Thursday, August 30, 2007

Miller, "What Can Automation Tell Us About Agency?"

Miller, Carolyn. "What Can Automation Tell Us About Agency?" RSQ 37.2 (2007): 137-157.

Miller questions the general apprehension teachers have toward the machine scoring of rhetorical performances (i.e., compositions), both oral and written. She begins by accounting for the expanding adoption of automated assessment where computational processes "read" texts and return statements about the text's efficacy that are then used to assess the text. Miller has a bit of fun with this idea, announcing a fictional "new product of interest to rhetoricians" called AutoSpeech-EasyTM, a "computer system for the automated assessment of oral performance in public speaking" (139). Miller refers to AutoSpeech-EasyTM as a thought experiment. She builds up the application, elaborating its many technological advancements, and then conducts an informal survey of 25 people to consider their reactions to the promises of the software. The survey responses represent a range of responses that suggest some of the ways that "automated assessment systems create a situation in which Burkean symbolic action directly confronts nonsymbolic motion in the form of the machine" (140).

Miller deals with this direct confrontation (viz., the clash of symbolic action and mere motion) carefully and with nuance, going over the results of the surveys and commenting on the most evocative statements related to "three dimensions of rhetoric that may help us understand what we want from a concept of agency" (142): performance/performativity (145), audience/addressivity (147), and interaction/interactivity (149). Citing Ronald Green, Miller notes that agency is deeply entangled with "a vision of political change"; this is one reason why agency is so guarded. Another is the "capacity to act"--in both the rhetor and the audience (145). Before dealing with each of the "three dimensions" in turn, Miller restates concerns about computational reduction: "The AutoSpeech-EasyTM thought experiment challenges this double understanding of agency [as a distributed capacity between the rhetoric and audience] by radically truncating the pedagogical situation, leaving the student in a rhetorical desert, demonstrating her capacities to an 'audience' capable only of motion, turning effects into algorithms" (145).

Later in the article, Miller acknowledges the limits of educational assessment as it applies to thinking about automation and agency: "In most educational situations, the possibilities of agency as rhetorical effect are artificially truncated: there is no exigence beyond educational accounting, and the teacher's role is that of a grader, not that of a rhetorical audience capable of enacting change" (148). This sense of truncation is absolutely crucial--and is as much to blame for the objections as is the computational process itself. I mean that the assessable ends of quasi-rhetorical performances in education arbitrarily constrain the performativity of the act--with or without computational assessment. Reconsidering the quotation at the end of the last paragraph, the process of turning "effects into algorithms" applies on every graded occasion, doesn't it? Miller concludes that much of the kinetic energy and possibility for ongoing action is diminished when computer-based algorithms stand in for human audiences. She writes that "out of respect for our students we should not ask them to make such attributions [of "human decency and respect"]" (153) to automatons and robotic graders. Certainly this is a question of responsibility--and adds a protectionism to the agent function where rhetoric is concerned.

One other point: I'm interested in the other places where automation and agency collide, where "effects into algorithms..." ends with an ellipsis, a breech opening to a larger sequence: "effects into algorithms[...back into effects]." I see this loop (is it a cycle? A rough-cycle.) happening with distant reading, with the computational methods put to use in a system such as CCCOA. For pedagogy, sure, automation becomes problematic; it does many of the things the respondents' intuitions suggest to Miller. Next we need to renew questions about automation-inflected rhetorics, where, nonhuman things participate in the network, activating new content, new associations, rather than truncating, reducing, or excising agency.

"Better system design with more interactivity could help bring the rest of us around to this view, as could simple habituation on our part: given sufficient experience and exposure, we may accept these machines as Latourian hybrids to which we unproblematically delegate rhetorical agency, just as we delegate the function of a doorman to an automatic door closer (Latour, "Mixing Humans")" (152).

"Research in interpersonal communication, human-computer interaction, and computer-mediated communication has suggested that we have a very low threshold for ethopoeia: in other words, it doesn't take much for us to be willing to attribute character to an interlocutor, no matter how primitive the cues are" (151).

Phrases: machine scoring (137), automated scoring systems (137), agent function (151), ethopoeia (151), Eliza effect (151), [rel. agentic shift from Milgram and Postman]

Bookmark and Share Posted by at August 30, 2007 8:10 AM to
Comments