Augmented teachers? What AI can do for educators

Share:

The debate surrounding artificial intelligence in classrooms often revolves around students: the work done with ChatGPT, whether they learn less using it, and how to assess them. This concern about learning is legitimate, and the PISA 2025 results , published in September, have reignited it.

But AI can also help teachers reduce their repetitive workload and reclaim time for what no machine does well: supporting their students. To analyze which uses might be most beneficial, we’ve been experimenting with real students for several years.

Unlimited practice multiple-choice questions

Preparing multiple-choice questions for students to practice is very time-consuming. You have to write the question, the correct answer, and several incorrect but plausible options, and update them every year because students end up memorizing or copying them.

To simplify this task, we developed AIQUIZ , an open-source platform where students choose the topic and difficulty level, and AI generates new questions (tailored to them) with explanations. Over two academic years, 325 students across four subjects used it, generating 38,752 questions . (Of all these automatically generated questions, we found errors in 229 of them, around 0.6%: poorly worded, with the correct answer incorrectly marked, or outside the syllabus).

And did the students find it useful? Most considered the tool helpful, especially for reviewing before exams or checking their understanding of each topic. This is based on the students’ perception; we don’t know if they actually learned more because of it.

It’s likely that some errors went unnoticed, so teacher supervision is still necessary. Even so, the number is low, especially considering we used a small and inexpensive model (GPT-40 mini). Furthermore, the platform provides the teacher with a dashboard showing the topics where students struggle most. The time previously spent writing questions can now be dedicated to reinforcing those specific topics.

An assistant available at any time

The second example goes beyond saving time: AI allows us to do something that was not feasible before.

In educational escape rooms , students, usually in teams, must solve a series of challenges within a limited time, applying what they have learned in class. They therefore constitute a form of game-based learning . They are highly motivating, but have a weakness: if a team gets stuck and no one helps, frustration sets in. When the activity is done remotely and in flexible shifts, the teacher cannot be available at all times to lend a hand.

To fill this gap, we incorporated an AI agent into a database-based escape room we use in four courses. Creating it was as simple as filling out a web form on an open-source platform we developed. We gave it detailed instructions for each challenge and specifically asked it to guide students step-by-step using a Socratic method, without ever revealing the final solution. The result is a chat function integrated into the game itself.

A total of 148 students participated, rating the agent 4.1 out of 5. They said it helped them progress and become less frustrated, and they requested fewer predefined hints than those in the previous course, who played without the AI ​​agent. Upon reviewing the conversations, we did not detect any instances of the agent deviating from the instructions.

The same idea applies to other activities where immediate support is needed, such as practices that students do without the teacher present.

Review materials and streamline paperwork

In the report, we gathered many other uses based on a survey in which we asked professors at the Polytechnic University of Madrid how they were already using AI in their daily work. One very common use is asking it to review materials: presentations, notes, or practice problems. The AI ​​detects errors, points out what is missing from an explanation, and suggests real-world examples that are relevant to students and help them understand the material.

Others use it to prepare practical exercises and assignments, or to brainstorm problem ideas. It’s also increasingly common to request multimedia elements to illustrate the course material: images, videos, simulations, or even an interactive website.

Another promising use is administrative: compiling and organizing evidence for each subject (i.e., documents demonstrating what and how students have learned, such as graded exams, assessed assignments, or rubrics) for the quality assurance processes the university must undertake each academic year, such as annual progress reports or official accreditations. A chatbot configured with the course syllabus can answer frequently asked questions about dates, rules, or submissions and reduce repetitive emails.

The grade is up to the teacher.

There’s one area where AI shouldn’t take over: grading. Assigning a grade requires human judgment. It’s necessary to know how the material was explained in class, what practical exercises the students completed, and how much each student has progressed from their starting point; and in many cases, empathy is essential.

European law follows the same line. The European Union’s Artificial Intelligence Regulation classifies systems that evaluate learning outcomes as “high risk.” They are not prohibited, but must meet strict requirements, including human supervision. In practice and self-learning, as in the two previous examples, AI can be a great ally, always with the teacher in charge.

When properly integrated, understood as a co-pilot, it does not dehumanize teaching: it takes care of the mechanical part and gives us back time for what requires human interaction, which is explaining, accompanying and listening to our students.

Author Bio: Enrique Barra Arias is Associate Professor of Telecommunications Engineering at the Polytechnic University of Madrid (UPM)

Tags: