SIDLIT 2026

Higher education's dominant response to generative AI has been swift and understandable, but wrong. Institutions deployed AI detection tools as the first line of defense against academic dishonesty with generative AI, and many built punitive consequences into courses and policies. The research on those tools is now substantial, peer-reviewed, and damning. Detection tools produce false positives at rates that should disqualify them from high-stakes use. They disproportionately flag non-native English speakers, neurodivergent writers, and students from educational backgrounds that emphasize clarity over elaboration, making this an equity problem, as well as an integrity issue:
- Weber-Wulff et al. (2023) found available tools neither accurate nor reliable.
- Liang et al. (2023) documented that detectors misclassified over 61% of essays by non-native English speakers as AI-generated while achieving near-perfect accuracy on native speaker writing.
- The MLA-CCCC Joint Task Force explicitly cautions against their use.


This session argues that the problem is a misdiagnosis.  Institutions have identified the symptom while misidentifying the cause.  The actual cause is assessment design. When a generative AI can complete an assignment without genuine engagement with the learning it was designed to measure, the assignment is measuring the wrong thing.  AI did not create that problem; it exposed it.

The fix is not better detection; it is better assessment design. Well-designed assessments that require transfer learning (application of knowledge to novel, specific, contextually embedded situations) are not nearly as vulnerable to AI completion as retention-focused assessments are. This session presents the evidence, makes the reframe, and demonstrates a live before-and-after assessment redesign so participants leave with concrete strategies they can apply immediately.  Responsible use of AI in education begins with this honest reckoning.