The Network Phenomenon: Empiricism and the New Connectionism
Stephen Downes, 1990
(The whole document in MS-Word)
TNP Part I Previous Post
A. What I Mean By Empiricism
When I speak of "empiricism" I wish to make it clear that I am not discussing logical positivism or other contemporary theories which have been described as empirical. Rather, what I mean has a much closer affinity to the philosophies of David Hume and John Stuart Mill. To employ Hume's terminology, what I wish to assert is that all ideas are copies of impressions. Modern terminology demands a more precise definition.
We may identify three distinct levels of human cognition.  The first, or lowest level is the "hardware" level, that is, the physical structure in which cognition occurs. The second ;evel is the "software" level, that is, a set of rules or procedures which govern cognitive processes. Third, there is the "data" level, which contains the contents of cognition, for example, mental representations.
Empiricists assert that all content at the data level has its origin in experience. What this means is that all content must have been, at some time or another, input through one or another of the senses. The contemporary dispute between empiricists and other philosophers concersn the origin of the software level. Empiricists believe that these rules or processes are learned, while other philosophers believe that they are innate or otherwise directly intuited.
Insofar as we are talking about formal rules, for example, the rules of logical inference, grammar, or mathematics, then I am in agreement with the empiricists. However, I believe that these rules belong to the data level. I think that they describe and do not govern. The rules or principles which actually govern cognition are of a different type: they are associative, not formal, principles. In this way, I believe my thesis differs significantly from contemporary or positivist forms of empiricism.
Like Hume, I believe that human cognition is governed by human nature. Thus, I believe that the associative principles which govern human cognition are a part of, or instantiated in, human nature. Therefore, according to the sort of empiricism I am proposing, instead of there being three levels of cognition, there are only two lvels: the hardware level, which contains the associative principles which govern cognition, and the data level, which contains the content of cognition including descriptive 'formal' principles.
Why am I the sort of empiricist I say I am? First, I do not believe that human cognition is governed by formal rules. Otherwise, we would never be able to break these rules, and there is substantial evidence that we can. So I believe we are not governed by formal rules, and further, I believe that these rules cannot be innate.
And second, even were we governed, innately or otherwise, by formal rules, I would argue that we should not be. For formal rules are abstractions, and while abstractions are useful, they are insufficient to respond to sufficiently complex phenomena. As Wittghenstein  points out, we can always find an exception to a rule, and we need to be able to respond effectively even in the case of an exception.
B. Association, Rules and Categories
Before the advent of logical positivism, empiricists such as Mill and Mach argued that general principles, such as formal rules or laws of nature, are summaries of previously experienced phenomena.  In my opinion, this view of rules is correct. Rules, as Wittgenstein argued, describe, and they do not prescribe.
Generalized descriptions such as rules or laws of nature may be derived employing associative principles. Simply put, the idea is that, when we observe a sequence of similar events in which two things go together, we generalize and say those things generally go togethr.
There is in my mind a close link between rules and categorizations. When we place two things into a category, we are saying that those two things are similar in some way. We use categories in order to generalize. If most members of a given category are associated with something, say, some sort of behaviour, we tend to associate all members of the category with that behaviour.
The concept of "similarity" is central to empiricism, for all association and categorization depends on similarity. By "similarity", I wish to emphasize, I do not mean identity of a set of observational predicates. In my opinion, similarity is a pre-linguistic concept. below, I will say that two things are similar if they have sufficiently overlapping vectors.
The entire principle of associationism may be defined by the following paradigm example: "dpgs are similar to cats, dogs are associated with fur, therefore, cats are associated with fur." While this may seem to be a very weak principle, once it is recognized that the 'cat', 'dog' and 'fur' in the example can be anything, for example, '1101', '1100', and '0001' respectively, then we can see that this is a very powerful principle.
TNP Part III Next Post
 I use the term "levels" in much the same manner as Pylyshyn, Computation and Cognition.
 On Certainty. "A rule is shewn by its exception."
 John Stuart Mill, A System of Logic, and Ernst Mach, The Analysis of Sensations.
 Ludwig Wittgenstein, Philosophical Investigations.