💻 computer science

Teaching AI Interactively: An Experience Report in Higher Education

This experience report demonstrates that redesigning an introductory AI course to incorporate embodied, unplugged simulations and collaborative labs—while maintaining traditional assessments—significantly improved student attendance, engagement, and perceived course effectiveness without compromising academic rigor.

Original authors: Jennifer M. Reddig, Scott Moon, Kaitlyn Crutcher, Christopher J. MacLellan

Published 2026-07-21

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: Jennifer M. Reddig, Scott Moon, Kaitlyn Crutcher, Christopher J. MacLellan

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine trying to learn how to drive a car by only reading a manual about engines, gears, and traffic laws. You might memorize the rules, but when you finally sit behind the wheel, the real world feels overwhelming, confusing, and scary. This is often what learning Artificial Intelligence (AI) feels like for students. AI is the branch of computer science where we teach machines to make decisions, learn from patterns, and solve problems that usually require human brains. But the "manual" for AI is heavy with complex math and abstract ideas that can make students feel like they aren't smart enough to join the club.

For a long time, teachers tried to fix this by giving students big projects to build AI programs at home. But the paper argues that the real magic happens in the classroom, before the homework starts. The researchers wondered: What if we stopped just lecturing and started letting students "act out" how AI thinks? They used a method called "CS Unplugged," which is like learning to dance without music first—using your body to feel the rhythm before you ever touch a speaker. By turning abstract math into physical games and group activities, they hoped to make the scary parts of AI feel like a fun, solvable puzzle.

The Experiment: Turning Class into a Playground

The researchers at Georgia Tech took a standard university course called "Introduction to Artificial Intelligence" and gave it a major makeover for a summer session. Usually, this class is a big lecture hall where a professor talks for two hours, and students take notes. For this summer version, they kept the same homework, the same exams, and the same difficult topics (like how computers search for answers or make guesses with uncertainty). The only thing they changed was what happened during the class time.

Instead of just listening, the students played games. For example, to understand how a computer decides the best move in a game, the students played a physical game called "Grid Knockout." They stood in pairs, holding grids of numbers, and took turns erasing half the grid to try to force their partner into a high or low number. They were physically acting out the logic of the computer's brain. After playing, they discussed what happened, and then they wrote the code to make a computer do the same thing. It was a bridge: first, they felt the concept with their hands; then, they built it with their fingers on a keyboard.

What They Discovered

The results were surprisingly clear, even though the students were doing the exact same difficult work as the previous semester.

First, the students in the new, game-filled class showed up more. The data showed that students in the summer session were nearly 7 times more likely to report attending a high percentage of classes compared to the lecture-only students. They didn't just show up because they had to; they showed up because they felt like they were missing out on something fun and useful.

Second, the students felt the class was more effective. Even though the tests and projects were identical, the summer students were about 4 times more likely to say that their homework and exams actually measured what they knew. They felt that the in-class games had prepared them so well that the assignments felt like a fair test of their skills, rather than a random hurdle.

Third, the students felt more confident. In interviews, they described the classroom as a "safe space" where it was okay to be confused. One student said they used to think, "I don't have the skills to be an engineer," but after the class, they realized, "I want to do this now." The physical games broke the scary, complex math down into manageable pieces, making the students feel like they were actually building something rather than just memorizing formulas.

The Takeaway

The paper suggests that you don't have to lower the difficulty of a course to make it better; you just have to change how students experience the learning. By letting students "embody" the ideas—standing up, moving around, and playing games—they built a stronger foundation for the hard math that followed. The researchers found that when students feel a sense of progress and see the value in showing up every day, they engage more deeply and feel more capable. While this was a small summer experiment with about 50 students, it hints that the way we teach difficult subjects might need to be less like a lecture and more like a collaborative workshop where everyone gets their hands dirty.

Technical Summary: Teaching AI Interactively

Problem Statement
Introductory Artificial Intelligence (AI) courses in higher education present significant learning barriers due to abstract concepts, mathematical complexity, and diverse student technical backgrounds. Traditional lecture-based instruction often fails to make probabilistic reasoning, search algorithms, and learning concepts accessible, leading to reduced student engagement, lower self-efficacy, and high attrition rates. While existing research emphasizes project-based learning, these approaches often prioritize out-of-class deliverables over the in-class instructional moment where students first decide if AI is accessible to them. There is a need to investigate how in-class instructional design, specifically using "CS Unplugged" (technology-free, embodied) activities, can support engagement and conceptual understanding without compromising academic rigor.

