You are not currently logged in.

Login through your institution for access.


Log in through your institution.


Anigrafs: Experiments in Cooperative Cognitive Architecture

Whitman Richards
Copyright Date: 2015
Published by: MIT Press
Pages: 168
Stable URL:
  • Cite this Item
  • Book Info
    Book Description:

    In this book, Whitman Richards offers a novel and provocative proposal for understanding decision making and human behavior. Building on Valentino Braitenberg's famous "vehicles," Richards describes a collection of mental organisms that he calls "daemons" -- virtual correlates of neural modules. Daemons have favored choices and make decisions that control behaviors of the group to which they belong, with each daemon preferring a different outcome. Richards arranges these preferences in graphs, linking similar choices, which thus reinforce each other. "Anigrafs" refers to these two components -- animals, or the mental organisms (agents or daemons), and the graphs that show similarity relations. Together these two components are the basis of a new cognitive architecture. In Richards's account, a collection of daemons compete for control of the cognitive system in which they reside; the challenge is to get the daemons to agree on one of many choices. Richards explores the results of group decisions, emphasizing the Condorcet voting procedure for aggregating preferences. A neural mechanism is proposed.Anigrafspresents a series of group decisions that incorporate simple and complex movements, as well as aspects of cognition and belief.Anigrafsconcludes with a section on "metagrafs," which chart relationships between different anigraf models.

    eISBN: 978-0-262-32911-8
    Subjects: Biological Sciences, Technology
    × Close Overlay

Table of Contents

Export Selected Citations
  1. Front Matter (pp. i-vi)
  2. Table of Contents (pp. vii-x)
  3. Preface (pp. xi-xiv)
    W. Richards
  4. Part I Preliminaries: From Babble to Barter and Beyond (pp. 1-14)

    Decision-making is a reflection of intelligence. The lowest forms of animals make reflexive actions, whereas advanced mammals, living in complex environments, make decisions for actions that are often also complex. In such environments, many possible alternatives need to be evaluated before a commitment is made. There are two aspects of this part of intelligence that interest us here: first, there is the procedure for making a decision among many alternatives; second, there are the possible relations among these alternatives, which we dub thecognitive architecturesof the system. These alternatives are possible choices for actions. Cognitive architectures thus capture the...

  5. Part II: Animacy:: Action Agents
    • Anigraf1: Simple Precursors (pp. 17-26)

      Figure 1.1 shows two prototype anigrafs. One of them, a pentagon, provides a clear but simple example of periodic behaviors. In both prototypes, each node represents one of the mischievous daemons or agents that controls the state of its special physical actuator. Each actuator initiates an action or behavior that moves the system closer to the goal the agent desires. Although one might attempt to capture these same relationships as parts of each actuator, such a model would confuse two levels of description we wish to separate: the level of the physical mechanism versus the level of the relations between...

    • Anigraf2: Swimmers: Beginning to Move (pp. 27-36)

      Precursor anigrafs had very limited behaviors, constrained for didactic purposes to simple ring forms. This simplification hid the potential complexity of designs and behaviors that could be attributed to primitive life forms. Here we begin to expand this repertoire. The component mental organisms, or “agents,” will have access to different kinds of interfaces with the environment, with interface hardware that affects the behavior of the system as a whole. Each agent will have control over those additions, and will “vote” depending upon the strength of its desire to achieve a preferred goal. In effect, then, we are formalizing a two-tier...

    • Anigraf3: Walkers: Syncopated Limbs (pp. 37-48)

      Swimmers have limited freedom of movement. Their actions are thus limited, and, consequently, we do not expect such creatures to possess a high level of intelligence. Our most advanced swimmers use wavelike motions of a segmented appendage for locomotion. Resident within each segment is an individual agent. A chain of these agents guides a wave motion of the body or limb segments. To enhance the behavioral repertoire of such anigrafs, an obvious next step is to add more limbs. Alternatively, the chain of agents itself can be augmented, such as adding branches so the anigraf resembles a simple tree. In...

