Higher Education & ResearchIssue 396 records
Student Success
For 2026-10-16, student-success signals shift from tool access to guided use. UMD told instructors in all 40 courses with its virtual study assistant about trial findings of lower grades and engagement, issued guidance, and its provost said unguided access was not effective. A University of Washington preprint offers a structured alternative: instructor-configured LLM homework graded on the whole conversation, with 134 students reporting confidence gains but no control group. Babson folded AI literacy into its writing centre with consultant training and AI-use intake questions, and City St George's began a 2,000-licence StudyStash Moodle pilot gated by a committee review. Student surveys point the same way: THE reports a Bett survey where most Russell Group students say AI teaching is little or mainly rules, and a 186-student Discover Education survey finds policy and training the top barrier. Operating takeaway (interpretation): pair any AI access with guidance, instructor visibility and decision-grade evaluation. Limits: one institutional study release, two self-reports, one preprint, two survey summaries.
- Evidence records
- 6
- Cross-source patterns
- 3
- Evidence classes
- 2 academic research2 vendor claim2 independent reporting
- Outcomes
- 3 emerging2 cautionary1 mixed
- Source freshness
- 6 new this fortnight
- Research completed
- 2026-10-17
Choose a role to see its takeaway beside every record in the ledger.
Patterns across the evidence
Synthesis · Lighthouse Advisory interpretation
Institutions are moving from open AI access toward guided, structured use
UMD's provost says making its course study assistant available without clear guidance was not effective, and the university has now issued guidance to all 40 courses with access. UW's LLteacher takes the structured route, with instructors configuring the AI's behaviour per assignment and grading the full conversation. Babson builds guidance into its writing centre through consultant training and AI-use intake questions. Interpretation: the operating unit is shifting from licences to guided use, though only UMD's evidence is randomized and it measures the failure of unguided access, not the success of guidance.
Operating questionFor each AI tool students can access, what instructor guidance, structured use mode and usage visibility accompany it?
Supporting evidenceUniversity of Maryland College of Education, Division of Information Technology and Office of the ProvostUniversity of Washington (Furfaro, Lovelace, Zhang, Mosciatti)Babson College Writing Center & AI Literacy Lab
Students report that institutional AI support is mostly rules, not teaching or access
In the Bett survey reported by THE, 77% of Russell Group undergraduates say they are taught little or nothing about AI, and 62% say AI teaching is mainly about assignment rules. In the Discover Education survey, students' top requests were institution-paid access and clearer policy, policy or training was the leading barrier, and professor encouragement trailed peer support. Interpretation: across two different samples, students see a gap between their everyday AI use and institutional instruction. Both are perception surveys with methodological limits.
Operating questionBeyond integrity rules, which programmes teach AI skills, and do students have sanctioned access to practise them?
Supporting evidenceBett UK survey reported by Times Higher Education; Russell Group; Universities UKDiscover Education study led by Deddy Gusman (University of Debrecen; Universitas Pahlawan Tuanku Tambusai), reported by Scienmag
New AI study-tool pilots plan perception-based evaluations, while the one randomized trial found harm
City St George's will judge StudyStash on student and staff feedback about accessibility, efficiency and engagement before a committee decision. UW's LLteacher evidence rests on self-rated confidence and survey agreement without a control group. UMD's randomized trial, the strongest design here, found lower grades and engagement with access. Interpretation: renewal decisions resting on satisfaction data risk missing effects that controlled comparisons reveal.
Operating questionDoes the pilot include a comparison group or baseline on grades and course engagement, not only satisfaction?
Supporting evidenceCity St George's, University of London (LEaD and DiTO)University of Washington (Furfaro, Lovelace, Zhang, Mosciatti)University of Maryland College of Education, Division of Information Technology and Office of the Provost
Evidence ledger
Full record · every source keeps its link and limitations
UMD tells instructors in 40 VSA courses about its negative AI-tutor trial findings and issues guidance; provost says unguided access was not effective
On 16 October UMD described what it is doing after its randomized trial of course-integrated virtual study assistants (VSAs), which involved 30 instructors and about 2,500 undergraduates from 11 of 12 schools in fall 2025. Forty courses currently have VSA access; instructors in all of them were informed of the findings and given best-practice guidance. The release restates that access was associated with lower final grades and substantially lower ELMS-Canvas engagement, with larger estimated effects for first-generation students, that only about 15% of students with access used the tool, and that access may have reduced office-hour attendance and time with instructors and peers. Lead author Jing Liu says students mostly asked for answers rather than using tutoring features. Provost Jennifer King Rice says making tools available without clear guidance was not effective, and proposes giving instructors usage feedback and steering students toward deeper engagement.
