Since the cognitivist revolution around the 1950s, it’s become commonplace that cognition involves computation/information processing. The two terms are generally used more or less interchangeably.
But it seems to me that “computation” and “information processing” mean two clearly distinct things. Paradigmatically, computation is the processing of digits according to appropriate rules, whereas information processing is presumably the processing of information, i.e., the processing of signals that carry information about a source, where carrying information means raising the probability that the source is a certain way.
Given the above, computational inputs and outputs may or may not carry information; information processing may or may not be done by means of computing.
Of course, the terms “computation” and “information processing” are used in all kinds of ways, and under some usages, their meanings may well coincide. But in light of the previous comments, it still surprises me that they are used interchangeably so often and seemingly without a second thought. I suspect the reason for this conflation goes back to the cybernetics movement and their effort to blend Shannon’s theory of communication (which measured information) and computability theory (as well as control theory). The cybernetic effort was especially influential on psychology and AI, and later on neuroscience. I’m not saying that the cyberneticians conflated computation and information processing. They were clear on the difference between the two. But they established “computation” and “information” as buzzwords that belonged together in a theory of cognition, and after them, many people stopped paying attention to the difference between the two.
Any thoughts on this? Does anyone know of previous discussions of the relationship between computation and information processing? I don’t remember ever reading anything explicitly about this.
Computation vs. Information Processing
9 Comments
Comments are closed.
Computation and information processing are probably used interchangeably as some sort of generic notion. They are different when you track them back to the technical work of Turing, Post, Church, et al., in the 1930’s and thereafter, and to the technical work of Shannon and Weaver, et al., in the 1940’s. So, for example, if you say that computation is symbol manipulation, that’s not the kind of technical definition found in any branch of recursion theory or automata theory. Maybe it’s the kind of thing you would find on the first page or so of informal introductory exposition.
Now, while one might want to give pride of place to the technical notions, in truth, for many philosophical and psychological projects, those technical notions are not in play. (For many where one might have thought them in play, they are not in play.) Instead, it is only particular features of some computational formalisms that are in play. So, for example, in the Chinese room debates, the technical notion of a Turing equivalent computational formalism is not really crucial. The central idea is that syntactic manipulation is not supposed to give the symbols meaning. Evidence for this reading is that Searle was pretty happy to reformulate the problem in terms of the Chinese gym.
Some understandings notwithstanding, the TM-equivalent formalisms are also not at stake in the systematicity arguments, nor at least one reading of the competence performance distinction, nor in the Searle/Putnam ideas regarding pancomputationalism.
Hi Gualtiero,
I think this is a good and interesting question. I’ve long had a similar worry to the one you express.
One thing that I’ve read that helps tie together the notions of computation and info processing is Richard Feynman’s Lectures on Computation. I don’t have the text on hand, but my recollection is that part of it has to do with regarding computation and the utilization of a memory as the transmission of a signal over time.
The argument appears me very interesting. I believe that the origin of the confusion of levels, is in the supposition that for processing information, the information must be in a digital code. Then it seems that “computation” and “information processing” are equivalent.
(Sorry for the english)
In computational neuroscience (the subfield investigating with computer models and simulations the anatomical connections between cerebral areas to know which functions are performed by which areas)is a commonplace the use of information processing and computation interchangeably.
I believe it is too difficult to reveal the historical intricacies about the synergestic influence between computer sicence or informatics, biology, neuroscience, AI, enginerring… in relation to the usages of information processing and computation.
But many authors not distinguish between computation as paradigmatically understood, and information processing as the way a signal carries information. They even use the whole catch-up expression “computational processing” when they want to refer to the workings of synapses, neurons, neural networks or even other physical devices implementing computations.
So, the confusion is total.
I’m glad to see you emphasize the distinction between “computation” and “information processing”. The difference is clear when you consider the kinds of tasks that each approach must perform, and the kinds of mechanisms that are competent to perform the different tasks. As you point out, cognitive science has often muddied the distinction.
Gualtiero,
This is definitely a distinction worth making! As to previous references, Brian Cantwell Smith mentions this in his chapter “The Foundations of Computing” in Scheutz’s ‘Computationalism: New Directions’. He distinguishes seven(!) different construals of computation, and information processing is one of them. I’m not really sure what the distinction is between all of these construals, but that’s one of them anyway.
Well, Feynman linked computation and information processing in his famous paper where he introduced the notion of quantum computing as simulating natural processes with other natural processes. Don’t remember if he was explicit on that but it was surely one of the presuppositions.
Well, what is important is not how many meanings you can introduce (and it’s really plausible to say it’s ambiguous in some situations), the point is rather what meaning is relevant for computational explanations in neuroscience, and which for computational explanations in weather modelling. These two are two kinds of explanation, one based on the assumption that the system itself processes information, and the other that the system is simply approximizable with some toy models but is not processing any info by itself.
So basically, I’d start looking at how computational explanations are used, and not at Searle’s hand-waiving about computation.
In defense of Brian Cantwell Smith, it’s not sheer number of construals of computation that interests him, but rather developing a theory of computing (and computation) that meets several (empirical and conceptual) desiderata. These are: (1) account for the full range of computational practice; (2) discharge core notions such as computation, representation, and semantics in a way consistent with pratice; and (3) provide a suitable substructure for the computational theory of mind. Part of the foundational project, then, requires disentangling the various conceptually distinct construals of computation at play in the various fields that traffic in notions of computation such as computer science, psychology, neuroscience, etc.
The plea in the last post to focus on how computational explanations *are used* is a close ally to supplying an adequate account of how, for lack of a better phrase, computation itself (rather than just computational explanations) is used. And this is a central component of Brian’s overall project. This amalgam of contemporary computational techniques, practices, machines, etc. Brian dubs “computation in the wild”. His view is that any theory of computation worth its salt, should do justice to it. Gualtiero’s recent paper in Philosophy of Science also has this desideratum of doing justice to computational practice as a stated goal.