Research Foundations › Module 2: Asking a good question

From a topic to a question

Lesson 1 of 3 · 2 min

Previous

Most first projects start with a topic: "AI in healthcare", "climate change", "fake news". A topic is a place to look, not a question. You cannot answer "AI in healthcare". You can answer a question.

Narrow it in steps

A useful way is to go from topic to problem to question.

Topic
Example: AI in healthcare
Narrower topic
Example: AI for reading chest X-rays
Problem
Example: Models trained on hospital X-rays often do worse on X-rays from other hospitals
Question
Example: How much does accuracy drop when a model trained on one public chest X-ray dataset is tested on another, and does simple image normalisation reduce the drop?

The final question names what you will measure (accuracy drop), on what (two named datasets) and what you will try (normalisation). Someone reading it knows what the project will do.

Signs a question is still too broad

  • It contains words like "impact", "role" or "effect" with nothing measurable attached.
  • You cannot say what data you would need.
  • It would take a team of experts years to answer.
  • Any result at all could be called an answer.

Signs a question is too narrow

  • The answer is already known and easy to look up.
  • Nobody would care about the result, even people working on the topic.

Read before you fix the question

You will only know if a question is new after reading what has been done. Expect your question to change two or three times as you read. That is normal, and it is a good sign: it means you are learning where the open problems are.

RememberA good question is one sentence long, and a stranger can tell from it what you will measure.