Gains:
- Ability to apply the principles of informed consent, nonmaleficence, and confidentiality to social research
- Ability to properly anonymize data by clearing all direct and indirect identifiers before giving it to artificial intelligence
- Ability to query the data retention policy of the artificial intelligence tool used and use only approved media for sensitive data
Social research is done with people, and this comes with a heavy responsibility. People tell you about their lives, their opinions, sometimes their most fragile experiences. Research ethics is a set of rules for maintaining this trust: not harming the participant, obtaining their consent, protecting their identity, and not misusing their data. In the age of AI, this responsibility has grown even more, because pasting participant data into an AI tool can now mean unknowingly leaking that data. In this unit, you will learn the basic principles of research ethics, how to protect confidentiality when using AI, and how to properly anonymize data (making the person unrecognizable).
Three basic principles are the backbone of every social research. Informed consent: the participant must understand what the research is and how his data will be used and say "yes" freely. Non-maleficence: research should not cause physical, psychological or social harm to the participant. Confidentiality and confidentiality: participant's identity and data must be protected. AI tools can facilitate these policies (e.g. drafting consent text, anonymization control) or compromise them (e.g. uploading personally identifiable transcript to a cloud). The difference is whether the user is conscious or not.
Step by step: ethical and safe research
1. Obtain ethical approval. In most institutions (universities, institutes) research involving humans requires approval from an ethics committee (the committee that ensures that the research does not harm the participant). AI helps draft the application text; The approval itself belongs to the institution.
2. Write the consent text clearly. The participant should be informed in a language he/she can understand, not in heavy legal language. AI drafts a simple and complete consent text; You confirm each item.
3. Anonymize data before giving it to AI. Name, address, phone number, workplace, and identifying details should be removed. Just deleting the name is often not enough; A description such as "with three children, the only pharmacist in neighborhood X" also gives away the person.
4. Know which tool stores data and how. Data you paste into an AI tool is processed on that tool's servers and may be used in training in some services. For sensitive data this may be unacceptable; Use tools with data protection assurance that are approved by your institution.
5. Store data securely and delete it in a timely manner. Comply with the retention period and deletion commitment you promised in your consent.
Caution: Pasting raw participant data containing personally identifiable information (named transcript, contact information, sensitive personal data) into any AI tool is a serious ethical and legal violation. Anonymize the data first; If you cannot anonymize it, never give that data to an external tool.
three mini cases
Case 1 — Insufficient anonymization. A researcher deleted names from the transcripts and gave them to the AI. But one participant said, "Last year, as the only female inspector in the municipality..."; This definition alone gave the person away. By telling the AI to “find identifying details in every transcript,” 11 such latent identifiers emerged and were generalized.
Case 2 — Consent text simplified. One team's consent form was written in heavy legalese; Participants were signing without understanding. YZ reduced the text to the 8th grade reading level and made each article in plain Turkish. Participants now truly understood and gave consent; The ethics committee also approved it.
Case 3 — Vehicle choice averted crisis. A PhD student was about to upload sensitive interviews with victims into a general AI tool to summarize them. His advisor reminded that the data was not anonymized and the vehicle's data retention policy was unknown. The data was first fully anonymized and processed only in a data protection-assured environment approved by the institution.
Four copyable templates
1) Anonymization control:
Mark EVERY detail in the text below that could directly or indirectly identify the person: name, location, workplace, unique roles (“one of a kind”), date, relationship, disease, rare characteristics. Suggest an anonymous response to each (e.g., “Participant 7,” “County X”). Text: [here]
2) Informed consent draft:
Draft a plain-language (8th grade level) informed consent text for a study. It includes: purpose of the research, what will be done, duration, voluntariness and right of withdrawal, how data will be stored/deleted, confidentiality assurance, communication. Avoid legal jargon. Subject: [x]
3) Ethical risk screening:
My research design: [summary]. Participants: [who]. List possible ethical risks: potential for harm, power imbalance, vulnerable group, privacy risk, consent issues. Give a mitigation recommendation for each risk. If there are fragile groups, indicate special points of attention.
4) AI vehicle safety control:
I have the following type of data: [type]. Come up with a checklist of security questions I should ask before processing this in an AI tool: where is the data stored, is it used in education, is it anonymous, is there institutional approval, is it sensitive data. Suggest decision criteria.
Weak prompt / Strong prompt
Weak prompt:
Summarize this interview record: "I am Ayşe Yılmaz, a teacher in Kadıköy, my wife was last year..." [text with full name, ID]
This leaks credentials directly to the external tool — a serious privacy violation. The data is not anonymized at all.
Powerful prompt:
Summarize the text below. The text was pre-anonymized: names were replaced with codes like "Participant 3", locations were replaced with generic phrases like "County X", identifying details were omitted. Adding a new identifier; do not make an inference that would violate anonymity.Text: [anonymised text]
The difference: in the second approach, the participant's identity does not reach the tool at all; Privacy is protected from the start.
Ethical principles and AI practice
principle
Meaning
risk with AI
protection
Informed consent
Informed free consent
—
Simple consent text
do no harm
No harm to the participant
misattribution
Loyalty audit
Privacy
Identity is protected
data leak
Anonymization
Data security
Safe storage
cloud processing
Approved vehicle
honesty
No fitting
hallucination
source verification
Common mistakes
- I think I just deleted the name and anonymized it. Indirect descriptors ("the only one") give the person away.
- Pasting the raw ID data into the external tool. It is one of the most serious privacy violations.
- Writing the consent text in jargon. Consent that is not understood is not real consent.
- Not knowing the vehicle's data policy. Find out where your data is stored and used.
- Treating ethical approval as a formality. The ethics committee protects the participant; take it seriously.
In summary
The heart of social research is responsibility to the participant. Informed consent, nonmaleficence, and confidentiality are the backbone of every study. AI; It can assist in consent text simplification, ethical risk screening, and anonymization control. But giving raw ID data to an AI tool is a serious violation; Completely anonymize the data first, clear indirect identifiers, use only secure tools approved by your institution, and keep every promise you make. Ethics is not an adornment of research, but a prerequisite.
Application task
Prepare a short sample transcript (edit yourself) that includes identifying information. Find and generalize all direct and indirect identifiers from the AI with the "anonymization control" template; Count how many secret identifiers come out. Then, create a simple consent text using the "informed consent draft" and have a friend read it to test whether it is truly understandable.
checklist
- [ ] I have checked the ethical approval requirement and process.
- [ ] I wrote the consent text in simple, understandable language.
- [ ] I cleared all direct and indirect identifiers before giving the data to the AI.
- [ ] I learned the data retention policy of the AI tool I use.
- [ ] I used only approved, secure media for sensitive data.