The Network Phenomenon: Empiricism and the New Connectionism
Stephen Downes, 1990
(The whole document in MS-Word)
TNP Part II Previous Post
III. Three Objections to Empiricism
A. Objections to Associationism
There are three essential objections to empiricism. The first objection is that associative principles are not sufficiently powerful to explain human cognition. The second is that there is no means to distinguish input from other cognitive phenomena. And the third is that associative inferences can never be justified. I will discuss each in more detail and show what I need to prove in order to meet the objection.
The first objection always has the following form: "Human beings can know or do X, no associative system can know or do X, therefore, humans use something other than associative principles."
A paradigm example of such an objection mentions humans' use of abstractions. Consider the following example.  Suppose we had to determine whether a string of letters is a well formed formula (wff) in a language L. Language L is a "mirror image" language; only strings the two halves of which are mirror images are wffs in L. In order to distinguish wffs in L from non-wffs in L, it is necessary to employ an abstract term. Since it is impossible for an associationist principle to employ an abstract term, then the principle we employ to distinguish wffs and non-wffs cannot be an associaionist principle.
Another example may be found in Leibniz.  While we need experience to suggest to us universal eneral principle principles, we cannot derive these principles from experience. For experence consists entirely of particulars, and no set of particulars is ever sufficient for the derivation of a universal. Therefore, we must emply some means other than experience in order to derive universal general principles.
These examples could be multiplied, but they give the general idea. Since the argument is valid, then the only means of answering such objections is to deny either (a) that human beings can know oe do X, or (b) that associationist principles are insufficient for X. In general, I take the following approach. If the claim is that we know X, then I deny 9a). If the claim is that we do X, then I deny (b). Classical scepticism is all that I need to deny (a). The real challenge lies in the denial of (b).
B. Theory-Laden Perceptions
The second objection to empiricism reaches the conclusion that, since there is no means of distinguishing perception from other aspects of cognition, it follows that we cannot say that other aspects of cognition have their origin in perception.
The premise is well supported, For example, Quine argues that we cannot distinguish between the analytic (for example, formal rules of inference) and the synthetic (empirical content).  A similar point is made by Hanson. According to Hanson, what we see is affected by what we believe, that is, all our experiences are "theory-laden".
Since the premise of this argument is well supported, the only means of responding to this argument is to show that the conclusion that there are no pure perceptions does not follow from the premise. There are two ways of stating the premise. I will consider each in turn.
The first way is to state the premise is to say that perceptual terms are theory-laden.  The second is that perceptions themselves are theory-laden. The first formulation I embrace, since all aspects of language are theory-laden. Language depends to a large degree on rules and categories, and these, I believe, are theories. However, it does not follow from this premise that there are no pure perceptions, since perceptions are distinct from descriptions of them.
The second way of stating the premise is to the effect that perceptions themselves are theory-laden. I can agree that at some level, perceptions are theory-laden. This is a natural and expected consequence of the theory of learning which I will propose below. If it is true that perceptions are always theory-laden, then the conclusion, that there are no pure perceptions, follows. So, in order to show that this conclusion does not follow, I will need to show, first, that there is some level of perception that is not theory-laden, and second, there was some point in time at which no perception is theory-laden.
This will not be an easy task. It is arguable, for example, that even if there are pure perceptions, they cannot be used unless combined with some theoretical structure or another.  It is also arguable that in order to perceive objects in three dimensions, some higher-level built-in constraints are required.  Therefore, in order to meet this objection, not only is it necessary to show that perceptions are pure at some level at some time, it is also necessary to show that these perceptions are sufficient for all other cognitive activity.
The whole idea behind justification is that of distinguishing between right and wrong (correct and incorrect, justified or unjustified) inferences, and the third objection to empiricism is that it cannot distinguish between right and wrong inferences.
The ground for this objection is that associationism does not distinguish between truth-preserving infrencs and other sorts of inferences. For example, suppose we adopt a causal theory of cognition (following, say, Armstrong and Goldman). Some causal interactions, for example, the triggering of relays in a computer, are truth-preserving. Others, for example, a bat striking a ball, are not.
What is needed, the objection continues, is a formal representation of the sort of causal interactions which are truth-preserving, in order to distinguish fropm those which are not. This representation must exist at a level over and above mere physical instantiation. An example of the sort of principles that we require is the set of rules of logical inference.
In response to this argument, I will argue, first, that the distinction between right and wrong inferences is sufficiently drawn by the concept of relevant similarity, and second, that associationist systems allow only inferences which preserve relevant similarity. Therefore, associationist systems, by the fact that they are associationist systems, provide sufficient justification for their conclusions.
An objection to this conclusion may state that there is a rather large difference between truth and similarity, and thus one cannot equate a mechanism which preserves similarity with a mechanism which preserves truth. Therefore, associationist systems do not provide sufficient justification for their conclusions.
The reason why I believe that justification must be defined in terms of similarity are complex. They will be discussed below. At this point, however, may I say that, if it is indeed the case that justification can be defined in terms of similarity, then the third objecyion can no longer be sustained.
TNP Part IV Next Post
 T.G. Bever, J.A. Fodor, M. Garrett, "A Formal Limit of Associationism," from Verbal Behaviour and General Behavour Theory, T.R. Dixon and D.L. Horton, editors. Prentice-Jall, 1968.
 New Essays Concerning Human Understanding.
 W.V.O. Quine, "Two Dogmas of Empiricism", From a Logical Point of View.
 N.R. Hanson, Patterns of Discovery.
 Eg., Paul Churchland, "Two Kinds of Evidential Bias". I have only a manuscript of this.
 Lawrence Bonjour. The Structure of Empirical Knowledge.
 Marr, Vision. See also Phillip Johnson-Laird, The Computer and the Mind.