Propose an idea or join an available project. Prepare the one-page note below and bring it to a discussion. Your topic does not need to be polished or proven novel; reading and a small investigation will help shape it.

1. Choose a Starting Point

• Your own idea: Explain the problem you have noticed and why it interests you.
• An existing project: Identify the project and the part you would like to explore. You can begin with a project-interest note if the research question is not yet clear.

Review our Research Areas and note the skills you already have, what you need to learn, your degree, and the time available.

2. Prepare a One-Page Idea

Use this AIT Brain Lab discussion outline for your initial note. For formal proposals, follow your programme’s requirements; see Resources for templates.

Include a working title, your name, degree, and proposed project area. Keep the main note to about one page; list references separately if needed.

It is fine to write “I do not know yet.” Distinguish what you have verified from your assumptions. Do not invent references, results, or claims of novelty.

For example, “build an AI sales agent” is a starting direction. A more focused question could be: “Does explicitly stating product limitations improve customers’ understanding and decision quality compared with an agent that only describes benefits?” This is an illustrative question; its novelty and feasibility still need checking.

Example: Download a one-page sales-agent idea (PDF). It illustrates the outline above; the study is proposed and its novelty still needs checking.

3. Discuss and Refine Together

Bring your note and the closest papers to the first project discussion. We will consider the importance of the problem, overlap with prior work, fit with the lab, available resources, and an appropriate scope.

• PhD: Work toward an original contribution and a research direction you can increasingly lead.
• Master’s: Define an identifiable piece of work, usually within a PhD or grant project, that meets your learning goals and degree requirements.

The outcome may be a revised question, a short literature review, a baseline reproduction, or a small feasibility study. We will agree on the next deliverable and when to review it.

4. Working Together

Meeting Frequency

• Early stage: weekly. Use frequent discussions to refine the question, assess prior work, and plan a feasible study.
• Once the direction and plan are established: every two weeks. Review evidence, resolve bottlenecks, and agree on the next steps.
• Later, when the work is progressing independently: on demand. Request a meeting when a decision, difficulty, or important milestone needs discussion.

Agree changes in meeting frequency together. Ask for help promptly when stuck.

Read, Question, and Write Early

Read strong, relevant papers from ACL, CHI, NeurIPS, and specialist venues. Examine the question, study design, and evidence supporting each claim.

For your first reading discussion, bring a comparison of three close papers, selected with the supervisor if needed. For each, record the question, main evidence, limitation, and what it changes about your idea. Trace their references backward and newer papers citing them forward. Search neighbouring terms and relevant specialist venues too; venue reputation alone does not establish that a claim is correct.

Finish with one paragraph: “These studies establish …; they leave … unresolved; answering this matters because …; our first useful test would be ….” Three papers are a manageable starting exercise, not proof that a gap exists. Keep updating the comparison as you find closer work.

Begin drafting the first three chapters early, following your programme’s required structure. These will usually cover:

1. Introduction: Background, problem, research questions, objectives, and expected contribution.
2. Literature review: A critical synthesis of the closest work and the gap your project addresses.
3. Methodology: The proposed approach, study or system design, evaluation, analysis, and feasibility.

Revise these chapters as you learn. Distinguish planned methods and expected contributions from completed work and findings.

Bring Your Thinking to Meetings

Before each meeting, prepare:

• Progress and evidence: What did you try, and what did you find?
• Interpretation: What do you think the evidence means? What remains uncertain?
• Bottleneck: What is preventing progress, and what have you already tried to resolve it?
• Proposed next steps: What do you recommend doing next, and why? Bring alternatives when useful.
• Support needed: What decision, feedback, or resource do you need from the supervisor or team?

Bring your interpretation and proposed action, even when the answer is uncertain.

When the Project Gets Stuck

Our first approach is to keep the same overall topic and adjust the project so it can continue. We may narrow the question, change the method, use more suitable data, simplify the system, or revise the evaluation. Together, we choose a feasible next step that preserves a meaningful contribution and meets the degree requirements.

Agree on a short review point and the evidence needed to choose the next step. If repeated attempts cannot answer the question, revisit the assumption or approach together. Staying with the topic should not mean repeating an uninformative experiment indefinitely.

Before Booking a Proposal or Defense

Send the report to the supervisor before booking your proposal presentation or final defense. Allow time for final comments and any necessary revisions. Agree on the review timeline with the supervisor before making booking arrangements, and follow the programme’s formal procedures and deadlines.

Keep a Small Research Record

Keep papers, claim–evidence comparisons, and meeting decisions in one research decision note. Use only the sections needed now.

A reference manager such as Zotero can organise papers and citations. Git can track code and changes. Write in the programme’s required format, using Word or Overleaf as appropriate. Start with tools you can use reliably; a complicated setup is not a research contribution.

For computational experiments, record data splits, model and code versions, settings, and failures. For human studies, keep the protocol, consent and ethics records, sampling decisions, and analysis notes. Keep protected data in approved storage.

Use AI With Judgment

AI can synthesise papers and organise evidence. You must judge what is missing, what may be novel, and why the question matters. Verify sources and explain your decisions in your own words. See AI for Research.

Shared Responsibilities

Agree on communication, feedback expectations, responsibilities, access to resources, and milestones. Discuss individual contributions and authorship early and revisit them as the work develops.

Use the relevant Scorecard when it helps identify the next one or two priorities. Consult the Research Guide for practical help with ideas, novelty, methodology, and writing.

Ideas Behind This Approach

Chaklam’s How to choose a good topic argues for a narrow, feasible problem. How to find related work efficiently teaches citation tracing. Essence of great research connects novelty, significance, feasibility, validity, and clarity. Use these as questions to reason through, rather than numbers to maximise.

Further perspectives: Hamming asks whether a problem matters and has a plausible route forward; Uri Alon connects problem choice to feasibility, interest, and student development; Karpathy emphasises a focused contribution and learning from research discussion. DARPA’s Heilmeier questions are useful for explaining a project’s purpose, difference, risks, resources, and success checks in plain language.

These are perspectives to adapt to your project. They do not guarantee acceptance or prescribe one method for every kind of research. Check the target venue’s current guidance: for example, CHI’s contribution types and ACL Rolling Review’s reviewer guidelines.