Perception, models, and data: Some comments

Abstract

Some comments are offered on the papers given at the conference, which are divided into three groups: visualperception, models and neural networks, and data analysis.The analysis stresses the complexity faced by scientific theories in each of these three areas, and consequently why the demand for computing capacity will continue to increase, with no practical bound in sight.I have divided the papers on which I comment into three groups: visual perception, models and neural networks, and data analysis. VISUAL PERCEPTIONI put the four papers on visual perception in the order in which we heard from James Cutting (1997), Daniel Kersten (1997), Mary Kaiser (Kaiser & Montegut, 1997), and Jeffrey Mulligan (1997. These are the most hopeful four papers I have heard in psychology in a long time. They show how many things there are under way and yet to be done.Anybody can see a need for high-performance computing in psychology by considering how much of the cerebral cortex is taken up by the visual system. It is a complicated, big system, from a computational standpoint. There will be no end to what we want to try to do. One concept I want to introduce in reflecting on James Cutting's paper, but on the others as well, is this: We should think of the processing in the visual cortex as being computationally irreducible. It is a nice concept that was, in fact, introduced by the father ofMathematica, Steve Wolfram(1985). He introduced it in the context of cellular automata.If we take, for example, the beautiful classic case of two bodies interacting by gravity, then we have a wonderful computationally reducible theory. We do not have to wait to see where the bodies are going to go (e.g., the sun and the earth). We have a good theory that predicts accurately, and we can compute very quickly. This case is computationally reducible. But, for many cellular autoCorrespondence should be addressed to P. Suppes, Department of Psychology, Ventura Hall, Stanford University, Stanford, CA 94305-4115 (e-mail: suppes@ockham.stanford.edu). mata, we are not going to do better than run the system. The kind you actually run is a direct simulation of cellular automata, in order to know what it is going to do, what the configuration looks like after, say, 10,000 trials. The conjecture is that this may be true of the visual system in many essential ways. I will discuss this point more later in this paper.This conjecture implies that we will have an insatiable appetite for computing. There are some differences here about how much we believe in extensive computing. It seems to me that these four papers, each in its own way, demonstrate very well the extraordinary desirability of having high-performance computing in psychology, if you are concerned with the details of visual systems at all, in any kind of elaborate way.Let us compare the computational demands of mathematical proofs. From a computational standpoint, the mathematical cases are all rather trivial. For example, we all know something about the proof ofthe four-color theorem, which involv...

Commentary on high-performance computing and psychological models: Theory-building and model-fitting processes

“…An attendant problem, as Suppes (1997) notes, is that there seems to be a lack of ready-to-use modeling software, especially in the behavioral sciences. As a consequence, those who fit models to data somehow end up repeating a great deal of what has already been explored.…”