Interactive toolToolsEthicsEnvironment
Interactive visual resource2026-06
Andy Masley
AI Literacy & Faculty Practice
An interactive calculator that estimates the carbon and water footprint of everyday AI use across models, output lengths, and usage patterns.
Why it matters: Useful for putting environmental-impact claims into proportion without turning them into a vague reason for either blanket adoption or blanket avoidance.
Evidence note: Use as an awareness and estimation tool, not a precise accounting system. Estimates depend on model, output length, regional grid assumptions, and exclusions such as training, image/video generation, and retries.
Open resource
PreprintEquityPolicyOpinion
SocArXiv (OSF Preprints)2026-05-23
Mendoza
Equity & Student Differences
Drawing on distributed cognition, writing studies, and disability studies, this conceptual paper argues scholarship has always relied on unevenly distributed supports - and distinguishes accountable scaffolding from substitution.
Why it matters: A simple AI prohibition may cost most for students with the least access to human, linguistic, or institutional support - a more equitable basis for assignment-level decisions.
Evidence note: Conceptual preprint rather than an empirical study.
Read the paper
PreprintPolicyResearchTools
arXiv2026-03-04
de Silva, Song, Yang & Humayoun
AI Literacy & Faculty Practice
Focus groups with 29 STEM faculty at a large U.S. public university show GenAI adoption is often student-driven; faculty describe pedagogical uses alongside concerns about learning, assessment validity, and integrity, and name the institutional support they need.
Why it matters: Faculty development cannot stop at tool demos - discipline-specific spaces to redesign assessment and compare practices matter more.
Evidence note: Qualitative preprint from one institution and one broad disciplinary cluster.
Read the paper
Peer-reviewedAssessmentResearchEquity
International Journal for Educational Integrity2026-02-02
Hadra, Cambridge & Mesbah
Assessment & Learning Assurance
Testing Turnitin and Originality on 192 texts - student EFL writing, professional writing, AI output, and hybrids - found modest accuracy (0.61 and 0.69), poor performance on hybrid texts, and signs of unfairness toward non-native English writing.
Why it matters: Detector output cannot carry a misconduct decision on its own. The findings support detection, if used at all, only as a prompt for human inquiry within clear policy.
Evidence note: Two commercial detectors at one point in time; performance shifts as tools and models update.
Read the paper
Peer-reviewedAssessmentResearchTools
Computers in Human Behavior Reports2026
Lau, Low, Tay, Guevarra, Gasevic & Hartanto
AI Literacy & Faculty Practice
Across six studies (N = 1,365), the authors validate a 13-item scale measuring verification, motivation, and reflection in GenAI use. Higher scores predicted more diverse verification strategies and better accuracy in a ChatGPT-powered fact-checking task.
Why it matters: Provides measurable outcomes for activities that ask students to check, challenge, and reflect on AI outputs - beyond confidence or completion.
Evidence note: First posted as an arXiv preprint (Dec 2025), now published; validate locally before high-stakes use.
Read the paper
Peer-reviewedAssessmentPolicyTools
Journal of University Teaching and Learning Practice2025-12-03
Perkins, Roe & Furze
Assessment & Learning Assurance
The authors of the AI Assessment Scale revise it after two years of worldwide adoption: clarified theoretical grounding, a new 'AI exploration' level, and a reframing of the scale as an assessment redesign framework that strengthens validity rather than a simple permission label.
Why it matters: This is the framework the course on this site uses to communicate assignment-level AI expectations - the revision shows how to use it to redesign tasks, not just label them.
Evidence note: A conceptual framework paper informed by implementation experience, not a controlled outcome evaluation.
Read the paper
Peer-reviewedAssessmentResearchTools
International Journal of Digital Literacy and Digital Competence2025-09-11
Ranieri, Biagini & Cuomo
AI Literacy & Faculty Practice
Develops and validates a 24-item Critical Artificial Intelligence Literacy Scale covering knowledge, operational ability, critical understanding, and ethics, tested with 314 first-year student teachers with a reliable four-factor structure.
Why it matters: A concrete way to assess AI literacy as more than tool fluency - usable for learning outcomes, reflection prompts, or program evaluation.
Evidence note: Validated with student teachers; confirm fit before using with other populations.
Read the paper
Peer-reviewedPolicyResearch
AI in Education2025-09-02
Lund, Mannuru, Teel, Lee, Ortega, Simmons & Ward
Policy, Integrity & Governance
A survey of 401 U.S. university students found that students' ethical beliefs predicted perceived misconduct and actual AI-assisted writing use better than awareness of institutional policies. Many students see AI use as distinct from plagiarism.
Why it matters: Rules alone may not change behavior - explain the learning rationale behind boundaries and give concrete examples of acceptable and unacceptable use.
Evidence note: Relies on self-reported perceptions and behavior.
