SCIENCE@home

IMAGE AND SUBSTANCE

Kenneth D. Forbus

Published 1993 in International Conference on Climate Informatics

ABSTRACT

Imagery is an important psychological phenomenon. However, to write a paper whose “primary goal is to bring the imag,ery debate to the artificial intelligence (AI) community” ignores the fact that the imagery debate has included the A1 community all along. The citations of work by Funt, Hayes, Hinton, Man, Minsky, Simon, and Ullman within the article itself-all quite respectable A1 researchers-are evidence that this goal was accomplished long ago. In fact, developing a computational understanding of imagery has been a motivation of many A1 efforts in spatial reasoning. For example, the Metric DiagramlPlace Vocabulary model of spatial reasoning was formulated to account for the use of perceptual-like operations in spatial reasoning (Forbus 1980; 1983; Forbus et a f . 1987; 1991). Most of the work on spatial reasoning in qualitative physics attempts to replicate the powerful spatial reasoning abilities of human engineers (Davis 1987; Faltings 1990; Forbus et al. 1991; Joskowicz 1987; 1989; Joskowicz and Sacks 1991; Kim 1990; 1992; Nielsen 1988; 1989; Yip 1991). Glasgow does a good job of pulling together literature on imagery. Interwoven with this survey are claims about Glasgow’s particular array representation as a computational model of the psychological phenomena of imagery. Her arguments against sparse propositional representations as a model for imagery are reasonable, but the case made for her array theory is weak. She appears to conflate using propositional structures for representations with using theorem proving methods for reasoning. (Without this conflation, it would be hard to argue that sparse arrays are not computationally equivalent to simple propositional structures.) There are many other alternatives, including the use of mixed propositionaI/numerical representations (Hinton 1979; Forbus 1980; 1983). The choice of which representation best accounts for the psychological data remains an open, empirical question. Computational arguments can be used in several ways in psychological modeling. Glasgow sets up two competing models and attempts to distinguish between them. Another way is to determine limits on the applicability of models by demonstrating computational constraints on the nature of the task being modeled. The rest of this note makes such an argument, based on evidence froin qualitative physics.

PUBLICATION RECORD

CITATION MAP

EXTRACTION MAP

CLAIMS

  • No claims are published for this paper.

CONCEPTS

  • No concepts are published for this paper.

REFERENCES

Showing 1-17 of 17 references · Page 1 of 1

CITED BY

Showing 1-7 of 7 citing papers · Page 1 of 1