**Mail folder:**SRKB Mail**Next message:**Tom Gruber: "KR Standards meeting trip report"**Previous message:**Tom Gruber: "frames and logic"

Date: Fri, 22 Jun 90 17:28 PDT From: bwilliam@parc.xerox.com Subject: Function vs Behavior To: shared-kr@sumex-aim.stanford.edu Cc: bwilliam@parc.xerox.com Message-id: <19900623002842.0.WILLIAMS@SHERMAN.parc.xerox.com>

Read on only if your interested in Function vs Structure During last meeting's discussion of function and behavior, one proposal was that they are not that dissimilar concepts, and that both are describable within roughly the same language. Coincidentally, that morning I completed the final version of a paper for the automatic generation of abstractions/approxiamtions workshop that elaborates on essentially that position. This might be of interest to the working group as a simple but concrete example. If anyone is interested in the paper I would be glad to give them a copy. Here is a brief description of the approach presented, within the context of extracting function from behavior: The task described in the paper is to automatically generate a teleological account of how a device achieves a behavior of interest (intended, faulty or whatever). The class of devices studied are those commonly explored in the qualitative reasoning community --- continuous devices whose behavior is described qualitatively (specifically in terms of regions of the device's state space). Specifically, the behavior and function of components are both described as equations in a qualitative/quantitative algebra called Q1. The perspective of the paper is that the function of a device's component is described exactly by those features of the component's behavior that contribute to the overall device's behavior of interest. Additionally a device "works" by establishing a network of local interactions (described by Q1 equations) that are produced by the components and connections. The problem then is, given a qualitative description of a behavior of interest (e.g., intended behavior) and a quantitative description of the component behaviors, construct the most abstract description of the component interactions that allow the behavior of interest to be derived. The approach is demonstrated on a very simple fluid regulation device. (Note that this can also be viewed as modelling approach, where the task is to construct the most abstract model of a device that is sufficient to answer a set of questions of interest.) The following is the paper title and abstract. Capturing How Things Work: Constructing Critical Abstractions of Local Interactions. Central to most scientific and engineering tasks, such as design, diagnosis, analogy and design capture, is a representation of a device that captures its salient features with respect to how it works. We take the perspective that a device works by constructing a topology of interactions between quantities. To construct a parsimonious description of an interaction topology we introduce the concept of a critical abstraction --- a most abstract description of a topology relative to a set of queries, that preserves the link beween the individual mechanisms of a device, and the behaviors mentioned in the queries. We present a constructive modelling technique for computing critical topologies using only the mathematical properties of the interaction topology representation. This approach provides a foundation for more sophisticated teleological reasoning techniques.