Task-contingent risk configurations of Generative Artificial Intelligence in education: a URL-clustered analysis of a risk-enriched multi-platform corpus
Primary research
#651
- Topic
- unassigned (set during synthesis)
- First seen
- 2026-07-23 07:15:51
- Last seen
- 2026-07-23 07:15:51
Source raw items (1)
- Semantic Scholar2026-07-23 07:15:18Task-contingent risk configurations of Generative Artificial Intelligence in education: a URL-clustered analysis of a risk-enriched multi-platform corpus
Generative Artificial Intelligence (GenAI) has raised concerns about academic integrity, cognitive dependence, professional roles, educational quality, and equitable participation. Research tends to report these concerns as a unified list of risks, even though their educational significance varies with the task in which AI is envisaged or applied. This study analysed a separately curated, risk-inflated corpus of 3,000 public-text records from six Chinese social media platforms and and 300 source Uniform Resource Locators (URLs). Five non-exclusive risk constellations and five educational task contexts were reconstructed from indexed keyword fields. Binary logistic generalised estimating equations clustered by source URL modelled task associations while adjusting for platform and text length. Cognitive-reliance language formed 35.5% of the corpus, academic-integrity language 22.7%, and teacher-role disruption 18.4%. Assignment/writing and assessment/feedback contexts were strongly associated with academic-integrity language (OR = 3.59, 95% CI [2.86, 4.51]; OR = 3.56 [2.90, 4.37]). Teacher-work contexts were associated with teacher-role disruption (OR = 1.39 [1.10, 1.74]). Variation between platforms was slight for four configurations; teaching-quality concern exhibited modest distinction. The results advocate task-based governance that combines process evidence with assessment, verification with teacher judgment, and access support with institutional policy. Since the corpus was intentionally risk-inflated, its percentages characterise semantic composition rather than platform-user prevalence.