright Date: 2014
Published by: MIT Press
Pages: 304
  • Cite this Item
  • Book Info
    Unifying the Mind
    Book Description:

    Our ordinary, everyday thinking requires an astonishing range of cognitive activities, yet our cognition seems to take place seamlessly. We move between cognitive processes with ease, and different types of cognition seem to share information readily. In this book, David Danks proposes a novel cognitive architecture that can partially explain two aspects of human cognition: its relatively integrated nature and our effortless ability to focus on the relevant factors in any particular situation. Danks argues that both of these features of cognition are naturally explained if many of our cognitive representations are understood to be structured like graphical models. The computational framework of graphical models is widely used in machine learning, but Danks is the first to offer a book-length account of its use to analyze multiple areas of cognition. Danks demonstrates the usefulness of this approach by reinterpreting a variety of cognitive theories in terms of graphical models. He shows how we can understand much of our cognition -- in particular causal learning, cognition involving concepts, and decision making -- through the lens of graphical models, thus clarifying a range of data from experiments and introspection. Moreover, Danks demonstrates the important role that cognitive representations play in a unified understanding of cognition, arguing that much of our cognition can be explained in terms of different cognitive processes operating on a shared collection of cognitive representations. Danks's account is mathematically accessible, focusing on the qualitative aspects of graphical models and separating the formal mathematical details in the text.

    eISBN: 978-0-262-32544-8
    Subjects: Psychology, Philosophy

Table of Contents

  1. Front Matter
    (pp. i-vi)
  2. Table of Contents
    (pp. vii-viii)
  3. Acknowledgments
    (pp. ix-xii)
  4. 1 “Free-Range” Cognition
    (pp. 1-12)

    Suppose you find yourself in the wilderness, hungry and lost. The particular reason you ended up there is relatively unimportant; perhaps you were hiking through a cloud bank and lost your bearings, or perhaps you were a contestant on a popular game show. The key challenge, regardless of what previously happened, is to figure out how to survive. More precisely, you need to engage in different types of cognition to, for example, determine which objects might potentially be food, observe other animals’ eating habits to determine what is possibly safe, and make decisions (possibly involving hard trade-offs) about what exactly...

  5. 2 Computational Realism, Levels, and Constraints
    (pp. 13-38)

    This book argues that we can fruitfully understand significant parts of human cognition as different processes operating on a shared representational store of graphical models. My approach throughout is unabashedly computational: cognition will be understood as precise operations on mathematically well-specified objects. More precisely, a background assumption of this work is that it is appropriate to think about aspects of cognition as fundamentally involving learning, transforming, making inferences using, and manipulating structured representations about the world.¹ One challenge with computational models of the mind is that it is often unclear just what commitments—metaphysical, epistemological, and methodological—are intended for...

  6. 3 A Primer on Graphical Models
    (pp. 39-64)

    The previous chapter laid the philosophical groundwork for the cognitive architecture that I will present. This chapter introduces and explores the framework of graphical models, the computational and mathematical basis of that architecture. The main thread of the chapter focuses on a conceptual introduction, with mathematical details provided in sections marked with an asterisk (*); the reader can skip those sections with little loss of (qualitative) understanding. Section 3.1 explores the common elements shared by all graphical models, with subsequent sections examining the details of specific model types. As with many formal frameworks, a high-level grasp of different graphical model...

  7. 4 Causal Cognition
    (pp. 65-98)

    We now have the pieces in place to express particular areas of cognition as operations on cognitive representations structured as graphical models, and to actually understand what that means. We begin in this chapter with causal cognition. Causation is one of the unifying threads of our cognition (Sloman, 2005). We interpret disparate events as parts of a coherent causal structure. We use our causal knowledge to predict future states of the world. Our choices and actions in the world are influenced and guided by our understanding of the causal relations around us. Perhaps most importantly, we understand the difference between...

  8. 5 Concepts, Categories, and Inference
    (pp. 99-128)

    The previous chapter showed how to use graphical models to capture our cognitive representations of causal structure, and the operations we perform in causal learning and reasoning. As I noted there, however, causal cognition was arguably an easy target, as one of the principal uses of graphical models over the past twenty years has been to model causal structures in the world. If our cognitive representations track the world to any significant degree, then we should expect that much of our knowledge about causal structures should be structured approximately as a DAG-based graphical model. A far more challenging task is...

  9. 6 Decision Making via Graphical Models
    (pp. 129-150)

    The previous two chapters focused on using graphical models to represent our causal and conceptual knowledge. Those cognitive representations are, however, essentially impotent on their own; they are only useful if they are connected in some way with decision-making processes that can (intelligently) use them. The first section of this chapter aims to show that causal knowledge—represented as a DAG-based graphical model—plays a key role in much of our decision making. In particular, causal knowledge can guide us to attend to the proper factors and enable us to better predict the outcomes of our own actions. One type...

  10. 7 Unifying Cognition
    (pp. 151-174)

    The previous three chapters explored three distinct types of cognition, focusing on causation, concepts, and decision making. Although these cognitive processes were considered separately, the analyses were bound together by the thread of graphical models: all three types of cognition were understood using that computational/mathematical framework. This chapter now argues for the position that was explicitly introduced in chapter 1 but left largely implicit in chapters 4 through 6: I contend that large swaths of human cognitive activity can fruitfully be understood as different operations on a shared representational store, where those cognitive representations are (approximately) graphical models. In some...

  11. 8 Alternative Approaches
    (pp. 175-204)

    The previous chapter provided an integrated cognitive architecture that aimed to unify many aspects of cognition as different processes on a shared store of cognitive representations, structured as graphical models. The goal of a unified model of much of cognition is obviously not new but rather has been a perennial target; we can even understand Plato’s account of the soul in thePhaedoas an attempt to build a unified understanding of the mind. The belief that my mind actually is a relatively unified, coherent object or process naturally arises from both self-introspection and other-observation. Demonstrations that aspects of cognition...

  12. 9 Broader Implications
    (pp. 205-222)

    At this point in the book, I want to step back from the details of particular cognitive models and architectures to consider some of the broader, more philosophical implications of the cognitive architecture that I have advanced and defended in previous chapters. That is, how (if at all) does it matter if our cognition really is partly unified by virtue of a shared store of cognitive representations that are structured (approximately) as graphical models? Many connections could be drawn, but I focus here on three topics for which there are both significant implications and also interesting open research questions. Section...

  13. 10 Conclusions, Open Questions, and Next Steps
    (pp. 223-228)

    This book has covered a lot of ground to characterize and evaluate the integrated cognitive architecture, but much remains to be done. On the empirical front, many different experiments should be run (and some are in progress) to better specify aspects of the architecture. In earlier chapters, I described multiple experiments that should be informative about the exact cognitive representations and processes within each type of cognition. Perhaps more importantly, many experiments remain to be done to better understand exactly what types of learning and goal effects arise from multiple processes acting on the same, shared representation. For example, suppose...

  14. Appendix: Graphical Models and Concepts
    (pp. 229-238)
  15. Notes
    (pp. 239-250)
  16. References
    (pp. 251-284)
  17. Index
    (pp. 285-288)