Methodology
The authors conducted an experience report comparing two offerings of an undergraduate "Introduction to Artificial Intelligence" course at the Georgia Institute of Technology.

Control Group (Spring Semester): A traditional 16-week, lecture-based format with ~300 students. Instruction consisted of large-hall lectures, with remediation available via office hours.
Intervention Group (Summer Semester): A 12-week offering with ~50 students. The curriculum content, assignments, exams, and assessment structure remained identical to the Spring offering. The primary variable was the redesign of in-class time.
Instructional Design: The Summer course utilized a systematic "unplugged-to-plugged" active learning loop:
1. Unplugged Simulation: Students engaged in embodied, collaborative, technology-free activities (e.g., "Grid Knockout" for adversarial search) to physically enact algorithmic decision-making.
2. Discussion & Formalization: The class discussed strategies, followed by a brief introduction of the formal algorithm and pseudo-code.
3. Plugged Implementation: Students transitioned to Jupyter notebooks to implement the algorithm in Python, using scaffolding to bridge the physical simulation to code.
4. Collaborative Labs: Students experimented with parameters in collaborative coding labs.
5. Formative Feedback: Students completed "One Minute Papers" (OMPs) at the end of each session to guide subsequent lesson planning.
Data Collection: The study utilized institutional end-of-course surveys (Likert scale) comparing student ratings of preparedness, attendance, assignment relevance, learning amount, and course effectiveness. Additionally, the authors conducted semi-structured retrospective interviews with 11 students (4 female, 7 male) to explore qualitative experiences.

Key Contributions
The paper contributes a specific instructional framework for higher education AI courses that integrates embodied learning with technical implementation. The core contribution is the demonstration that replacing passive lectures with a structured cycle of unplugged simulation, guided discussion, and collaborative coding can improve the student learning experience while maintaining identical assessment rigor. The study provides empirical evidence on how "concreteness fading"—moving from physical embodiment to abstract mathematical formalization—can be operationalized in a university setting.

Results
Quantitative analysis using ordinal logistic regression revealed statistically significant differences between the two semesters in three key areas, despite identical assignments and exams:

Attendance: Summer students had 6.8 times higher odds of reporting higher physical attendance ( $p < 0.001$ ).
Assignment Validity: Summer students were approximately four times more likely to agree that assignments accurately measured their knowledge and understanding ( $p < 0.016$ ).
Course Effectiveness: Summer students had 3.4 times higher odds of rating the course as overall effective ( $p < 0.05$ ).
Non-Significant Findings: There were no statistically significant differences in self-reported "amount learned," homework completion, or initial preparedness.

Qualitative analysis of interviews identified four themes driving these outcomes:

Active Knowledge Construction: Students valued the hands-on breakdown of algorithms into core components.
Safe, Supportive Environment: Collaborative activities and constant instructor presence reduced anxiety and encouraged questioning.
Sense of Progress: Daily tangible outputs (labs, activities) made attendance feel worthwhile.
Increased Confidence: The scaffolding helped students, particularly those with less technical background, view AI concepts as attainable rather than intimidating.

Significance and Claims
The authors claim that the redesign of in-class instructional time is a critical lever for improving student engagement and confidence in AI without lowering academic standards. The paper argues that when instructional time offers distinct, valuable learning experiences that students cannot replicate independently (such as guided embodied simulations), attendance becomes intrinsically motivated rather than policy-driven.

The study suggests that tightly coupling in-class activities (unplugged simulations, collaborative labs) with out-of-class assessments helps students perceive coursework as a coherent learning trajectory rather than disconnected tasks. The authors modestly note that these results are based on a single pilot at one institution with a smaller class size and significant teaching assistant (TA) support. They posit that scaling this approach to larger classes requires strategic TA utilization for facilitation and the aggregation of formative feedback. The paper concludes that embodied activities and explicit bridges between physical experience and formal notation are essential for helping students navigate the cognitive demands of introductory AI.

Technical Summary: Teaching AI Interactively

More like this