Beyond the Zero-Sum Debate: Designing for Agency and Efficiency in the Age of GenAI

In the bustling world of education, a quiet tension has long existed between two ways of thinking about how people learn. On one side, there is the belief that students need freedom to explore, make mistakes, and struggle through difficult problems to truly understand a subject. This approach values the messy, often frustrating process of figuring things out as the very engine of learning. On the other side, there is the desire for efficiency, where teachers provide clear structures and step-by-step guidance to ensure students master facts and skills without getting lost or discouraged. For decades, educators and researchers have debated how much help to offer and how much independence to grant, wondering if these two goals could ever coexist. Now, a new force has entered this conversation: generative artificial intelligence. These are computer systems capable of writing essays, solving math problems, and generating ideas in seconds, often making the hard work of learning seem unnecessary. This sudden arrival forces a re-examination of old questions, asking whether such powerful tools will help students learn faster or simply trick them into thinking they know more than they actually do.

Two researchers, Marcela Borge from Pennsylvania State University and Todd Shimoda from Thinkertools, Inc., have stepped into this debate to offer a practical way forward. They argue that the arrival of artificial intelligence does not require us to choose between total freedom and rigid control. Instead, they propose a framework that treats learning activities as having two distinct parts: the core work that students must do themselves to build deep understanding, and the supporting work that can be handled by technology to keep things moving smoothly. Their goal is to show how educators can use these new tools to enhance human thinking rather than replace it, ensuring that students still engage in the necessary struggle that leads to genuine insight.

The researchers begin by identifying a problem they call the "Construction Paradox." This is the idea that learning often requires the difficult, time-consuming work of building knowledge from scratch. When a student writes an essay, solves a complex equation, or designs an experiment, they are not just producing a final product; they are constructing a mental model of how the world works. If a student uses artificial intelligence to generate that essay or solve that equation instantly, they bypass the mental construction. They end up with a polished result, but they have skipped the very process that creates understanding. The paradox is that while students and teachers might feel satisfied with the finished product, the student has actually learned less because they avoided the cognitive effort required to make the knowledge their own. The researchers suggest that if we let artificial intelligence do all the hard thinking, we risk creating a generation that can produce perfect answers but cannot think critically when faced with a new or unfamiliar problem.

To solve this, the authors draw on decades of research in computer-supported collaborative learning, a field that studies how people learn together with the help of technology. They propose a simple but powerful distinction between "focal" and "supporting" tasks. A focal task is the main thing a student needs to learn to do. If the goal is to learn how to argue a point, then the act of constructing that argument is the focal task. If the goal is to learn how to manage a team, then the act of negotiating and coordinating with others is the focal task. These are the parts of learning that must remain under the student's control. They are the parts where the struggle, the errors, and the eventual breakthroughs happen. Supporting tasks, on the other hand, are the background activities that help the focal task happen but are not the main learning goal themselves. These might include formatting a document, organizing data, or looking up a definition. These are the parts where technology can be most useful, stepping in to reduce the mental load so the student can focus their energy on the core work.

The researchers illustrate this with examples from their own past work designing educational software. In one system designed to help students learn science, the software did not tell students exactly what to do or how to solve every problem. Instead, it left the scientific inquiry—the asking of questions, the forming of hypotheses, and the testing of ideas—entirely up to the students, though it did incorporate software advisors that provided general advice on critical parts of the inquiry cycle. This was the focal task, the part that required their full attention and effort. However, the software did provide structured support for the reflection that happened after the experiment. It guided students on how to look back at what they did, directing their attention to specific, theoretically grounded dimensions of collaborative competence and constraining what they could reflect on, so they could focus on learning from their mistakes rather than getting bogged down in the mechanics of writing a report. By separating the two, the system allowed students to experience the full weight of scientific discovery while still receiving the guidance they needed to make sense of it.

Applying this logic to generative artificial intelligence, the researchers suggest that teachers and designers must ask a specific question before using the technology: "What is the main thing we want the student to learn, and what part of this activity is essential for that learning?" If the answer is that the student needs to learn how to think critically, then the thinking process itself must be preserved. The artificial intelligence can be used to generate examples, offer alternative perspectives, or help organize information, but it cannot be allowed to do the thinking for the student. The researchers warn that if we use these tools to simply speed up the production of homework or essays, we are likely to undermine the very skills we hope to develop. The danger is not in using the technology, but in using it to bypass the cognitive work that is necessary for growth.

The paper also emphasizes that this balance cannot be achieved by technology alone. It requires the active involvement of teachers who understand the goals of the lesson and can make real-time decisions about when to let students struggle and when to offer help. In the past, educational software often followed a fixed script, guiding students through a pre-determined path. With generative artificial intelligence, the path is no longer fixed; the technology can respond in unpredictable ways, offering unexpected answers or even incorrect information. This makes the teacher's role more important than ever. The teacher must act as a conductor, deciding when to let the students and the artificial intelligence interact freely and when to step in to guide the conversation. They must create a classroom environment where students feel safe to make mistakes and where the group works together to verify information and solve problems.

Ultimately, the researchers argue that the value of learning lies in the process, not just the product. The frustration of not knowing the answer, the effort of trying different approaches, and the satisfaction of finding a solution are what build deep understanding and the ability to apply knowledge in new situations. Generative artificial intelligence has the potential to be a powerful partner in this process, but only if it is used carefully. It should be seen as a tool to support the human effort of learning, not a replacement for it. By distinguishing between the core work that students must do and the supporting work that technology can handle, educators can harness the power of these new tools without losing the essential human elements of curiosity, collaboration, and critical thinking. The goal is not to make learning easy, but to make it meaningful, ensuring that students emerge from their education not just with a collection of facts, but with the ability to think, create, and solve problems on their own.