Why it matters, evidence and limitations
- Why it matters
- New since the archived RCT coverage: the institution's operational response (guidance to 40 courses, no campuswide rollout) and the provost's explicit diagnosis that unguided access failed. This is a current governance model for any campus enabling course AI tutors.
- Evidence and measured results
- Randomized trial summary (working paper already archived) plus institutional statements. Effects are associations from the trial; office-hour and peer-time effects are described as suggested by the findings. No new data beyond the paper are reported.
- Limitations and uncertainty
- Self-reported institutional interpretation of its own study; most students with access did not use the VSA yet effects were observed, so mechanisms are unclear; single university.
Babson renames its writing centre the Writing Center & AI Literacy Lab, adding AI-use questions to appointment intake and scenario training for consultants
Babson College has renamed its writing centre the Writing Center & AI Literacy Lab. Director Kristi Girdharry expanded its mission so students can discuss how to use AI without surrendering their voice or replacing critical thinking. Faculty consultant Weston Miller describes the stance as neither policing nor cheerleading. Consultants have received training built around AI-use scenarios, and appointment intake now asks whether students want to use AI, whether their professor expects or prohibits it, or whether they prefer not to. Peer consultant Lily Stambouli encourages using AI as a brainstorming partner rather than an author; an MBA student for whom English is not a first language describes using AI to manage long case readings while staying responsible for direction and revision.
Why it matters, evidence and limitations
- Why it matters
- Published 16 October, this is a concrete student-support redesign: AI literacy embedded in an existing tutoring service with intake that captures course-level AI rules, rather than a new chatbot.
- Evidence and measured results
- Interviews with consultants, a faculty consultant and students in a college publication. No usage, satisfaction or learning measures reported.
- Limitations and uncertainty
- Promotional institutional story; no evaluation; single private college.
City St George's pilots StudyStash in Moodle with 2,000 optional licences, anonymised staff analytics and a committee review before June 2027
City St George's, University of London has launched a pilot of StudyStash for 2026-27 in selected cohorts from Optometry, Speech and Language Therapy, and Economics. The tool sits inside Moodle modules and lets students create flashcards, practice tests and podcasts from existing teaching materials. Participation is optional, 2,000 student licences are available, and staff can view anonymised usage analytics to spot engagement patterns. A Speech and Language Therapy year lead hopes it will encourage active learning for first-year students. The pilot runs to June 2027; student and staff feedback will be evaluated for impact on resource accessibility, study efficiency and engagement, and the AI Accelerator and Change Committee will recommend whether to continue. Learning Enhancement and Development manages the pilot and the Digital Innovation and Transformation Office oversees it.
Why it matters, evidence and limitations
- Why it matters
- A second UK university pilot of the same course-material revision tool within days (UEA's Blackboard pilot is already archived) shows vendor spread; City's post adds licence scale, analytics scope and a named decision committee.
- Evidence and measured results
- Institutional announcement with staff quote and vendor tagline. Evaluation planned via feedback; no measured results.
- Limitations and uncertainty
- Feedback-based evaluation cannot isolate learning effects; vendor marketing quoted; outcome claims are aspirations.
THE: Bett survey finds 77% of Russell Group undergraduates taught little or nothing about AI, with most AI teaching focused on assignment rules
Times Higher Education reports a Bett UK survey of more than 1,000 UK undergraduates. More than three-quarters (77%) at Russell Group universities say they are taught little or nothing about AI, against 64% at post-1992 universities, and 62% of Russell Group respondents say that when AI is taught it is mainly about how far it may inform an assignment. Health and medicine students were the most likely to say AI is not covered at all. Only 34% felt their university had been straight about realistic graduate outcomes, and 78% would have been interested in a shorter technology-personalised degree. The article also notes early results from an ongoing 450-student UCLA study, to be presented at a Cambridge summit, suggesting frequent answer-giving AI use raises short-term performance but weakens independent thinking. The Russell Group said its universities are setting clearer assessment rules and adapting curricula.
