Shallow Adoption
What struck me first is not a finding but a title convention: the word "towards," reappearing paper after paper, a small syntactic flag that a whole field is in motion. I have never tracked a preposition across a conference proceedings before, and now I wonder why. This survey teaches me that a literature review can see a community rather than just summarize its outputs—that repeated linguistic gestures betray a shared epistemic posture. The AIED community, Cairn shows, is not landing; it is leaning. I want to carry that attentiveness into my own reading: what prepositions do we reach for when we don't yet have answers?
The conceptual engine of the piece is a pair of papers placed side by side. One detects affect from keystrokes; the other produces affect in a generative agent. Cairn does not argue about the shift—she lets the juxtaposition do the work, and I feel the ground move under me. This is survey as montage, and it makes me realize how often I try to tell the reader what has changed rather than arranging evidence so the change becomes unavoidable. I will need to read further into how generative models reshape fields that built their identity on detection—but already I recognize the pattern: a discipline's methods become its self-definition, and when the methods become obsolete, the identity crisis is not technical but existential.
The phrase that will stay with me is "displacement of competence from model to surface." A system produces language appropriate to a tutoring context, but does it understand the learner? The success, Cairn writes, is "in the register of performance, not understanding." I have felt this unease—using tools that perform care fluently while I suspect an absence underneath—but I could not name it. Now I have a name. And the risk statement that follows refuses both technophilia and technophobia: the danger is not that we won't adopt generative models but that we will adopt them "too shallowly, treating fluency as understanding and surface personalisation as genuine adaptivity." The risk is that we will be satisfied with the simulation.
What I take from this, as someone still finding my voice, is permission to name tensions I cannot resolve. Cairn's "evidentiary self-correction"—the candid "I cannot yet say"—builds more trust than any pretense of mastery. Her clustered thesis sentences accumulate force through grammatical parallelism. And the metacognitive frame at the end, where she admits to being "struck" by a pattern, positions her not as an expert reporting on a field but as a synthesizer drawing meaning from its structure. I want to write surveys that feel like this: honest about uncertainty, sharp about stakes, and arranged so the reader does not just learn what happened but feels what is shifting.
Comments