How We Turn Big Problems into Research We Can Actually Do

When we mentor student researchers, we do not usually start with "what's your research question?"

We start before that.

For a recent conference held by the Computational Social Science group at UC Davis, I asked the students to start with a topic or issue about artificial intelligence that genuinely interested them.

They came back with broad topics: teenagers, menopause; which, if nothing else, gave us range.

From there, we worked backward in two teams. What was the actual problem? What did we want to understand? Who was affected?

And then, perhaps most importantly, what could we realistically study?

This question was the hardest.

Students often want to design the perfect study: the one that answers everything, includes the ideal sample, and solves the whole problem at once. But research rarely works that way.

We spent a lot of time around a whiteboard drawing out possibilities, identifying constraints, and asking a different question: What is the smallest useful study we can do next?

One project began with a problem in sexual health information for adolescents. Many young people do not receive school-based sex education that feels relevant or inclusive, and AI tools are increasingly available as another place to ask sensitive questions. That raises an important question: how accurate, trustworthy, and inclusive are the answers those systems provide?

Our first instinct was to ask adolescents about their experiences, but that created a serious design problem. Adolescents might not want to share their AI chat history with us, and we would need to factor in many ethical safeguards and support systems based on information we learned. It would be ethically and practically complicated.

So we changed the first step.

Instead of beginning with adolescents, the team is now collecting responses from large language models and evaluating those responses internally against established best practices. That lets us begin assessing how these systems handle sexual health questions and gives us insight into what is being answered well vs. what might need improvement.

The second project started from a very different problem: menopause care.

The students were interested in how AI might improve communication between patients and physicians. But here, too, the first idea ran into a constraint. We are not physicians, we do not work inside a health center, and changing how doctors communicate is a large and difficult intervention.

So we moved the point of intervention.

Instead of trying to change physicians, the students developed a study in which an AI chatbot would help patients prepare for an upcoming appointment. The tool would not diagnose or recommend treatment. It would help users clarify their symptoms, identify priorities, and practice how to communicate concerns and questions before entering the exam room.

The proposed comparison was simple: an AI coaching condition versus a more traditional preparation worksheet.

Both projects were recognized with Best Innovation awards at the conference.

What I liked most about that recognition was not that the projects were about AI. Plenty of researchers are studying AI. It was that the students had learned to take complicated social problems and make them researchable without pretending the constraints did not exist.

That is a skill we try to build in the Youth Media Design Lab.

We often borrow from UX and design-studio approaches when planning research. We define the problem, identify the people affected, map the constraints, generate multiple possible approaches, and sketch them out together before committing to a study.

Sometimes the most creative part of research is not inventing a complicated method. It is finding the first question that is small enough to answer but meaningful enough to move the problem forward.