Not one studied how people actually learn with AI.
We’re measuring:
• Performance
• Efficiency
• Outcomes
But we’re ignoring:
→ What people do when they don’t know what they’re doing
→ How they iterate with AI
→ How they handle bad outputs
→ How confidence is built… or lost
Let me be more direct:
We’re studying AI like it’s a calculator.
Not like it’s changing how humans think.
So I tested it in my own workflow.
I built a Python agent to process my Google Scholar alerts, rank relevance, and generate daily research briefings:
👉 https://github.com/robazprogrammer/google_scholar_agent
122 papers this week.
Zero that actually examined the learning process with AI.
That’s the gap.
And it matters because AI is doing something education isn’t prepared for:
👉 It’s pushing people into unfamiliar tasks faster than we’ve ever trained them to handle.
Not just faster answers.
Faster exposure to uncertainty.
This is where my work is heading:
AI-Augmented Exploratory Learning (AAEL)
Not:
“AI helps you complete tasks”
But:
→ How humans approach unfamiliar problems
→ How they iterate when the first answer is wrong
→ How they decide whether to trust the output
→ How confidence is built through friction, not avoided by it
Hot take:
If AI can complete your assignment…
you weren’t measuring learning in the first place.
I’m designing a study around this right now:
→ How professionals use AI to solve unfamiliar tasks
→ AI-only vs AI + coaching conditions
→ Measuring workflow, iteration, and self-efficacy
Because the real question isn’t:
“Should we allow AI?”
It’s:
👉 What does learning look like when the answer is always available… but understanding isn’t?
Curious:
Are we studying AI…
or avoiding the harder question of how people actually learn?
Robert Foreman
Doctoral Student, Educational Technology
Central Michigan University