Education and the Wild West of AI:
What the Research Actually Shows
Lisa Nieuwenhuizen, Ph.D.
William Woods University
Two things are true at once, genuine learning gains and measurable cognitive costs.
This session is about telling them apart, using research from 2024–2026.
Education and the Wild West of AI
FRAMING
Reality Check: What’s working, what should worry us, and
what all this means for teacher preparation
AI is producing real, causally-measured learning gains and real, measurable cognitive costs. Our job as teacher educators isn't to pick a side. It's to teach our candidates, and ourselves, to read the conditions that separate one outcome from the other.
Not a technology question
“Is AI good or bad for learning” is the wrong frame. The evidence says: it depends on task design, verification habits, and who has access.
A design question
Every study in this session turns on how the tool was used, not whether it was used. That's the lever we actually control.
Education and the Wild West of AI
CURRENT REALITY: ADOPTION VS. POLICY
Adoption is running ahead of governance
57%
of institutions now call AI a strategic priority, up from 49% a year earlier
22%
have an institution-wide AI strategy; the rest is departmental or ad hoc
55% describe their AI rollout as happening informally, department by department. Students are already using these tools; our institutional policies are still catching up.
What is the reality in your institution?
Education and the Wild West of AI
EQUITY CHECK
The access divide is inside our own walls
50%
of institutions grant students no institutional access to generative AI tools at all
83%
of EDUCAUSE respondents worry AI adoption will widen the digital divide
Among institutions with no student access, cost, not academic integrity, was the top-cited barrier. Larger, better-resourced institutions are also the ones most likely to plan for more access.
Does your institution grant students access to generative AI tools? Google Education Accelerator?
Education and the Wild West of AI
WHAT’S WORKING? STRONGEST OPPORTUNITY EVIDENCE
AI tutoring: some of the cleanest causal evidence in ed-tech in years
A 2025 Harvard randomized controlled trial found AI tutoring outperformed in-class active learning in an authentic course setting, not a lab study, a real class.
A Brookings synthesis of 2024–25 RCTs: Harvard, the World Bank, Stanford, UK researchers, found a consistent pattern: gains in learning outcomes, knowledge transfer, motivation, engagement, and efficiency across independent studies and contexts.
Education and the Wild West of AI
CHANGING THE DESIGN
The gains are conditional; the design is the pedagogy
“AI tutoring” is not one intervention. Across the studies, the gains held only when the system was built with:
Socratic questioning, not direct answers
Scaffolded, sequenced content — not open-ended chat
Explicit safeguards against hallucination, like step-by-step verification
Feedback at both the answer level and the reasoning-step level
A teacher-guided hybrid model, not stand-alone chatbot use
Education and the Wild West of AI
How are you using AI in your pedagogy?
BEYOND TUTORING
Three more places the evidence is promising
Feedback at scale
Formative feedback on drafts, in-process, before a grade is at stake — where students report it's most useful.
Access supports
Reading-level adjustment, translation, and first-draft scaffolding for multilingual and disabled learners.
Instructor time
Drafting, rubric-building, and administrative relief — freeing time toward higher-value teaching moves.
These are lower-evidence than the tutoring RCTs, mostly early and mixed-methods work, but converge with the same conditional pattern: value shows up when a person stays in the loop.
How are we training future teachers to use these tools?
Education and the Wild West of AI
WHAT SHOULD WORRY US? THE HEADLINE STUDY
MIT Media Lab: “Your Brain on ChatGPT”
54 participants wrote essays across three conditions: LLM-assisted, search-engine-assisted, brain-only, while wearing EEG, across multiple sessions over four months.
Neural engagement
Brain-only writers showed the strongest, most distributed neural connectivity. LLM users showed the weakest.
Ownership & memory
LLM users reported the lowest sense of ownership over their own essays and struggled to accurately quote their own work back.
Effects persisted: participants who wrote with an LLM first, then switched to brain-only writing, still showed reduced connectivity months later — researchers call this “cognitive debt.”
Education and the Wild West of AI
THE BIGGER PICTURE
It's not uniform: a 123-study review finds two patterns, not one
A 2025–26 scoping review of 123 peer-reviewed studies found generative AI can support or erode learner agency and the difference is systematic, not random.
