Several authors urged that whether or not something is a computer is not a matter of objective fact, but depends on what attitude one brings to the question.

[Valerie Hartcastle:] We define computations and computers relative to some specific set of explanatory concerns. That is, it is largely a pragmatic decision whether we consider the digestive system a computational device, as is whether we consider my Mac a computational device. How we talk about things depends upon what about those things we are trying to explain.Istvan Berkeley suggested that that the notions of ``computation'' and ``computer'' are inherently ambiguous.Searle would counter that this view makes computation a purely artificial kind; that is, there is nothing out there that makes anything computational. This means that computers and computations are a different breed of concept than our real scientific notions, like force or mass or species, which presumably do exist apart from our pragmatic concerns.

My response to Searle is that he misunderstands science. Scientific theories are not formed in a vaccuum. They all are produced in a particular context and are designed to answer certain questions (with a particular contrast class) and not others. Furthermore, they are abstract idealizations of some phenomena we have observed (or would like to observe). Consequently, physicists use the term ``force'' or ``mass'' in particular contexts when answering particular questions and these terms get their meanings from those particular usages. Similarly, cognitive scientists (say) use the term ``computer'' or ``computation'' when referring to certain abstractions embedded in a particular sort of explanation, given to a particular sort of audience. In sum, what Searle complains about as the problem with the notion of ``computer'' is a ``problem'' for all scientific notions. (And hence, not really a problem at all.)

[Berkeley:] I think that this ambiguity might explain why there are conflicting intuitions about the suitability of definitions, and also explain why Searle believes that he can make the kind of claims he does.Well, maybe not. Attacks on the ``computationalist hypothesis'' by Searle and others sometimes confuse these senses. The philosophical literature has taken to using the term ``computational'' as an adjective on kinds of machine, but confuses it somehow with ``computable,'' which is a mathematical property of functions on the integers. Computability is connected with Berkeley's computationT, but arguments about what a computer can or cannot do seem to be inevitably referring to computationC. (My Powerbook can receive a Fax transmission, an ability not mentioned anywhere in the theory of computation; fax-reception is clearly computational, but - since it has nothing particularly to do with functions on the positive integers - not computable.)I am pretty certain that the term ``computer refers to a number of distinct classes of objects. Perhaps it would be useful to attempt to find a set of definitions for ``computation'' which served to capture diferent aspects of the entire class of computational entities. I certainly do not feel up to this task alone, but what I will try and do here is offer two versions of the notion of computation, which seem to lie at the extreme range of the term.

The first notion of computation, which I will call computationT, short for Technical computation, is the most restrictive notion. Entities which have the properties of a Turing machine or perhaps a von Neumann machine would fall within the scope of this notion. An entity is computationalT if it engages in the rule-governed manipulation of complexes of structured symbols (this is vague, I know - please fill in your own favorite list of properties, as appropriate). Many connectionist systems might fail to be computational in this sense (assuming that the claims made in the standard connectionist literature are correct, something which I am not too convinced of). Similarly, it is far from clear whether biological cognition would be an instance of computationT (Pylyshyn would probably say ``yes,'' but others would say ``no''). On this restricted notion of computation though, it is clear that Searle and Putnam's Wordstar-instantiating walls and computational stones would be ruled out.

There is another notion of computation which is also common though. This I want to call computationC, short for common sense computation. This notion of computation is about as broad as is possible. Roughly, something is computationalC if it is something which can be done by a computer. So, to cite a ridiculous example, playing a Bach fugue would count as being computationalC, as a machine with a sound card, CD drive etc., can play a Bach fugue. Pretty clearly this notion of computation is way too broad to be of much use for anything serious, as it makes many too many things ComputationalC.

Until we can say what a computer is, computationalC is meaningless, unfortunately. Which brings us back full circle to where we began. Also, it is not obvious that computationT rules out the Searle-Putnam examples. The final point of discussion is whether any of these proposed definitions manages to avoid Searle and Putnam.

Fri Jul 25 22:00:35 MEST 1997