@Article{,
author = {Brian V. Funt},
title = {Problem-solving with diagrammatic
representations},
journal = {Artificial Intelligence},
year = {1980},
OPTvolume = {13},
OPTnumber = {3},
OPTpages = {201--230},
}
Whisper is a traditional AI problem solver that can request observations from a diagram. It can also make changes to the diagram. The domain is blocks world. It knows about stability and falling objects. It can visualize something rotating in the diagram and determine when it will hit another object. The goal is to move all blocks til they are stable. It moves them, then simulates in the retina til finished.
The retina in Whisper has best resolution at the center. It can focus on different parts of the diagram. [p37] The retina is made up of bubbles that are grouped as rings and wedges. Bubbles also communicate with their nearest neighbors. The retinal supervisor tells the bubbles what to process. [38]
The perceptual primitives are: find the center of a shape, find the points of contact between a shape of one color and the shape of another, examine curves for abrupt slope changes, test a shape for symmetry, test the similarity of shapes, and visualize the rotation of a shape while watching for a collision with another shape.
To rotate, all bubbles with the object in it ask the next wedge to turn on, then turn themselves off. Collisions are detected when you ask a bubble to turn on with object 1 when it's already on with object 2.[40] In this work, the diagram is a 2d square of pixels that the retina can focus on. It produces a series of snapshots as its solution. Things do not fall simultaneously in Whisper.
Limitations: no model of velocity (which would require equations), no simultaneous motion,
It also implements sliding.