Generative AI in higher education: How students learn to row with digital peers
Source PublicationScientific Publication
Primary AuthorsRojas, Guerra, Frez et al.
"Working with an AI is like rowing a tandem kayak. If you communicate and match your partner's strokes, you move forward. If you sit back and expect them to do all the work, you just spin in circles."
Rowing a two-person boat
Imagine trying to row a tandem kayak across a wide, foggy lake. You are sitting in the front seat, and your partner is in the back. Because of the thick fog, neither of you can see the entire lake at once. You can only see the obstacles directly ahead, while your partner has the map and compass. If you talk to your partner, share exactly what you see, and match your paddle strokes to their rhythm, you glide smoothly across the water. You reach your destination safely. But what happens if you just sit back, yell out a command to row, and expect your partner to do all the heavy lifting? The boat simply spins in circles. You go absolutely nowhere. Working with artificial intelligence is surprisingly similar. It requires a shared effort to move forward.
Generative AI in higher education
As digital tools rapidly enter university classrooms, educators want to know exactly how students interact with them. A recent study looked closely at Generative AI in higher education. Researchers wanted to measure what happens when students are forced to collaborate with a large language model to solve difficult calculus problems. They set up an experiment with 30 undergraduates. To make the test realistic, the researchers used a 'jigsaw' design. This means they split the information in half. The student had one piece of the puzzle, and the AI had the other. They had to talk to each other to find the final answer. Collaboration was not just an option; it was structurally necessary.
The eight types of digital rowers
The researchers recorded the chat logs and analysed the behaviour of the students. They noticed that human-AI teamwork is not all the same. In fact, they identified eight distinct roles that students took on during the test. Some students acted as 'Co-Regulators' or 'Recalibrating Collaborators'. These students shared their reasoning step-by-step. They negotiated with the AI. In our boat analogy, they were the ones communicating clearly and matching their paddle strokes. Other students became 'Delegating Directors' or 'Low-Effort Guessers'. They tried to force the AI to do the work without sharing their own knowledge. The results were clear. Students in the highly interactive group solved the maths problems 71 per cent of the time. Those in the low-interaction group only succeeded 19 per cent of the time. If students sustain a shared thought process with the AI, then they achieve much better outcomes.
What this suggests for future classrooms
This study measured how students solved specific maths problems, but it suggests something much larger about learning. Simply giving a student an AI tool is not enough. The way teachers design the assignment matters immensely. If educators want students to learn, they must build tasks that make true collaboration necessary. The AI cannot just be an answer machine. It must be a partner in the boat, requiring the student to keep paddling.