The Sentence That Decides Everything (Judy's Question 1)

Chapter 2 Episode 2

1. What is the problem? What are you going to solve?

This marks the beginning of everything. The first requirement, no matter which research field you are targeting, is to define a specific, valid problem in a concrete, clearly defined context, also known as the research question (RQ). When someone says, “I want to study human-AI collaboration,” or “I want to investigate interdisciplinary team collaboration,” it indicates the person’s research topic interest, but not a problem that can be solved.

The way you frame the problem determines which research community you are speaking to. A challenge in AI-supported clinical decision-making for diagnosis, for instance, could be studied through a machine learning lens (e.g., developing predictive models that make accurate risk predictions from patient data), a health informatics lens (e.g., evaluating the impact of AI integration on diagnostic accuracy and patient outcomes), or an HCI lens (e.g., designing interfaces that align with clinicians’ cognitive processes and do not impose additional cognitive burden). The framing of the RQs differs in the vocabulary they use, the analytical lens they apply, and the aspects they value most, so it is critical to ensure you are asking a question that falls within the scope and interests of your target domain and is posed through the lens that domain recognizes — otherwise, it will be desk-rejected without doubt.

Every Word in the RQ Matters

The research question requires a level of precision that most junior researchers underestimate. Specifically, you need an exact description of the context you are focusing on, the specific variables or factors of interest, and the evaluation outcomes or dimensions you are investigating, and nothing beyond that. Why? Because each term in the RQ carries weight, and readers perceive it as the critical information for defining the scope of your work. As a result, anything you put in the research question that is not within your scope of interest, such as a single vague or misplaced word, will introduce bias and invite confusion about your scope, your contribution, or the type of evidence readers should expect.

Moreover, the vocabulary you use should match the vocabulary your domain uses in a way that people (in this domain) can understand the RQ consistently and appreciate its novelty and significance in the way you expect. If your RQ uses language that the community does not recognize or asks a question that the community does not value, you should go back to the highly relevant papers published at top-tier venues in your target domain and study how they formulate their research questions, what aspects the community cares about, and how the vocabulary maps to the concepts you are trying to express.

Novelty and Significance

Getting the wording right is necessary but insufficient, because the question should also be worth asking and investigating. Two simple rules apply here.

First, you need to confirm that the question has not already been addressed by prior work, since a question you believe to be novel may already have been studied in a venue or context beyond your initial search. Leveraging the literature review methodologies I shared before, this can be relatively easy to address as long as you are following the guidelines.

Second, the question needs to be significant enough to be worthy of a scientific investigation. By arguing “significance,” I want to be clear that I am not saying the problem needs to be big, bold, or revolutionary. Instead, it means that the problem should be motivated in a logically coherent way that persuades readers it leads to a substantial goal or benefit. For instance, knowing what type of information clinicians collect for informed diagnosis is a necessary prerequisite to the subsequent question of knowing how they use that information in the reasoning process to make a decision, and the latter question is a necessary prerequisite to designing AI-assisted technology that provides the right assistance at the right time.

RQ Was Not Created in a Single Attempt

There is a common misconception among junior students that they want to lock in the RQ as soon as possible so they can move on to the actual work quickly. I would view this as the wrong optimization goal, and I often tell my students to spend more time and effort in the question formation stage to ensure the knowledge underlying your RQ is as concrete as possible. “Slow is smooth, smooth is fast,” a saying famously associated with U.S. Navy SEALs, shares the same mental framework. It will always cost you and your collaborators more if you rush toward near-term goals, even when those goals feel productive, because mistakes made at this stage often require starting over.

You will iterate and refine the formulation of your RQ as you work through the subsequent questions in the chain, for instance, as you figure out more precisely what has already been addressed by prior work (Q3) and scope down to a narrower, specific context (Q4 and Q5). And even if you have started the study design or are halfway through the study, it is still possible — and very likely — that a teammate raises a critical question about your motivation or that you find a highly relevant paper you missed at the beginning. When that happens, you need to revisit the questions here, do more literature review, and potentially revise the RQs, along with the motivation, significantly.

Given the prevalence of AI tools today, I would recommend using them for the purpose of critiquing your motivation and RQs by asking them to role-play as experienced researchers in your field and assess and critique objectively. I personally prefer substituting “objectively” with “harshly” to make sure I am prepared for the worst cases.

From Q1 to Q2

Once you have a working version of the RQs, even one you know will iterate, the next question in the chain asks who cares about the problem you just defined and why the stakes are high enough to justify the work. The next post covers Q2 and Q3 together, since the stakeholders and the landscape of prior work are tightly connected: the people who care about the problem are the same people whose prior solutions you will review and whose gaps you will identify.