Friday, November 21, 2003

Reading:: Social Studies of Science v22 (1993)

Originally posted: Fri, 21 Nov 2003 09:02:28

Social Studies of Science v22 (1993).

A colleague has been working on a task modeling technique, and when he described it, I thought about an article I had just read by Bruno Latour, Philippe Mauguin, and Genevieve Teil. So I recommended it. He was interested enough that I went back yesterday and reread the entire special issue so I could have an informed discussion with him. The approach and the issue are both very interesting. They point to a direction that I think Latour explored and ultimately abandoned -- at least I haven't seen any trace of it since.

That direction is what Latour calls a "quali-quantitative" one. In "A Note on Socio-Technical Graphs," Latour, Mauguin, and Teil describe "a visual and conceptual space" that -- they emphasize -- explicitly works against the social vs. technical factors that are often described in social studies of science by introducing a dichotomy that explicitly cuts across those factors. Social vs. technical is a false division, they say (reasonably, to my mind), so the new dichotomy serves to jolt us out of its use. This visual-conceptual space, meant to map scientific controversies by examining statements of all parties, draws on the notions of association and substitution -- or in the linguistic terms used here, syntagm and paradigm -- for the mapping structure. These two notions serve as the axes for a two-dimensional graph in which narratives are analyzed both quantitatively and qualitatively. The graphs were implemented in HyperCard, a program that researchers in Europe had gone nuts over. (For instance, at about the same time Latour and his colleagues were connecting these graphs, Bodker and Gronbaek were using HyperCard for cooperative prototyping in Denmark.)

Now, these graphs were specifically developed to analyze narratives in science and technology studies, not observations or other naturalistic collection methods and not other sorts of narratives. So I'm not sure how well they travel. But in the framework of STS, they do really interesting things. For instance, Latour et al. suggest that they can be used to identify black boxing: if a series of actants stays together through successive versions of narratives without "defecting," then "they may be aggregated in a black-box and given either a new name or the name of one of the actants" (p.41). Black boxes are a staple in actor-network theory, so being able to identify black boxes across many narratives at a glance -- or better yet, being able to automate this identification -- is a real boon.

Just as the graphs can be used to identify coherent assemblages across accounts, they can also compare contradictory accounts. This allows us to avoid the essentialism in functionalist accounts by identifying divergences without making a priori distinctions about which ones are useful (right, reasonable) and which ones are meaningless (wrong, irrational).

Near the end of the article, on p.45, Latour et al. have a little dialogue with one of the reviewers. The reviewer objects: but explaining success can't just involve counting the actants involved and assuming that the longer network will be the stronger one. He gives the example of American Bell, a relatively small company that ultimately prevailed over Western Union with the help of a "small and unassailable set of patents." So, the reviewer asks, don't you need to account for other, more qualitative factors? Latour et al. reply that yes, the longer network did win: American Bell used its patents to align itself with a vastly longer network, the legal system. This argument is a neat trick, but I want to emphasize that it's an escape hatch: as far as I can tell, there really is no way to count the number of actants that make up a network. Just as Latour et al. lift the curtain to reveal the entire legal system standing behind American Bell, his reviewer could choose to lift another curtain to reveal other networks standing behind Western Union. And those competing networks, if extended enough, will eventually overlap and actants will find themselves oscillating between them! The "longer, stronger network" argument is very difficult to quantify, even with these socio-technical graphs. Particularly because Latour et al. do not give any methodological guidelines for coding narratives!

This article is followed by responses. The first is by James K. Scott, who had used these graphs in his own work and gives us sort of a testimonial about them. He makes the point that the socio-technical graphs are inscriptions -- ways "to reduce (translate) complex processes to features that can be graphically represented in two dimensions" (p.60) -- and that actor-network theorists were thus using the same strategy for success that they had so often witnessed in use by the scientists they had studied. He goes into some of the methodological details about coding that Latour et al. had avoided, and he supplies an example from his own research. He also includes a number of objections, including the objection that this method doesn't provide any way to discriminate among the texts (narratives) to analyze.

More strenuous objections come from W. Bernard Carlson and Michael E. Gorman. The first author is unmasked here as the unnamed reviewer whose argument Latour et al. had dismissed. Like Scott, these authors remind us that many texts are after-the-fact reconstructions (often dubious narratives); sources are heterogeneous and difficult to code in a uniform manner, particularly visuals; no criteria are discussed for what to include or exclude on the graphs; and finally Latour et al., in their opinion, did not answer their concern about counting actants. In the place of socio-technical graphs, Carlson and Gorman push their own cognitive mapping scheme, which attempts "to map the mental processes" of the actors (p.87).

If you've read much Latour, you can guess what happens next. In his own reply to Carlson and Gorman, Latour heaps scorn on the notion of mapping mental processes. "Either the authors have access to completely new types of documents unknown in France which allow them to directly observe mental processes, goals, and details, or they have simply misunderstood the whole argument of our method" (p.94). The whole argument is that socio-technical graphs allow the triangulation of accounts, not essences -- which, Latour charges, is what Carlson and Gorman are after with their attempt to map mental processes. Latour, who famously declared a ten-year moratorium on cognitive explanations, emphasizes again why this is a good idea.

Can these graphs be used to model tasks? No, not as such. But they can provide an inspiration for such modeling. And as my colleague pointed out, my own genre ecology diagrams are all about associations and substitutions too. Guess what I'm going to be working on for the next couple of years.

Note added 11/3/2004:

I said that this line of work appeared to have been abandoned. In fact, it shows up in the 1999 book Pandora's Hope.

Blogged with Flock

1 comment:

Peter Jones said...

Hi Clay

Really rich seam here! Researching paper on socio-technical structures in health informatics. I have have only recently arrived at Latour through Michel Serres. Look f/w to returning tomorrow 0:40 here. If you have any links to your ecology diagrams I'd be interested in linking at:

http://www.p-jones.demon.co.uk/linksTwo.htm

Best,
Peter Jones
http://www.p-jones.demon.co.uk/
Hodges' Health Career - Care Domains - Model
http://hodges-model.blogspot.com/
h2cm: help 2C more - help 2 listen - help 2 care