Implementing Augmented Thinking Protocol in Education

New research (Gerlich 2025) published last week confirms that structured training on prompting & thinking with AI makes it much more likely students will use it as a dialogic partner rather than a cognitive shortcut. Guided support followed these crucial 5 steps: 1. Participants initial unaided reflection (i.e. no AI access) on a preset task. 2. Use of ChatGTP for targeted research to develop context and fact-finding with constrained prompts. 3. Participants revision of original reflections to improve argument construction (without copy-paste access). 4. Critical review of new hybrid outputs using ChatGTP to stress test argumentation. 5. Final revision and reflection with ChatGTP rooted in participants' reasoning and judgement. This process reduced offloading, created higher rubric-rated critical reasoning and higher self-reported reflective engagement. Guided use also narrowed demographic performance gaps and produced what participants described as a “seminar-style challenge”. Another interesting finding . . . Users in other test groups suffered an illusion of non-offloading: they thought they were doing the cognitive work while their behaviours showed otherwise. Interaction design is clearly key. Structured support is a valuable adoption lever that preserves student agency and fosters deeper learning when AI is rolled out at scale. Full study here: https://lnkd.in/dfE7WKsf

For educators seeking practical implementation, not just theory: 1. The 20-Minute Rule Require 20 minutes of unaided work before any AI collaboration Application: Human: Draft initial climate policy arguments AI: Generate counterarguments Toggle: Develop a more nuanced position 2. "Blind Bake" Assessments Students submit three components: Original handwritten draft (scanned) AI-enhanced version "Toggle Map" documenting their decision-making process Key Benefit: Makes cognitive offloading visible and intentional 3. Skill-Specific Toggle Zones Designate certain skills as AI-free territories: History: Primary source analysis Science: Hypothesis formulation Math: Initial problem structure and approach 4. Feedback Roulette Peer reviewers identify: Which sections appear AI-assisted Their reasoning behind these assessments Builds meta-awareness for both creator and evaluator 5. Cognitive Time-Stamping Leverage document version history to: Compare thinking before and after AI assistance Identify when AI bypassed valuable struggle Evaluate process quality, not just final output Free Resource: I've created a Toggle Method Lesson Planner template – comment "TOGGLE TOOL" for access. "AI shouldn't make thinking easier – it should make thinking deeper." #EdChat #AIStrategy #CriticalThinking #ToggleTeaching Pragmatic AI Solutions Alfonso Mendoza Jr., M.Ed. Amanda Bickerstaff Vriti Saraf Pat Yongpradit France Q. Hoang Mike Kentz

Important new evidence on ChatGPT in education: Wang & Fan's (2025) meta-analysis of 51 studies shows we're at an inflection point. The technology demonstrably improves learning outcomes, but success depends entirely on implementation. The research reveals optimal conditions: sustained use (4-8 weeks), problem-based contexts, and structured support for critical thinking development. Effect sizes tell the story; large gains for learning performance (g=0.867), moderate for critical thinking (g=0.457). Quick fixes don't work. Thoughtful integration does. Particularly compelling: ChatGPT excels in skills development courses and STEM subjects when used as an intelligent tutor over time. The key? Providing scaffolds like Bloom's taxonomy for higher-order thinking tasks. As educators, we have emerging empirical guidance for AI adoption. Not whether to use these tools, but how to use them effectively - maintaining rigor while enhancing accessibility and engagement. The future of education isn't human or AI. It's human with AI, thoughtfully applied.

The biggest threat AI poses to student learning is not cheating. 🚨 It is the illusion of competence. When an AI tool instantly generates a highly articulate answer, students often mistake the machine's fluency for their own mastery. Bypassing the productive struggle required for true comprehension leads to metacognitive laziness, weak retention, and an inability to transfer knowledge to new situations. We cannot simply ban these tools. We must fundamentally redesign the workflow. To combat this, I utilize a framework called The Cognitive Sandwich. It structures learning into three distinct phases to ensure AI supports human thinking rather than replacing it. First, students must attempt the work independently to engage in the necessary productive struggle. Next, AI is introduced strictly as a Socratic coach to challenge reasoning and provide hints rather than direct answers. Finally, the student must synthesize the feedback and produce the final outcome entirely in their own words. Alongside this framework, we also have to implement strict safeguards for foundational skills and require unaided checks to verify true understanding. If a student cannot perform the task without an AI assistant, they do not truly know the material yet. Guiding leadership teams to build and implement instructional strategies exactly like this is what I focus on when partnering with educational institutions. Designing comprehensive AI literacy training that protects real learning while supporting neurodiverse students and meeting federal compliance takes deliberate, strategic planning. Technology should elevate our classrooms, but we absolutely must protect the human learning process. How is your campus balancing AI exploration with foundational skill building? Let us talk about it in the comments. 👇 #AILiteracy #EducationLeadership #GoogleEdu #FutureOfLearning #InstructionalDesign #EdTech #TeachingWithAI #SpecialEducation #UniversalDesignForLearning

