The Next Layer of AI-Resistant Learning: What the Research Says (and What It Looks Like in Practice)
In an article I wrote a few years ago I shared 10 AI-Resistant Practices for the Classroom. These were simple ways to design learning experiences that prioritize human thinking, discussion, and creation.
But AI-resistant learning isn’t just a reaction to ChatGPT or generative AI.
It’s actually rooted in decades of research about how people learn best.
When you look closely at the science of learning (constructivism, cognitive psychology, and social learning theory) you realize something that most effective learning experiences were already AI-resistant.
They emphasize critical thinking, discussions, experimentation, reflection, and human interaction.
So let’s take the next step and build on theAI-resistant practices along with some research that supports them and examples of what they look like in real classrooms.
1. Retrieval and “Thinking Without Tools”
The Research
Cognitive science consistently shows that retrieval practice, which is pulling knowledge from memory, improves long-term learning more than rereading or passive review.
Research by cognitive psychologists such as Roediger and Karpicke shows that testing improves retention far more than additional study.
If students rely on AI to generate answers, they skip the mental work required for retrieval. This is why we have to design AI-Resistant practices, that align with our purposes in the classroom.
Classroom Example
The 3-Minute Brain Dump
At the start of class, students answer a prompt without devices like…
“Write everything you remember about photosynthesis.”
“Explain the causes of the American Revolution.”
“Sketch how Boyle’s Law works.”
Then students compare answers in groups. AI can't do the retrieval for them so students must rely on their own memory and reasoning.
Easy and effective.
2. “Defend Your Thinking”
The Research
Students explaining their reasoning is one of the most powerful learning strategies available.
Research on metacognition shows that students learn more when they explain their thinking and reflect on their reasoning.
John Hattie’s synthesis of research identifies self-reported grades and metacognition among the highest-impact strategies for learning.
Classroom Example
The Thinking Defense
Instead of just turning in work, students answer questions like these…
Why did you choose this strategy? What was the hardest part? What mistake did you fix?
In math have students record a 60-second video explaining how they solved a problem.
In English class have students explain why they chose a particular piece of evidence.
We need to DEFEND learning, and not focus all of our attention on assessing final products (which can be easily created with AI), but instead on assessing the learning and thinking.
3. Real-World Messy Problems
The Research
AI is excellent at solving structured academic tasks. It struggles a bit with ambiguous real-world problems. These are the types of problems that students who have a base of academic knowledge can begin solving to transfer that knowledge and skill to new and unique situations.
A big piece of the research shows that authentic problems increase motivation and deeper learning because students must apply knowledge in meaningful contexts.
Learner-centered learning research also emphasizes authentic tasks as key to engagement and transfer.
Classroom Example
Instead of “Write an essay about water pollution.”
Maybe your students design a local solution to the issue, after understanding the basics.
Students could Interview community members, research local data, come up with unique solutions, and present their recommendations to a local board or governing group.
AI can help with the research, but students must interpret, decide, and design.
4. Collaborative Sensemaking
The Research
AI can generate information. But learning often happens through social interaction.
Social constructivist theory (Vygotsky) shows that students build understanding through dialogue and shared problem-solving.
Research consistently shows collaborative learning increases deeper understanding and retention. Retention being a key word here!
Classroom Example
The Four Corners Argument
Students move to different corners of the room based on their opinion.
Example prompt: “Should social media companies regulate misinformation?”
Students then defend their position, respond to counterarguments, possibly switch sides. No tech needed or available. AI isn’t even in the picture.
The learning happens through interaction and negotiation of ideas (as much of life does!).
5. Physical and Experiential Learning
The Research
The more learning involves physical interaction and experimentation, the harder it is to outsource to AI.
Experiential learning theory shows students learn best when they experience, then reflect. Only after the experience can they conceptualize and apply.
Hands-on learning increases engagement and helps connect abstract ideas to real experiences.
Classroom Example
Instead of ONLY explaining physics concepts have tudents build rubber-band powered cars. I did this in school, and I still remember it!
Activities like this test your understanding of distance, friction and design implications.
AI can’t run the experiment. That’s what makes it so powerful of a learning experience.
BONUS: AI Fluency Through “Breaking AI”
The Research
One emerging area is teaching students to interrogate AI itself.
Recent research on AI literacy suggests students should experiment with AI systems to understand their limitations and biases.
When students explore how AI fails or misclassifies information, they develop critical digital fluency and a deeper understanding of technology.
Classroom Example
Students try to trick an AI image classifier by manipulating images or prompts.
Then they discuss things like, why did it fail? What patterns did the AI use? What does this reveal about machine learning?
The best way to understand when we should use AI for learning vs when we shouldn’t is to understand how it works, and build off that knowledge.
