Friday 1 November 2013

Disagree with Cziko, Campbell on instructional learning

Gary Cziko and Donald Campbell famously attempted to explain instructional learning in terms of variation with selective retention. This was part of a grand scheme to show that all knowledge was the product of such processes. In Without Miracles, Gary Cziko recounts a string of cases where instructional learning turned out - on closer inspection to involve selective retention.

After some reflection about this, I think they are wrong about the issue. Instructional learning (e.g. rote learning) is usually best modeled by copying - not by processes necessarily involving selective retention. It's true that selective retention is common - and that if you examine processes closely enough, you can usually find some selective retention going on somewhere. However, to model instructional learning using copying is simple, obvious and correct - whereas to model it using selective retention seems like a case of special pleading.

We don't need a grand scheme which attributes all cases of adaptive fit to selective retention. Such a scheme is mistaken. Some cases of adaptive fit are the result of copying. Not necessarily copying results of previous processes involving selective retention - just plain copying.

Henry Plotkin briefly makes much the same criticism in Evolutionary Worlds without End. I think he is correct - and that Cziko and Campbell were wrong. Shoehorning instructional learning into models involving selective retention is an "unhelpful" activity that evolutionary epistemology should distance itself from. Instructional learning is mostly just copying, which is perfectly compatible with modeling individual learning within Darwinian frameworks.


Update 2015-05-25:

I've changed my mind on this a bit. See: Selection is simple, general and explains goodness of fit for the details.

It is possible to see all copying as a form of selection. Maybe this is a bit contrived. It certainly depends a lot on the details of the definition of the term "selection" that you use. However, classifying copying as a form of selection is possible - and it helps to explain how Cziko and Campbell's position can be internally consistent.

2 comments:

  1. Thanks, Tim, for posting these welcomed insights!

    As a resolute selectionist - my own thinking, similarly to Cziko's and Campbell's, grounded in cybernetics as explaining the principle of trial-and-error(the eliminating of erroneous trials and selection of more successful ones), which appears to be a defining feature of biological systems at all organizational levels - I have to ask how, on any alternative basis, there could have evolved such a thing as the human brain, possessing not merely the capacity (i.e., requisite skills) for copying an unlimited diversity of entities, patterns, context-dependent relationships, abstract symbolic thought-processes and methodologies etc., etc., but moreover, the capability of selecting from that range what it predicts it needs to copy for furthering its biological survival. That, as I see it, pinpoints the exact argument that Campbell and Cziko were seeking to demonstrate with their skillfully selected, copied illustrations. Personally I cannot think of any way that any biological organism could circumvent the trial-and-error principle by substituting some other, apart from selecting to destroy itself. Can you?

    Best regards,
    Richard Traub.

    ReplyDelete
    Replies
    1. The issue I was discussing was not really about the evolution of the human brain. It was whether instructional learning (i.e. broadly copying) could be usefully modeled as a form of selectionist learning (i.e. trial-and-error). In learning theory social learning and individual learning are usually separated. Similarly in evolutionary biology copying and selection are normally disjoint primitive operations. We can break this down into two issues: are there definitions of "selection" for which selection can explain copying, and are those definitions of selection ones which we should adopt. It seems as though there's a useful distinction to be made between finding new knowledge and borrowing existing knowledge. It isn't clear whether it is helpful to choose terminology that obscures this distinction.

      Delete