Agency-supporting
Self-regulated learning, reflective evaluation of AI output, strategic revision, scaffolded problem-solving.
Agency-eroding
Offloading without verification, uncritical acceptance of output, displaced responsibility for the thinking.
Education and the Wild West of AI
What do you allow your students to do with AI?
A WORD OF CAUTION: ACADEMIC INTEGRITY
Our assessments were built for a world that's over
40→70%
AI-detector accuracy, 2020 to 2024 still unreliable against paraphrased or “humanized” text
Uneven false positives
Detectors flag non-native English writers at disproportionately higher rates than native speakers.
Education and the Wild West of AI
Responses to address the problem
WHAT THIS MEANS FOR TEACHER PREPARATION: OUR CANDIDATES
Preparation is uneven and mostly bolted on
A 2025 study gave 60 pre-service teachers an 8-hour foundational AI training. Self-assessed AI knowledge and general attitude toward AI both improved. Trust in AI, showed no statistically significant change.
Across the teacher-education literature, AI literacy is inconsistently offered, often elective, and rarely embedded across methods courses rather than isolated in a single ed-tech unit.
That's a design opportunity specific to this room: we are not just users of this technology. We are training the people who will make AI-use decisions in front of K-12 students for the next thirty years.
How do we prepare tomorrow’s teachers for AI?
Education and the Wild West of AI
SYNTHESIS
One framework, five moves
1
Treat AI as augmentation embedded in tasks that require justification, comparison, and explanation, not a shortcut around them.
2
Redesign assessments before you police them.
3
Teach verification and trust-calibration explicitly, not just tool operation.
4
Ask who has access before you assume adoption; build equity checks into syllabi and program design.
5
Model your own use transparently with candidates. Show your reasoning about when you do and don't reach for it.
Education and the Wild West of AI
Discussion
Where in your own courses would “cognitive debt” show up first? Would you notice it?
Which of your current assessments are most exposed right now?
What would sustained AI literacy look like across our whole teacher-prep sequence, not just one course?
Education and the Wild West of AI
References
This presentation was created with the assistance of Claude.AI
Education and the Wild West of AI
Bilbao-Eraña, A., & Arroyo-Sagasta, A. (2025). Fostering AI literacy in pre-service teachers: Impact of a training intervention on awareness, attitude and trust in AI. Frontiers in Education, 10, Article 1668078. https://doi.org/10.3389/feduc.2025.1668078
Burns, M. (2026, January 27). What the research shows about generative AI in tutoring. Brookings. https://www.brookings.edu/articles/what- the-research-shows-about-generative-ai-in-tutoring/
Evangelista, E. D. L. (2025). Ensuring academic integrity in the age of ChatGPT: Rethinking exam design, assessment strategies, and ethical AI policies in higher education. Contemporary Educational Technology, 17(1), Article ep559. https://doi.org/10.30935/cedtech/15775
Flaherty, C. (2025, April 21). Half of colleges don't grant students access to gen AI tools. Inside Higher Ed. https://www.insidehighered.com/news/tech-innovation/artificial-intelligence/2025/04/21/half-colleges-dont-grant-students-access
Kestin, G., Miller, K., Klales, A., Milbourne, T., & Ponti, G. (2025). AI tutoring outperforms in-class active learning: An RCT introducing a novel research-based design in an authentic educational setting. Scientific Reports, 15, Article 17458. https://doi.org/10.1038/s41598-025-97652-6
Kosmyna, N., Hauptmann, E., Yuan, Y. T., Situ, J., Liao, X.-H., Beresnitzky, A. V., Braunstein, I., & Maes, P. (2025, June 10). Your brain on ChatGPT: Accumulation of cognitive debt when using an AI assistant for essay writing task. arXiv. https://doi.org/10.48550/arXiv.2506.08872
Robert, J., & McCormack, M. (2025, February). 2025 EDUCAUSE AI landscape study: Into the digital AI divide. EDUCAUSE. https://library.educause.edu/resources/2025/2/2025-educause-ai-landscape-study
Wang, G., Wang, W., Yang, D., & Ren, J. (2026). Generative AI, cognitive offloading, and learner agency in higher education: A scoping review. Behavioral Sciences, 16(7), Article 1150. https://doi.org/10.3390/bs16071150