20 Feb 2013

Andy Clark. 2.4 of Being There, “Soft Assembly and Decentralized Solutions,” summary

summary by
Corry Shores
Search Blog Here. Index-tags are found on the bottom of the left column.]
[Central Entry Directory]

[Posthumanism Entry Directory]

[Andy Clark, Being There, entry directory]

Andy Clark

Being There:
Putting Brain, Body, and World Together Again
The Situated Infant

Part 2.4
Soft Assembly and Decentralized Solutions

Brief Summary:

Machines, including the ‘machinery’ of infant learning, can better adapt to complex and dynamic situations when information processing is decentralized.


Clark will discuss soft assembly in human development. (42)

Traditional robotic programming is hard assembled, because it does not make real-time adjustments, unlike the soft assembly of human motion.

A traditional robot arm, governed by a classical program, provides an example of "hard assembly." It commands a repertoire of moves, and its success depends on the precise placement, orientation, size, and other characteristics of the components it must manipulate. Human walking, in contrast, is soft-assembled in that it naturally compensates for quite major changes in the problem space. As Thelen and Smith point out, icy side- | walks, blisters, and high-heeled shoes all "recruit" different patterns of gait, muscle control, etc., while maintaining the gross goal of locomotion. Centralized control via detailed inner models or specifications seems, in general, to be inimical to such fluid, contextual adaptation. (42-43) […]

Multi-factor, decentralized approaches, in contrast, often yield such robust, contextual adaptation as a cost-free side effect. This is because such systems, as we saw, create actions from an "equal partners" approach in which the local environment plays a large role in selecting behaviors. In situations where a more classical, inner-model-driven solution would break down as a result of the model's incapacity to reflect some novel environment change, "equal partners" solutions often are able to cope because the environment itself helps to orchestrate the behavior. (43, boldface mine)

[Previously Clark describes childhood development and how many factors in both the child and his environment are equal partners in guiding its development.]

Pattie Maes invented a way for machines to determine among themselves how to distribute jobs, rather than having a centralized system handle all the data and make that decision. Each machine when creating a job asks the other machines to estimate how long they would take to perform it, and the machine most able given its current abilities and activities gets the job. Job scheduling then becomes an “emergent property” of the simple machine self-assessment and communication behaviors. (43d)


Soft assembly out of multiple, largely independent components yields a characteristic mix of robustness and variability. The solutions that emerge are tailored to the idiosyncrasies of context, yet they satisfy some general goal. This mix, pervasive throughout development, persists in mature problem solving and action. Individual variability should thus not be dismissed as "bad data" or "noise" that somehow obscures essential developmental patterns. Instead, it is, as Thelen and Smith insist, a powerful clue to the nature of underlying processes of soft assembly. (44a)

Thelen and Smith give the example of the development of child reaching behavior, where the factors and events leading up to the learned behavior vary widely between children even though the resulting behavior is similar for all. (44b)

One child began with fast flapping then dampened it. (44bc)

Another had to increase lift. (44c)

Other children exhibited other variations. The central nervous system is merely working with the physics and mechanics of the seemingly somewhat autonomous parts of the body that come to be adjusted. (44-45)

Clark writes:

the job is to learn to modulate parameters (such as stiffness) which will then interact with intrinsic bodily and environmental constraints so as to yield desired outcomes. In sum, the task is to learn how to soft-assemble adaptive behaviors in ways that respond to local context and exploit intrinsic dynamics. Mind, body, and world thus emerge as equal partners in the construction of robust, flexible behaviors. (45a.b boldface mine)

Andy Clark. Being There: Putting Brain, Body, and World Together Again. Cambridge, Massachusetts/London: MIT, 1997.

No comments:

Post a Comment