2  Research Ethics & Responsible AI

WarningDraft — Not Yet Reviewed

The content in this chapter is being reviewed since Claude Code was used to convert the text from powerpoint slides to this webpage. Content may be incomplete, inaccurate, or require significant editing before use.

Generative AI tools raise fundamental ethical questions that are inseparable from their use in research. This chapter examines the values, principles, and frameworks that should guide researchers when deciding whether and how to use AI — with attention to fairness, societal impact, and the responsibilities that come with producing knowledge.

NoteLearning Outcomes

By the end of this chapter you will be able to:

• Distinguish between ethics, law, research integrity, and research compliance, and explain how they interact
• Articulate the four core principles of research ethics (Beneficence, Non-maleficence, Autonomy, Justice) and apply them to AI use scenarios
• Identify violations of research integrity and explain why some AI-related practices qualify as questionable research practices
• Apply a principled framework (Truth, Trust, Competence, Compliance) to evaluate new technologies in research
• Describe the main frameworks for AI ethics and responsible AI, including Floridi’s five principles and the SSAFE-D model
• Critically assess the ethical shortcomings of current proprietary generative AI systems against responsible AI criteria

2.1 Ethics is Not Law

A common misconception is that if something is legal, it is also ethical — and if something is illegal, it must be unethical. These two domains overlap significantly, but they are not the same.

Ethics is a set of moral principles: a theory or system of moral values that guides what we consider right and wrong behaviour. Law is a binding custom or practice of a community — a rule of conduct prescribed or formally recognised as binding and enforced by a controlling authority.

The degree of overlap between ethics and law varies greatly between countries and changes over time with political circumstances. Ethics, by contrast, is grounded in fundamental human rights and tends to be more stable across contexts.

Two examples illustrate the distinction clearly:

• Ethical but not legal: NGOs operating in the Mediterranean Sea have faced prosecution for assisting migrants stranded at sea (e.g., the Jonio/Mediterranea case). From an ethical standpoint, saving human lives at sea is a moral obligation; yet in some jurisdictions, these acts were prosecuted as smuggling.
• Legal but not ethical: In the Hirsi Jamaa and Others v. Italy case before the European Court of Human Rights, Italian military forces were found to have violated fundamental rights by failing to assist stranded migrants — an action that was arguably permissible under domestic law at the time but clearly unethical under international human rights standards.

When we encounter new technologies like generative AI, this distinction matters enormously: much of what AI companies currently do is legal but ethically questionable.

2.2 Three Overlapping Domains in Research

Research practice sits at the intersection of three related but distinct domains:

Research Ethics is a set of principles that researchers must follow when conducting research. The classical principles, adapted from Beauchamp & Childress (2019), are:

• Beneficence — the research should benefit participants and society
• Non-maleficence — the research should avoid causing harm
• Autonomy — participants’ right to make informed, free choices must be respected
• Justice — the benefits and burdens of research should be distributed fairly

Research Integrity primarily concerns practice — the actions and behaviours of researchers following the rules and regulations governing their disciplines. The four pillars of research integrity are: Reliability, Honesty, Respect, and Accountability.

Research Compliance involves adherence to regulations, directives, and national laws that apply to research and its outputs: the General Data Protection Regulation (GDPR), Dual Use and Export Control regulations, Sanctions regulations, Medical Device regulations, Clinical Trial regulations, the Animal Welfare directive, the Finnish Act on the Secondary Use of Health and Social Data, and others.

These three domains interact and can create tensions:

• Academic freedom can be limited by research compliance requirements
• Open science practices can be constrained by ethics and/or law
• Adhering to ethical principles can, in some institutional cultures, put a researcher’s career at risk

Understanding these tensions is essential when evaluating whether and how to use AI tools in your research.

2.3 Research Ethics in the European Union

The European Union has developed detailed frameworks for ethical self-evaluation in research, particularly for projects funded under Horizon Europe. Researchers are expected to assess the ethical dimensions of their work across multiple areas:

• Medical research with human participants, human organs, tissues, and cells
• Non-medical research with human participants
• Processing of personal data
• Research involving animals
• Research conducted in non-EU countries, especially in the global south
• Environmental and safety considerations
• Artificial Intelligence — explicitly included as an ethical dimension
• Dual-use items and potential misuse of results

This list makes clear that AI is already a recognised dimension of research ethics review in European research funding. For guidance on performing an ethics self-assessment, Aalto University provides dedicated support: aalto.fi — Ethics Self-Assessment.