Whenever I post a concern about AI in education, someone pops up to say it has accelerated their learning. I get it. I have too. But the research on cognitive offloading keeps mounting, and I worry what that means for attention, memory, and genuine understanding, especially for younger learners. Of course, it mostly comes down to how we use AI. Tools set defaults. Defaults become habits. Habits shape minds. Which is why I was heartened by this new study. In a cross-country experiment with about 150 participants, unguided access to ChatGPT gave only a small bump over human-only work and often looked a lot like AI-only output. Add a simple scaffold and the curve bends. Reflect first. Use AI narrowly to gather evidence. Draft in your own words. Ask the model to attack your draft. Then revise. Under that guided workflow, critical-thinking scores jumped by roughly forty percent and people reported feeling more mentally engaged, even though the task felt harder. That harder-but-better point matters. The risk is not AI in education. The risk is the default, unstructured way many people use it. Unguided, the tool invites passivity and machine-shaped prose. Guided, it behaves like a sparring partner. The mechanism is reflective engagement. Slow down to take a stance. Use the model to surface evidence and adversarial feedback. Iterate. That desirable difficulty is where learning lives. There is also an equity signal. Younger or less experienced participants started lower, but the structured workflow helped narrow, though not eliminate, the gap. That is exactly what you want in schools, where anything-goes AI use risks widening disparities. The right defaults do not just lift averages. They compress variance. So what should classrooms do with this? - Teach AI as critic and evidence finder, not ghostwriter. - Make process visible and assessable. - Require a short pre-write. - Allow targeted AI look-ups. - Insist on drafting in the student’s own words. - Then require an AI red-team of the draft before revision. Grade the product and the receipts bundle: pre-write, sources gathered with AI, the critique transcript, and a brief reflection on what changed. In edtech and LMS design, tilt the experience toward question, critique, evidence by default and delay full-text generation until a claim is on the table. Set rails that make the reflective path the easy path. AI can speed learning. Without structure, it speeds forgetting. Research Caveats: one topic domain; short-run effects; a convenience sample around universities and workshops; some measures based on self-report. The comparative signal is strong, but we should want replication across subjects, age bands, and longer retention windows. Use it as a guide, not gospel.

A student recently asked me “Is AI making it harder for me to think critically, even while it’s helping me write better?” That question gets at what many of us are seeing. The outputs look smoother, but it is easy for students to slide into “accept and submit” mode. Introducing the Push-Back Protocol, a simple five-round framework that turns AI from a shortcut into a thinking partner: + Demand evidence + Surface assumptions (and bias) + Ask for alternative perspectives, including non-U.S. contexts + Stress test the argument + Synthesize and reflect The key idea: the first AI response is not the answer. It’s the starting point. If you’re teaching, designing assignments, or coaching students, I hope this gives you something practical to try. https://lnkd.in/gicsDwnp

Most AI conversations in education are stuck in one lane: policing. But the real question I wrote this book for is: What does learning look like when AI is present—everywhere—and we still want students to think for themselves? My book, Students First, AI Second: Practical Thinking Tasks for Every Classroom, is built around a simple principle: AI should never replace student thinking. It should reveal it, stretch it, and make it visible. Here are 3 practical shifts from the book that change the way AI fits into a lesson: 1) From “answers” → to “thinking traces.” Don’t grade the final product only. Grade the path: decisions, revisions, reasoning, evidence, reflection. 2) From “tool use” → to “task design.” AI doesn’t improve learning by existing in the room. Learning improves when teachers design tasks that require judgment, tradeoffs, and explanation. 3) From “Got it / didn’t get it” → to “show me your process.” Even a weak answer becomes valuable when students can explain why they chose it, what they ruled out, and what they’d improve. Here’s a ready-to-use classroom task (works with or without AI): ✅ The “Better Question” Task (10 minutes) • Give students a topic (or a reading / problem / image). • Ask them to write one “bad” question (too broad, yes/no, obvious). • Then rewrite it into three better questions: 1. a clarifying question 2. a challenging question 3. an application question • Students choose the best one and justify why. If AI is allowed, students can consult it—but they must also write: “What did the AI miss or oversimplify?” and “What question is still unanswered?” That’s the whole point of Students First, AI Second: use AI to increase student thinking—not reduce it. If you’re a teacher, trainer, or school leader and you want practical, repeatable tasks that protect student voice and modernize learning, this book is for you. #AIinEducation #Teaching #CriticalThinking #Assessment #TeacherDevelopment #HigherEd #ESL #ClassroomPractice

