Launching a Controlled Classroom Test (Image Credits: Pixabay)
La Crosse, Wisconsin — Concerns over artificial intelligence in higher education often center on students bypassing genuine learning through quick answers. Yet economists at the University of Wisconsin-La Crosse explored a different path. They developed a specialized AI tool for their macroeconomics course that prompted critical thinking instead of providing solutions outright. Their spring 2025 experiment revealed promising results for how technology can reinforce education.
Launching a Controlled Classroom Test
The study involved 140 undergraduates, primarily first- and second-year students, across four sections of an introductory macroeconomics class. Course content, assignments, and exams remained uniform for all participants. Instructors prohibited AI use and peer collaboration during in-person tests, ensuring scores captured individual comprehension without external aids.
Following the initial exam, researchers divided sections randomly into four study conditions. One group continued solo work without AI support. Another collaborated in peer groups sans technology. A third used the AI tool independently, while the fourth combined it with group discussions. This setup allowed direct comparisons of study methods’ impact on performance.
Crafting an AI That Challenges Minds
Researchers built the tool, dubbed Macro Buddy, via ChatGPT’s custom features, disabling web access to limit it to course-specific materials like lecture transcripts and slides. Unlike general chatbots, Macro Buddy functioned as a Socratic guide, posing targeted questions to nudge students toward insights.
Consider a query on falling prices boosting consumer spending. Rather than explaining purchasing power directly, the AI inquired about price drops’ effects on affordability. Students then articulated connections themselves, fostering step-by-step reasoning. This approach contrasted with answer-dispensing bots, which prior research linked to weaker retention once access ended.
Such prompting mirrored proven cognitive processes that solidify knowledge through effort. By withholding ready solutions, Macro Buddy compelled active engagement with concepts.
Clear Gains Emerge in Exam Outcomes
All groups saw average scores dip on the second exam after format changes. Differences sharpened by the third test, weeks into the new routines. Those pairing Macro Buddy with peer sessions achieved the top averages.
Students using the AI solo outperformed the no-AI individual group. Peer-only collaboration yielded modest gains relative to others. The combined method likely thrived as AI honed personal understanding while group dynamics demanded clear articulation and peer scrutiny.
- Solo without AI: Baseline control, limited progress.
- Groups without AI: Incremental improvement.
- Solo with Macro Buddy: Notable score uplift.
- Groups with Macro Buddy: Peak performance.
Shaping AI’s Role in Future Learning
Fears persist that AI erodes critical skills by handling tough cognitive work. Surveys from 2025 showed nearly 90% of over 1,100 U.S. college students employing generative AI for tasks from drafting to concept clarification. Critics highlight risks beyond cheating, including diminished deep learning.
This experiment countered such worries. Properly configured AI, emphasizing inquiry over output, paired with social interaction, bolstered results. General chatbots often default to direct responses, but targeted designs shifted engagement toward active problem-solving.
Peer elements introduced accountability absent in solo AI use, exposing diverse viewpoints. Researchers detailed findings in a paper available via SSRN, underscoring implementation’s importance over the tool itself.
These insights suggest AI need not undermine education but can amplify it through deliberate structure. Educators might adapt similar tutors across disciplines, prioritizing reasoning prompts and collaborative reinforcement. What approaches have you seen succeed with AI in learning? Share in the comments.
Key Takeaways
- AI tutors excelling at questioning outperform answer providers in sustaining student gains.
- Combining AI guidance with peer discussion yields superior exam results.
- Design choices determine whether AI supports or supplants critical thinking.
