Can Generative AI Help Identify Potential Research Problems?

Can AI Help You Find a Research Problem Worth Investigating?

You may know the general area you want to study without knowing what the actual research problem is. Perhaps you are interested in generative AI in education, climate adaptation, telemedicine, misinformation, employee well-being, or another broad domain. Having an interesting subject, however, is not the same as having a research problem.

This is where generative AI can seem remarkably helpful. Describe a field or phenomenon and it can immediately suggest unresolved challenges, conflicting outcomes, underexamined populations, practical difficulties, theoretical tensions, or unanswered questions.

Some of these suggestions may lead somewhere useful. The difficulty is that an AI system can formulate a convincing problem even when the evidence needed to establish that problem has not been checked.

AI Can Surface Candidate Problems, but It Cannot Establish Them for You

What It Means for AI to “Identify” a Research Problem

A research topic is not yet a research problem

Suppose you tell an AI system that you are interested in “university students' use of generative AI.” That is a subject area. It tells us what you are interested in, but not what requires explanation, investigation, intervention, or better understanding.

A research problem usually involves something more specific: an unexplained phenomenon, an unresolved theoretical issue, inconsistent evidence, a consequential practical difficulty, an inadequately understood relationship, a limitation in current knowledge, or another condition that creates a defensible reason for investigation.

Generative AI can help you move from the first toward possible versions of the second. It should not be allowed to quietly erase the distinction.

AI can generate different ways of problematizing the same phenomenon

A useful capability of generative AI is not simply finding “the problem.” It is showing that the same broad phenomenon can potentially be problematized in several ways.

Consider widespread student use of generative AI. One researcher might investigate changes in assessment validity. Another might examine unequal access to high-performing AI systems. Someone else might investigate student reliance on AI, disciplinary differences, epistemic trust, changes in writing practices, or how teachers redesign assessment.

These are not interchangeable descriptions of one underlying problem. They represent different possible objects of inquiry, assumptions, stakeholders, and bodies of literature.

Using AI to produce such alternatives can therefore extend the broader process of generating possible research ideas. Its value lies partly in preventing a broad interest from becoming prematurely locked into the first problem formulation that comes to mind.

AI can interrogate observations from practice

Research problems do not arise only from journal articles. Researchers may notice something in classrooms, hospitals, organizations, communities, laboratories, public services, or other settings that seems puzzling or consequential.

You might tell an AI system, for example, that instructors at the same university appear to be responding very differently to generative AI. Rather than asking, “What should I research?”, you could ask what distinct research problems might plausibly underlie that observation.

The model might suggest differences in institutional policy interpretation, confidence, professional autonomy, perceived risk, disciplinary norms, assessment practices, or access to professional development.

None of those possibilities has been established merely because the model generated it. But they give you candidate explanations and problem formulations to investigate.

AI can help expose assumptions hidden inside your problem statement

Sometimes the problem is not difficult to find. It is difficult to formulate without smuggling in an unsupported assumption.

Imagine starting with: “The problem is that students' use of generative AI reduces critical thinking.” The statement already assumes an effect that the proposed research may actually need to establish.

You could ask AI to identify assumptions embedded in the formulation. A useful response might point out that “use” encompasses very different practices, that critical thinking needs operational definition, that effects could vary by task or learner, and that the direction of the relationship has been presumed rather than demonstrated.

This is one reason AI can sometimes be more useful when asked to challenge assumptions about a research idea rather than simply generate additional material supporting it.

AI can suggest possible contradictions and tensions to investigate

Another productive approach is to provide the model with actual evidence rather than asking it to rely entirely on its internal knowledge.

If you have notes from several papers, verified findings, policy documents, or other appropriate source material, you can ask AI to identify apparent tensions among them. Perhaps one group of studies reports positive outcomes while another reports negligible effects. Perhaps a policy assumes something that empirical evidence does not clearly support. Perhaps two theoretical perspectives imply different explanations for the same phenomenon.

Here AI is assisting with comparison and synthesis. The underlying evidence remains inspectable.

That is considerably stronger than asking a model, without supplying or checking sources, to tell you what researchers “do not know.”

AI does not automatically know what the literature has established

This limitation matters enormously. A language model can produce statements that resemble scholarly problem statements without having conducted a systematic, current, and reproducible search of the literature.

It may claim that “few studies have examined” something, that evidence is “limited,” or that a particular population is “understudied.” Those phrases are common in academic writing, which makes them easy to generate. Their familiarity does not make them evidence.

If the proposed problem depends on missing knowledge, you are approaching the question of whether there is a genuine research gap. That conclusion requires literature evidence rather than AI assertion.

A real-world difficulty is not automatically a research problem

Generative AI may also identify practical problems: teachers lack training, patients experience delays, employees resist a system, communities have unequal access to a service, or organizations struggle to implement a policy.

These may be important. But research does not become justified merely by attaching a study to a practical complaint.

You still need to establish what is known, what needs explanation, which aspect can be investigated systematically, who is affected, what evidence supports the claimed condition, and what knowledge the proposed study could contribute.

Problem identification should move from possibility to evidence

A defensible workflow separates discovery from substantiation.

AI can help you ask better diagnostic questions

One of the more useful applications may be asking AI for questions rather than answers.

What evidence would establish that this is actually a problem? Who experiences it? Compared with what? What alternative explanations exist? Which concepts are underspecified? What would make the apparent problem disappear? Does the problem occur everywhere or only under certain conditions? Could the framing privilege one stakeholder's perspective over another?

Questions like these force the proposed problem to survive scrutiny before it becomes the foundation of an entire study. A proposal committee will eventually ask similar questions anyway. Better to meet them before Chapter 1 acquires emotional significance.

From “Teachers Are Not Using AI” to a Researchable Problem

The AI did not discover the final research problem. It helped destabilize an overly simple interpretation. The researcher's subsequent engagement with evidence determined whether there was a defensible problem to investigate.

Where AI-Assisted Problem Identification Can Go Wrong

Ask AI to Diagnose Possibilities, Then Verify the Diagnosis

Generative AI is most useful here when it prevents you from becoming attached to one interpretation too early. Give it an observation and ask for competing ways of understanding it. Give it a tentative problem and ask what assumptions you have embedded in the wording. Provide verified findings and ask where genuine tensions might exist.

Then stop treating the AI output as evidence and investigate.

You may eventually need to determine whether the resulting question is scientifically important and feasible to investigate. Those judgments should come after the underlying problem has been adequately established.

Before Accepting an AI-Suggested Research Problem

Frequently Asked Questions About AI and Research Problems

Let AI Suggest Where to Look, Not What to Believe

How to Cite This Guide

This guide is intended to be read, shared, and used in research, teaching, and academic work. If you draw on its ideas, explanations, or other content, please acknowledge the source by citing the guide. Doing so gives appropriate credit and helps your readers locate the original resource.