Question, not a theme
“The ethics of AI” is a subject area. A strong question gives you something specific to investigate and judge.

The best EPQ topic is not the most impressive-sounding one. It is the one you can investigate with focus, evidence and enough curiosity to stay with it.
Six tests before you commit
A strong question creates a manageable investigation. Use these checks before falling in love with the wording.
“The ethics of AI” is a subject area. A strong question gives you something specific to investigate and judge.
Set a population, place, period, case study or type of evidence. A smaller question usually produces deeper analysis.
Check that credible books, research, data or primary material exist before committing. Access matters as much as interest.
Avoid a title that only invites description. “To what extent…?”, “How effective…?” and “Should…?” can work when the criteria are clear.
Decide whether the project needs literature review, data analysis, comparison, experiment, design or an artefact—and whether you can do it safely.
Your experience and possible degree subject can shape the direction, but the title should remain an open question rather than a pre-decided answer.
Start broad, then choose the boundary that makes the research possible. You do not need to add every boundary to the final title, but you should know what is in and out.
What do you keep reading, noticing or arguing about?
What is uncertain, contested, changing or difficult to explain?
Are you judging effectiveness, causes, ethics, impact, interpretation or design?
Which place, population, period, case study or technology will you examine?
What sources or data could change your mind?
Can you complete this safely, ethically and within the time and access you have?
Illustrative starting points
These examples show how a broad interest can become a focused question. Do not copy them unchanged. Adapt the scope, evidence and method so the project reflects your interests and your centre’s requirements.
Mental health and social media
To what extent should evidence about sleep disruption shape social-media guidance for 16–18-year-olds in the UK?
How to make it yours: Define the age group, outcome and evidence base. Avoid trying to prove that one factor causes every mental-health outcome.
Artificial intelligence and law
How adequate is current UK copyright law when generative AI systems are trained on creative works?
How to make it yours: Choose one jurisdiction and one legal tension. Separate what the law currently says from what you think it should say.
The cost of university
How has the English student-finance system influenced participation among lower-income applicants since 2012?
How to make it yours: Set a period and population, then identify data that can test the claim rather than relying on opinion.
Bias in algorithms
Which technical and governance measures are most effective at reducing bias in facial-recognition systems?
How to make it yours: Specify the system and criteria for “effective”. Consider building a small evaluation framework rather than a large product.
Public memory
To what extent did museum interpretation of the British Empire change between 2000 and 2025?
How to make it yours: Select a manageable sample of museums and a consistent way to compare interpretation over time.
Sustainable cities
Could heat-pump adoption reduce household emissions in a typical pre-1945 UK terrace without worsening energy costs?
How to make it yours: Fix the building type and assumptions. Use published performance data and explain where modelling is uncertain.
The generator uses your subjects, possible university course and interests to produce adaptable research directions. You receive the initial results immediately—no email gate.