Why it matters, evidence and limitations
- Why it matters
- Fresh independent evidence that student AI instruction often stops at integrity rules, framing demand for curriculum-embedded AI literacy; transferability to U.S. institutions is limited by UK sector structure.
- Evidence and measured results
- Survey percentages reported by THE; no sample method, weighting or questionnaire shown. Sector bodies' responses quoted. UCLA findings are preliminary and not inspected.
- Limitations and uncertainty
- Survey commissioned by an edtech event organiser; methods not disclosed in the article; self-reported perceptions; UK context.
Survey of 186 students in Indonesia and Hungary finds near-daily GenAI use but top barriers are policy and training, with professor encouragement lagging peers
Scienmag reports an exploratory Discover Education survey of 186 students across Indonesian, English and Hungarian language versions. 78.0% used generative AI several times a week or daily, 96.8% used ChatGPT, and 79.6% said AI had improved their learning, while 51.6% saw it as equally beneficial and risky. The most requested institutional supports were free institution-paid AI access (37.1%) and clearer AI policies (24.2%), and educational policy or training was the most cited barrier (31.2%). Classmate support averaged 3.94 on a five-point scale versus 3.14 for professor encouragement. The authors call this practice-governance decoupling and stress limits: self-selected recruitment, no established measurement invariance, no preregistration or multivariable model, and a sensitivity analysis in which country contrasts shrank.
Why it matters, evidence and limitations
- Why it matters
- International, same-week evidence that student demand centres on sanctioned access and clear guidance; useful context for U.S. literacy and licensing decisions with explicit transferability limits.
- Evidence and measured results
- Descriptive survey statistics with confidence interval for usage; perceptions rather than measured learning; author-stated methodological limits relayed by the article.
- Limitations and uncertainty
- Small non-probability sample; translation without psychometric equivalence; secondary summary; primary paper not inspected in this run.
UW intro statistics course grades whole student-LLM conversations via LLteacher; 134 students report confidence gains, but no control group
A University of Washington team deployed LLteacher, an open-source tool for instructor-configured LLM-guided homework, in a large introductory statistics course serving mostly pre-major undergraduates. Assignments use recall, discovery and critical-thinking modes, and instructors grade the entire student-LLM conversation rather than only the final answer. Among 134 consenting students in an IRB-approved study, all submitted at least two of six assignments and 90% completed all six. Over 80% agreed the assignments improved their comprehension and ability to interpret statistical results, with larger self-rated confidence gains among students who started less confident. Interaction styles varied widely and reported hallucinations were rarely disruptive. The authors state the study had no control or comparison group, relied on voluntary participation in one course and quarter, and cannot establish that AI-guided homework outperforms traditional homework.
Why it matters, evidence and limitations
- Why it matters
- Posted 11 October and not previously captured, it offers a concrete counter-design to UMD's open-access tutor: structured, graded AI conversations. Evidence is preliminary and self-reported.
- Evidence and measured results
- Pre/post self-ratings and end-of-term survey items; interaction logs; descriptive results without comparison group. Authors developed the tool.
- Limitations and uncertainty
- No control group; self-reported confidence; single course; developer-authored preprint; not peer reviewed.
How to read this edition
Source findings, measured results and limitations come from the cited publications. Patterns, operating questions, role takeaways and implementation considerations are Lighthouse Advisory interpretation, stated as questions to validate locally rather than guaranteed outcomes. Vendor and operator claims are labeled as claims. Full research method.
- Academic research
- Research produced through an academic institution or peer-reviewed venue.
- Vendor claim
- A supplier-provided assertion that has not been upgraded to independent evidence.
- Independent reporting
- Independent reporting with attributable sources but without a formal evaluation design.