Is it possible that if McLaren, Scheuer, & Miksáko had been aware of the Explanatory Learner Model, it would have influenced their research model? Possibly, but it certainly should have, for a few reasons.
First, McLaren et al put forth a model that placed the emphasis on prediction over the learner, and what could not be predicted was eliminated from Argunaut, even if those eliminations were intrinsic or essential to student discussions. "Explanatory models do not just provide accurate prediction, but also offer actionable insights that may better advance both learning science and educational practice" (Rosé, et al., p 2943); to do so, they bring together an interdisciplinary team that spans the engineering, learning sciences and technology with an eye on the learner, learner process and instructional context.
Second, "if AI-based analytics are to have an impact on education, developers must attend to the human interaction and social context of using technology" (Rosé, et al., p. 2947). With Argunaut, interaction becomes reductive-abstracted to a meaningless algorithm (at the student level, anyway) in an environment that was stripped of social connection. Additionally, deep alerts beyond a single-shaped contribution (or assertion) quickly became to complex to model, outside of case example matching.
Third, rather a cognitive model that attempts to capture all the "richness of human task performance ... a KC [knowledge component] ...specifies ... a set of hypothesized component processes needed for a set of tasks... facts, skills or principles" (Rosé, et al., p 2949). This is a holistic approach, and one that differs markedly from McClaren's approach, who sought to simply capture stripped down exchanges. Moreover, the KC models become more dynamic across domains because they are splittable, malleable and mergeable (Róse, et al. p 2950). Finally, the Argunaut moderator dashboard indicates a wealth of information about who is contributing and by which 'shape'. But as Róse points out, teachers wanted to know why they are having difficulty and they these struggling students can best be supported. (Róse, et al., p 2950), or to determine where to focus in-class instruction.
Rosé, C. P., McLaughlin, E. A., Liu, R., & Koedinger, K. R. (2019). Explanatory learner models: Why machine learning (alone) is not the answer. British Journal of Educational Technology, 50(6), 2943-2958.
McLaren, B. M., Scheuer, O., & Miksáko, J. (2010). Supporting collaborative learning and e-discussions using artificial intelligence techniques. International Journal of Artificial Intelligence in Education, 20(1), 1–46.
Hello Danica,
ReplyDeleteGreat reflection. I enjoyed reading your reflection. I appreciate how you summarized the potential highlights and potential shortcoming from the McLaren article. You also referred to the takeaway points from the Rosé & McLaughlin article and described well the relevancy to the McLaren’s research approach. I would like to know in more detail your view on how McLaren’s research approach may evolve from Rosé & McLaughlin’s work, and how such changes may impact Argunaut in student-teacher in-class instruction and student performance prediction.