Read the paper
Peer-reviewedResearch
Computers and Education: Artificial Intelligence2025-08-06
Xia, Li, Yang, Weng & Chiu
Teaching & Student Learning
This meta-analysis finds positive effects of GenAI on university student motivation and engagement, varying by subject, learning strategy, group size, and mode of use - with individual and small-group use showing notable cognitive and emotional benefits.
Why it matters: Supports a conditional claim: GenAI improves engagement when the learning strategy is well designed - a reason to specify how, when, and with whom students use AI.
Evidence note: Engagement is not the same as durable learning, and effects vary across contexts.
Read the paper
PreprintEquityPolicyResearch
arXiv2025-08-01
Contractor & Reyes
Equity & Student Differences
Survey evidence from a selective U.S. college shows GenAI adoption above 80% within two years of ChatGPT's release, varying across disciplines, demographics, and achievement levels - used both to augment learning and to automate coursework.
Why it matters: Students are not a uniform user group: collect local evidence, separate learning-enhancing use from substitution, and check whether policies burden groups differently.
Evidence note: Preprint based on one selective college, which limits generalizability.
Read the paper
Peer-reviewedResearch
Journal of Computer Assisted Learning2025-07-15
Liu, Zuo & Lu
Teaching & Student Learning
A meta-analysis of 37 experimental studies (2022-2025) finds a moderately positive overall effect of ChatGPT use on academic achievement (g = 0.577), larger for social sciences, 5-10 week interventions, declarative knowledge, and traditional-classroom integration.
Why it matters: A calibrated, peer-reviewed answer to 'does ChatGPT help learning?': a moderate average benefit that depends on duration, discipline, and instructional design.
Evidence note: Depends on the quality and heterogeneity of included studies, many short with modest samples.
Read the paper
Peer-reviewedAssessmentResearchTools
Proceedings of the National Academy of Sciences2025-06-25
Bastani, Bastani, Sungu, Ge, Kabakci & Mariman
Assessment & Learning Assurance
A field experiment with nearly 1,000 math students: a standard ChatGPT-style tutor boosted performance while available, but students scored 17% worse than the no-AI group once access was removed. A tutor designed with learning guardrails largely prevented that harm.
Why it matters: Access to AI is not the key design decision - the interaction rules are. Tutoring designs that delay answers and prompt reasoning protect independent skill.
Evidence note: Conducted in high school mathematics; transfer to higher education should be tested, not assumed.
Read the paper
Peer-reviewedResearchTools
Scientific Reports2025-06-03
Kestin, Miller, Klales, Milbourne & Ponti
Teaching & Student Learning
In a randomized controlled trial with college students, a custom AI tutor built around research-based pedagogy produced more learning in less time than an in-class active-learning lesson, with higher reported engagement and motivation.
Why it matters: Strong evidence that a carefully designed AI tutor can support learning - but the tutor was purpose-built. Embed pedagogy, boundaries, and course context rather than deploying a generic chatbot.
Evidence note: One course and one tutor design; adoption should preserve the study's pedagogical conditions.
Read the paper
Peer-reviewedPolicyResearch
IFLA Journal2025-05-08
Wilson
Policy, Integrity & Governance
An examination of how UK universities developed GenAI policies covering academic integrity, assessment, student use, and staff support - finding uneven policy development and resource constraints outside highly resourced institutions.
Why it matters: AI policy is a maintenance practice, not a one-time document: collaborative development, shared resources, student input, and recurring review.
Evidence note: UK institutional setting, though the governance issues transfer readily.
Read the paper
Peer-reviewedResearchTools
CHI 2025 (ACM Conference on Human Factors in Computing Systems)2025-04-25
Lee, Sarkar, Tankelevitch, Drosos, Rintel, Banks & Wilson
Assessment & Learning Assurance
A survey of 319 knowledge workers (936 real AI-assisted tasks): higher confidence in GenAI predicted less reported critical thinking, while higher self-confidence predicted more. AI shifted critical effort toward verification, integration, and stewardship.
Why it matters: It defines the critical-thinking practices students will need in AI-mediated work - assignments can assess verification, integration, and judgment explicitly.
Evidence note: Knowledge workers rather than students, and self-reported measures.
Read the paper
PreprintPolicyResearch
arXiv2025-01-15
Simkute, Kewenig, Sellen, Rintel & Tankelevitch
Policy, Integrity & Governance
Interviews with 26 students and 11 educators across two universities reveal AI use shaped by unclear guidance, inconsistent communication, plagiarism-focused discourse, and unspoken rules - with concerns about confidence, skill development, and agency.
Why it matters: Explains why broad institutional statements fail at the assignment level: students need consistent, contextual examples of appropriate use.
Evidence note: Qualitative preprint based on two universities and a modest interview sample.
Read the paper
