Large Language Models (LLMs), such as ChatGPT and GitHub Copilot, are already part of the tools some students use for programming. They can generate code, explain concepts, detect errors, or help solve problems. But what makes a student decide to incorporate them into their learning?

A recent study analyzed this question among 311 computer science students from a public and a private university in Mexico. The researchers focused particularly on how the ease of use of LLMs relates to students’ perceptions of their usefulness and how frequently they use them.

The idea arose when Anas Wajid, a postdoctoral researcher at the Institute for the Future of Education (IFE), began teaching students to program and realized how widespread the use these tools was.

“It surprised him because it was very different from the way he had been taught,” says Claudia Camacho, a researcher at the IFE, a professor at the School of Engineering and Sciences (EIC), and one of the authors of the article.

Ease of Use Encourages the Adoption of LLMs

The results showed that the easier LLMs were to use for programming, the more effective these tools were perceived to be. Ease of use was also associated with greater trust in them, a stronger intention to continue using them, and more frequent use.

The strongest correlation they found was between ease of use and perceived benefit. This means that when interacting with these tools requires little effort, they tend to feel more comfortable experimenting with them and see them as a useful support for their programming activities.

“In this way, we verified with evidence what we intuitively thought was happening,” Camacho says.

However, the ease of use of a tool can also have a downside. The authors point to risks such as over-reliance on the tool and the possibility that students will incorporate AI-generated code without understanding how it works.

“It’s something that must be taken into account, that we must maintain certain fundamental skills,” the researcher reflects.

The overuse of LLMs can become a risk to programming learning because students may delegate to AI problem-solving processes that they need to develop themselves.

Over-reliance can lead to the acceptance of generated code without understanding or properly verifying it, which can weaken the retention of fundamental concepts.

Therefore, they argue that the goal of universities should not simply be to allow or prohibit the use of these tools, but to teach students how to use them critically.

The authors warn that ease of use—which encourages higher frequency—must be accompanied by strategies that force students to reason, verify, and explain the solutions generated by AI.

Using AI While Continuing to Learn

“LLMs are everywhere now, including in one of the most promising careers: programming,” says Camacho.

In the case of programming, they recommend incorporating activities in which students have to explain and verify the AI-generated code, document how they used these tools, and justify the decisions they made. They also propose that assessments consider the reasoning process and not just whether the final code works.

But beyond adapting the use of these models to the discipline to which they are applied to and developing institutional guidelines, codes and regulations for their use, there must be an education focused on the use of artificial intelligence.

In this education, which can begin in childhood, students must learn to use it consciously.

“Just as we learn to read and write, we should have basic literacy in what AI is, recognizing its ethical, environmental, and intellectual implications.”

However, the study has a significant limitation: its results are based on students’ perceptions, not on a direct measurement of how much they learn using LLMs. Furthermore, the participants came from only two universities, so the results cannot be generalized to all Mexican students.

Thus, the research does not prove that ChatGPT improves programming skills on its own. What it does suggest is that ease of use can become a driver of adoption.

When these tools are easy to use, students tend to trust them more, perceive greater benefits, and use them more frequently.

Thus, the educational challenge lies in ensuring that this ease does not replace learning, but rather accompanies it.

Did you find this story interesting? Would you like to publish it? Contact our content editor to learn more: marianaleonm@tec.mx