Professor and data scientist Susana Brás, in the Introduction to Machine Learning course of the third year of the Bachelor’s degree in Computer Science at the University of Aveiro (Portugal), turned group work into a learning journey that runs throughout the semester, instead of a product to be handed in at the end. Generative AI is not forbidden. It is encouraged, but with rules.
Students choose an open problem that interests them and develop it in stages, like sprints in a professional setting: phased deliveries, interim feedback from the professor, revision and, finally, a report, an oral debate and peer assessment. Groups plan in advance which tasks they will use AI for, and in the report they must name the tools, justify their choice, explain the criteria for using them and describe how they verified the results. The principle is simple: AI can support research, programming, writing or analysis, but the intellectual responsibility for the work remains with the student.
This course project runs alongside the classes, usually starting three weeks into the semester, once students are already familiar with the subject and content. It proceeds more or less in parallel and autonomously, with scheduled checkpoints and preliminary validation deliveries. The last 3 to 4 classes of the course are hands-on and entirely dedicated to the project, with the professor moving between all the groups to discuss and follow their work.
The professor presented the guide and explained that the work would be organized into sprints, each with a delivery and feedback, and that changes between versions were not only allowed but expected.
She also presented the rules for using AI: it may be used, but it has to be planned, declared and verified. She explained the two mandatory sections of the final report (Discussion and Critical Analysis, and AI Tools) so that, from day one, the groups knew they would have to justify every decision.
The groups began exploring possible problems and data sources that interested them.
The starting point is a problem open enough to allow students to make choices. Whenever possible, they are encouraged to select a topic or problem that interests them and to define a suitable strategy to study it. This freedom of choice came with guidance from the professor and clear criteria:
- choices must be technically grounded
- choices must be consistent with the defined objectives
- choices must be critically analyzed
The groups followed the same rule used in class: plan first, then execute, and consult when needed. Planning included the tasks in which each group expected to use AI, and why.
Transparency about the use of AI thus becomes part of the work. Depending on the context, it may include identifying the tools used, the tasks in which they supported the group and even a record of relevant interactions. The underlying principle is that AI can support research, exploration, programming, writing or analysis, but it should not replace understanding the problem, interpreting the results or taking responsibility for the decisions made.
In the interim delivery, the professor left comments, questions, suggestions and doubts. The purpose of the feedback was not to confirm whether the group had found the “right” solution, but to help students identify weaknesses and alternatives
Connecting the project and classroom learning
The project runs alongside regular classes, allowing new content to be gradually incorporated into the work.
“As new methodologies are discussed in class, students can question earlier decisions, try out alternatives and rethink their approach. The project thus becomes a unifying element of the course, rather than a standalone activity carried out only at the end of the semester. This integration aims to foster less fragmented learning.” Professor Susana Brás
One of the core principles of this approach is to shift part of the attention from the final result to the process that led to it.
The second delivery
In this second delivery, the groups presented their chosen methodology and preliminary results, explaining why they had chosen certain data, methods and evaluation approaches.
The professor again commented and asked questions, pushing students to think beyond the quantitative score. Getting a technically correct result was not enough. They had to understand why their proposed solution worked, under what conditions it would stop working and what other options they could have considered.
“At first it was a challenge, but the students came to understand why it mattered.” Professor Susana Brás
The last 3 to 4 hands-on classes of the course were entirely dedicated to the project. The professor moved between all the groups to discuss decisions, follow their progress and challenge their reasoning with questions such as:
“Does this decision make sense for the problem?”
“What alternative could have been considered?”
“What is the most important limitation of this approach?”
The groups worked on the report, which had to take into account the feedback from the interim deliveries.
The professor did not give up on the written report, because knowing how to write and organize information remains essential. The document had to tell a coherent story, supported by evidence from the data, and include two mandatory sections:
- The Discussion and Critical Analysis section describes what the group learned and understood, the limitations, what could not be avoided, the decisions that were not unanimous and, above all, why.
- The AI Tools section names the tools used, but above all justifies the choice, the criteria for using them and the verification strategy. It may include a record of the most relevant interactions, what worked and what produced unacceptable results.
The professor valued the reports in which students showed the ability to identify, understand and correct errors across the versions of the document they submitted. Changes between versions of the work are not only allowed but expected, and valued as evidence of learning.
Each group of students presented its work to the class:
- outlined the problem
- explained the key decisions
- interpreted the results
- acknowledged limitations and possible alternatives.
This was followed by a discussion with the professor and classmates. Students who were not presenting participated by asking questions, offering comments, and suggesting alternatives.
Peer evaluation was also conducted by classmates, and everyone had to be able to follow each group’s strategy and line of reasoning. This required communicating complex ideas clearly and evaluating not only the technical results but also the reasoning, collaboration, and the way AI was used.
The final assessment aims to reflect the overall maturity of the project, rather than just the quality of a single metric, graph, or isolated result.
Conclusion
Over the semester, during academic year 2025/2026, students brought together content they would otherwise have seen as isolated techniques, made decisions on problems with no single solution and learned to justify them. AI went from taboo subject to object of reflection: students were no longer afraid to say they had used it and began to explain how they used it, what worked and what did not. The value of the project lay not in reaching a perfect solution, but in building a well-grounded solution, understanding its limitations and communicating that reasoning clearly.
The Professor’s Reflection
“One change I made regarding AI was simple… I don’t forbid it, I encourage it, but with rules. I believe this method has had some impact because they stopped being afraid to say they used it and started explaining how, what worked and what produced unacceptable results.
Why use this approach? The main goal is to turn group work from a product to be handed in into a learning journey.
By creating room to experiment, make mistakes, receive feedback and revise without every stage being immediately tied to a grade, the aim is to encourage a more reflective relationship with learning.
At the same time, having a shared problem over several weeks provides a context in which students can integrate content that might otherwise be seen as unrelated.
Finally, by requiring students to explain and defend their own decisions, including how they used artificial intelligence tools, the aim is to reinforce an essential idea: the intellectual responsibility for the work still belongs to the student.
The value of the project, therefore, does not lie in reaching a perfect solution. It lies in the ability to build a well-grounded solution, understand its limitations, learn from the process and communicate that reasoning clearly.” Professor Susana Brás
