PHDE0067 Chap.4 Research Ethics and AI
Research Ethics and AI
Research Ethics and AI treats ethical reasoning as part of every stage of psychological research. Recruitment and consent must account for power, understanding and a genuine ability to refuse. Procedures should minimise avoidable harm and plan withdrawal, confidentiality and debriefing. Data handling raises questions about access, retention, re-identification and secondary use.
Scientific integrity requires transparent analysis, honest limits and accurate attribution. Classic controversial studies show why safeguards developed, while AI-supported and co-constructed research create changing risks around opaque processing, unequal error and accountability.
The chapter links each specific risk to a proportionate safeguard and asks whether the design produces knowledge worth the burden it places on participants.
What this chapter covers
- 01
Voluntary informed consent
- 02
Participant risk and withdrawal
- 03
Privacy and confidentiality
- 04
Debriefing and justified withholding
- 05
Research integrity
- 06
AI-supported analysis
- 07
Co-constructed research and accountability
Worked example · free
Review external AI processing
- 2Identify consent scope, privacy, access and retention risks.
- 2Explain possible analytic error, bias and opacity in generated themes.
- 2Propose approved processing, data minimisation, validation, documentation and renewed consent where required.
Key terms
- Informed consent
- A voluntary decision based on sufficient understanding of the relevant research.
- Debriefing
- A post-participation process that explains the study and addresses relevant effects or withheld information.
- Data minimisation
- Limiting collection, access and retention to information needed for the justified purpose.
Research Ethics and AI FAQ
Is signed consent enough for every later data use?
No. Consent has a scope shaped by what participants understood. New processing, external access or substantially different analysis may require additional review, stronger safeguards or renewed permission.
Why is a weak study also an ethical problem?
Participants may bear time, inconvenience or risk without a reasonable prospect of useful knowledge. Sound methods, honest analysis and calibrated conclusions are therefore part of respecting their contribution.
Who is responsible when an AI tool makes an error?
The researchers remain accountable for the evidence and interpretation. They must validate outputs, protect data, document the tool's role and reject generated claims that cannot be grounded independently.
How do I turn an ethical principle into a reasoned judgment?
Name the plausible harm, who bears it and the study feature that creates it. Then propose a safeguard matched to that mechanism and assess whether the remaining risk is proportionate to the knowledge aim. Consider less risky alternatives, participant understanding, withdrawal, data flow and accountability. For AI-supported work, add independent validation, bias checks and a named human decision-maker.
Finish with a conditional decision and the evidence that would change it; this demonstrates balancing rather than merely listing positive ethical vocabulary.
Exam move
Make an ethics table with rows for recruitment, consent, procedure, withdrawal, debriefing, data handling, analysis and communication. In each row, name the plausible harm, the person or group who bears it, and the safeguard that specifically addresses it.
Use classic studies to practise balancing knowledge value with participant experience, but avoid treating historical context as an automatic excuse or modern rules as unexplained verdicts. For AI scenarios, add columns for data destination, access, retention, validation, bias and human accountability.
Ask whether the proposed use remains inside the understood consent and whether the team could explain the transformation to a reviewer. Build counterexamples in which a positive-sounding safeguard does not solve the actual problem. In essay work, turn ethical principles into conflicts and proportionate judgments. In MCQ work, locate the stage of research before selecting the issue.
Finish by explaining how a stronger ethical design can also improve candour, trust, measurement and interpretability. Extend the table with an evidence column that distinguishes a possible risk from a demonstrated harm and a responsibility column naming who can change the design. Work through a deceptive-study scenario twice: first from the researcher's information needs and then from the participant's reasonable expectations.
Decide whether withholding information is necessary, proportionate and followed by a debrief that can genuinely restore understanding. For digital data, sketch the route from collection to storage, analysis, sharing and deletion. At each transfer, record whether identifiers remain, who can gain access and whether the purpose has changed.
Apply the same discipline to an AI-assisted task: specify the input, proposed output, validation check and human decision that follows. Do not let a fluent output stand in for evidence of accuracy. Construct a bias example by asking who is represented in the training or evaluation data, whose errors carry the greatest cost and which subgroup results must be inspected.
Then propose a safeguard that can be tested rather than a vague promise of fairness. For revision, pair principles that can conflict, such as openness and confidentiality or scientific value and participant burden. Write a qualified decision with a condition under which it would change.
This produces ethical reasoning rather than a list of rules and keeps responsibility attached to concrete choices throughout the research lifecycle.
Working through Research Ethics and AI in PHDE0067? Sia is AskSia’s AI Psychology tutor — ask any PHDE0067 Research Ethics and AI question and get a clear, step-by-step explanation grounded in how PHDE0067 is taught and assessed. Read this chapter free, then take your hardest questions to Sia.