I'm very excited to share the results of an #AI #teaching experiment that Aimée Skidmore and I ran in her Grade 12 classroom at the Collège du Léman - International School in Geneva, Switzerland. We aimed to leverage the principles of comparative textual analysis from writing and literature classrooms to teach students how to use #ChatGPT in ways that support and reveal thinking, rather than diminish it. Over four weeks, we layered in a sequential, scaffolded approach to analyzing interactions with AI tools to create a benchmark for its use in the context of an existing process. Then, we asked students to perform a specific set of skills while using AI. Aimee graded and gave feedback to each student. She also asked them to annotate their chats to better understand their approach. We surveyed students before and after the experiment, and their sentiments were eye-opening: 1. 85.7% of students reported changing their approach to AI use 2. 47.6% became "significantly more strategic" in their AI interactions 3. 81.0% endorsed the continued use of this approach in classrooms The full analysis of our experience -- including reflections around mistakes that we made and recommendations for future research in this area -- will be published in an Elsevier book titled "GenAI and Higher Ed" later this summer. But we wanted to provide a preview with the story and our high-level takeaways beforehand. You can find Aimee's write-up below on my blog - "How We Frame Machines." We hope this endeavor leads to further and deeper research in the field of AI literacy and meaningful AI use in educational settings. Nick Potkalitsky, PhD Pat Yongpradit Amanda Bickerstaff Tiffany Hsieh Mamie Rheingold Margaret Vo Steven Butschi Isabelle Hau Eric Tucker Rick Dakan Kevin Yee https://lnkd.in/evnJPn5M

Lost amidst the #AI tsunami in education is metacognition. We are racing to teach students which tools to use. But we are quietly un-teaching the skill that makes any tool useful. How to notice what you know, what you do not know, and what to do next. If students cannot think about their thinking, AI turns into: → Fast answers with shallow understanding → Confident nonsense with no internal error-checking → Outsourced judgment when judgment matters most In entrepreneurship, this gets exposed instantly. → Weak evidence. → No learning loop. → Bad assumptions. → Expensive mistakes. So here’s a practical fix you can run tomorrow. No new software required. The 5-minute Metacognition Loop (before, during, after any task) Predict “What will be hard about this? What do I already know?” Plan “What is my approach? What is step one?" Monitor “What is confusing me right now? What is my next question?” Validate “What would convince me I am wrong? What is my evidence?” Debrief “What did I learn about the topic, and about how I learn?” Want to use AI too? Great (I think you should). Make AI the mirror, not the brain: Ask it to challenge assumptions, generate counterexamples, and test reasoning. If we teach in a world of AI but we do not teach metacognition we are training prompt operators, not thinkers. Patrick Dempsey Lily Abadal, Ph.D. Jason Gulya Sam Illingworth Tawnya Means Emily Pitts Donahoe Marc Watkins

Today, I would like to share a recent article on integrating AI into education entitled "Integrating AI-generated content tools (AIGC) in higher ed: A comparative analysis of interdisciplinary learning outcomes" by Zhang and Tang (2025) (https://lnkd.in/e4mNchms ). Although AIGC tools are now widely adopted in higher ed, few studies systematically compare their impact across STEM, humanities, social sciences, business, and health fields. Zhang and Tang address this gap through a dataset that includes 1,099 students, 252 faculty members, 86 classroom observations, and both pre/post assessments and interviews across 15 institutions. Findings 1. Meaningful Gains in Interdisciplinary Learning Outcomes. When AIGC tools were strategically integrated interdisciplinary project outcomes increased 37%, measured through collaborative problem-solving, cross-domain knowledge synthesis, and peer communication. Improvements were strongest in: - Interdisciplinary communication (+23.6%) - Creativity (+17.4%) - Knowledge acquisition (+17.2%) - Skill development (+16.0%) These gains substantially exceed those typically associated with traditional EdTech tools, such as LMS. 2. Discipline-Specific Patterns Matter. The authors found that AIGC adoption varies markedly by disciplinary epistemology and instructional culture: - STEM fields show the highest usage (87% weekly), emphasizing code generation, simulation modeling, and structured prompting. - Humanities/social sciences adopt more slowly but display deeper pedagogical integration often using AIGC as a critical object of analysis. - Business and economics benefit most from AI-generated scenarios. - Medical/health sciences used for diagnostic simulations or case variation. 3. Pedagogical Design Determines Learning Quality. The study introduces a Quality of Integration Index (QII), showing that high gains correlate with: - Pedagogical coherence - Explicit alignment between AIGC use and learning outcomes - Depth of curricular integration 4. Students Treat AIGC as an Intellectual Partner. Students learn best when AIGC tools are framed not as answer generators but as collaborative partners. This aligns with emerging research on “AI-assisted sense-making,” where students refine, critique, and extend AI-generated output. Across all disciplines, the study identifies five success principles: - Faculty co-design rather than top-down tool implementation - Explicit alignment between AI capabilities and outcomes - Staged implementation with iterative refinement - Dual-track assessment (AI-assisted vs. independent work) - Transparency about AI limitations for students Institutions that followed at least four of these achieved 54% higher learning gains and 68% higher faculty satisfaction. Reference Zhang, Y., & Tang, Q. (2025). Integrating AI-generated content tools in higher ed: A comparative analysis of interdisciplinary learning outcomes. Scientific Reports, 15(25802), 1–14.