Clusters without cluster concepts

Philosophers try to understand things. One traditional way that goes is to provide a definition. Q: What is knowledge? A: Something is knowledge if and only if it is justified true belief. K=JTB.

Attempts to give necessary and sufficient conditions in this way typically fail. JTB is insufficient, and a literature pops up suggesting some additional condition, so that K=JTB+X. Some later philosophers pass quickly over the difficulty, not bothering to fill in a value for X and suggesting that for their purposes we can just focus on the JTB part.

A popular alternative since the mid-1900s has been to analyze in terms of cluster concepts. When X, Y, and Z all seem relevant but none seem either necessary or sufficient, say instead that the concept is a cluster formed by those criteria. This has been a common move for analyzing art.1

In work with Christy Mag Uidhir, I’ve argued that the cluster concept account is a mistake.2 Here’s a different way of seeing the problem.

It is common to think of a cluster concept as being like what’s going on in a connectionist network. Each element in the cluster has a weight determining how much contribution it makes to concept application. Given a stimulus, the network has a total activation for the concept.

We might imagine a network trained to recognize things. Looking at the activation of the output node for ART, it returns .99 when presented with the Mona Lisa but only .73 when presented with Duchamp’s Fountain.3

This is the wrong model, because the cluster of features does not have static weights. Instead, the cluster is a resource. In disputes over whether this or that is art, disputants will appeal to one or another feature. Appealing to a specific criterion is rhetorical move in giving and asking for reasons. It is an argument that this ought to matter, not a report that it simply already does matter. So each criterion has complex connections to reasons and consequences, but not a pre-existing weight.

What holds the various elements of the cluster together is not that they are all organized and stratified in a concept. Instead, there is a social convention of reasoning with elements in the cluster when using the word. The thing that makes the plurality of art concepts all count as art concepts is only that they live under the umbrella of “art” in our discourse.

  1. It is usually but somewhat erroneously traced back to Morris Weitz. He argues that “art” cannot be defined but doesn’t articulate a cluster concept. Instead, he says that we recognize new instances of “art” based on similarity to exemplary art works.
  2. See Art Concept Pluralism and its sequel.
  3. This is an idealization, because the relevant inputs for art judgements include function and social context. These aren’t readily coded as inputs to a connectionist network.

Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.