    • Anigraf4: Tally Machines (pp. 49-62)

      Swimmers and walkers illustrate the anigraf abstraction, being focused on cognitive aspects of knowledge and decision-making. Here we depart briefly from these abstractions to explore how agents’ voting might be implemented. For animate creatures, neurons or neural assemblies are the analogs of cognitive agents. Similarly, the tally process will also be a collection of neurons organized to aggregate inputs from various neural modules, which would be the physical analog of anigraf agents. Just as there will be advances in anigraf designs, so do we expect evolutionary advances in methods for counting votes. We follow this sequence, first with machines for...

  6. Part III: Cognition:: Agents With Beliefs
    • Anigraf5: Dancers: Mating Games (pp. 65-76)

      Goals such as “fight,” “flee,” “evade,” “approach,” or “be still” have a definite cognitive flavor. Supporting these goals are the limb movements we have characterized as run, walk, turn, halt, etc. It is useful to make explicit these two levels of description. Let us attribute the observed behavior to more cognitive, mental organisms, and use lower-level agents to describe the movements of the limbs. The external observer sees only a symphony of limb movements, and from these actions infers a creature’s intentions and goals. This inference is compelling (recall the labels naturally placed on Braitenberg’s vehicles). For example, the direction...

    • Anigraf6: Planners: Event Sequencing (pp. 77-84)

      A dance requires choreography. For humans, we create a variety of dance sequences from a rather limited set of limb movements. However, for many fish, birds, and other lower animal forms, there are “innate releasing mechanisms” with the choreography “built in” (Lorentz 1982; Tinbergen 1951). This is especially clear in mating behaviors, where consummation requires first various displays that arouse interest, then contact, and finally intercourse. Often the environment plays a key role in choosing the correct action where one act sets up the preconditions for the next. Whether the behavior is the building of a spider’s web, the burying...

    • Anigraf7: Explorers: New Worlds (pp. 85-98)

      As anigrafs become more engaged with their worlds, new relationships among events will be discovered. Some of these will involve inanimate objects and actions; others may follow from interactions with new species of creatures, perhaps with different embodiments, and hence with different internal models. These new relationships will require new, or revised, models that are appropriate for the context. Two options are obvious: edges can be added or deleted; or, nodes can be created or destroyed. Both come at a cost. If edges are revised, then the altered anigraf, appropriate for the new context, may not have the proper form...

    • Anigraf8: Alliances: Coordinating Diversity (pp. 99-112)

      Consider a society of identical anigrafs. Then, unlike the connectivity shown in figure 8.0, the social map will be the complete graph. However, with such homogeneity, there will be accurate, intelligible communications among members of the society. Sacrificed will be the ability of the group to see the world from different vantage points. Behaviors will be very predictable. Clearly, the flexibility and adaptability of a society depends upon its members possessing a range of talents to execute a host of different tasks (Page 2007).

      Fortunately, given even minimal environmental pressures, offspring are not carbon copies of their parents. As the...

  7. Part IV Metagrafs (pp. 113-124)

    Metagrafs are relationships between anigraf models. Examples would include how one model may be transformed or decomposed into another, or similarity relations between graphs. The simplest such relationship is the identity, when one graph is the same as another, but the nodes have different meanings in different contexts. Metagraphs have the potential to point out new anigraf designs by revealing different classes of graphical forms (e.g., see Gunkel’s networks). For advanced creatures, a metagraf can also provide an anigraf with self-induced insights. When one graph has the same form as another, but each are in different contexts, with different labels...

  8. Epilogue (pp. 125-128)

    Superficially, anigrafs may appear to be simple enhancements of Braitenberg’s vehicles, expanded to include aspects of cognition. This similarity however should not obscure the fact that anigraf mechanisms are quite different from vehicle servo-feedback systems. These differences in design favor mimicking behaviors set in quite different contexts. Anigrafs explore the consequences of viewing living creatures as a society of agents, with beliefs and preferences captured by relationships of a graph. These differences open new windows to understanding mind, revealing computational complexities, relationships, and structures not previously considered.

    The reality of “mind” nevertheless remains elusive. Considering mental events as part of...

  9. Bibliography (pp. 129-134)
  10. Appendix: Phase Plots (pp. 135-138)
  11. Glossary (pp. 139-146)
  12. Index (pp. 147-148)