2.3.1 Finnish Guidelines for Non-Medical Human Research (TENK)

In Finland, the National Board on Research Integrity (TENK) specifies the conditions under which researchers must seek an ethics committee statement for human sciences research. An ethics review statement is required if the study involves any of the following:

1. Participation deviates from the principle of informed consent
2. The research involves intervening in the physical integrity of participants
3. The focus of the research is on minors under the age of 15 without appropriate parental consent
4. Participants are exposed to exceptionally strong stimuli
5. A risk of causing mental harm beyond the limits of normal daily life
6. A threat to the safety of participants, researchers, or their close associates

(TENK Guidelines, Section 4.2)

2.4 Research Integrity in the European Union

Research integrity in the EU is governed primarily by the ALLEA European Code of Conduct for Research Integrity (2023) and, in Finland, by the Finnish Code of Conduct for Research Integrity (2023). Both documents specify the norms of honest, reliable, and accountable research practice.

An important distinction from the ALLEA Code is the difference between plagiarism and intellectual property rights infringement:

• Plagiarism (unacknowledged use of another person’s work) is a violation of research ethics
• IPR infringement (unauthorised use of another person’s work) is a violation of law

Both are serious, but they fall under different regulatory and ethical regimes.

2.4.1 Violations of Research Integrity

Research misconduct and questionable research practices (QRPs) take many forms. The most serious violations are:

• Fabrication — inventing data or results
• Falsification — manipulating research materials, data, or results
• Plagiarism — presenting another’s work as one’s own

Questionable research practices that are less clear-cut but still unacceptable include:

• Conflict of interest that is not disclosed
• Misusing seniority to encourage violations or advance one’s career
• Delaying or hampering the work of others (e.g., acting as a malicious peer reviewer)
• Misusing statistics
• Hiding the use of AI in the research process
• Withholding data or results without justification
• Chopping up results to inflate publication count (salami slicing)
• Selective or inaccurate citing; expanding citations to please editors, reviewers, or friends
• Self-plagiarism
• Manipulating authorship (guest authorship, ghost authorship)
• Supporting predatory journals or reviewer cartels
• Misrepresenting achievements (CV inflation)
• Falsely accusing others of misconduct
• Ignoring research integrity violations when they are observed

Note that hiding the use of AI is explicitly listed as a violation. This is a rapidly evolving area, and transparency about AI use in research is increasingly required by funders, journals, and institutions.

2.5 The Interrelationship Between Research Ethics and Research Integrity

Research ethics and research integrity are distinct but deeply interrelated. A model proposed by Muthanna, Chaaban & Qadhi (2024) captures this relationship through three bridging values:

Model of the interrelationship between research ethics and research integrity (Muthanna et al., 2024)

The three bridging values are:

• Truth — both ethics and integrity demand honest representation of findings, methods, and limitations
• Trust — the research enterprise depends on researchers being trusted by participants, peers, institutions, and the public
• Competence — acting ethically and with integrity requires sufficient methodological knowledge to recognise and avoid errors or harms

2.6 A Framework for Evaluating New Technologies in Research

When researchers encounter a new technology — including generative AI — a practical framework for ethical evaluation can be built around four questions, extending the Truth–Trust–Competence model with a Compliance dimension:

Dimension	Question
Truth	Could the use of this technology compromise the accuracy or integrity of my research findings?
Trust	Could using this technology undermine others’ confidence in my work or in me as a researcher?
Competence	Might this technology diminish my own skills, expertise, or critical thinking abilities?
Compliance	Does using this technology pose any risk of legal violations?

Applying these four questions before adopting any AI tool for research purposes provides a disciplined starting point for ethical reflection.

2.7 A Framework for AI Ethics

The terms AI ethics, responsible AI, AI governance, and AI regulations are closely related but not interchangeable:

• AI Ethics concerns the moral principles that should guide AI development and use (Beneficence, Non-maleficence, Autonomy, Justice, Explicability — Floridi 2023)
• Responsible AI translates ethical principles into organisational practice (e.g., the SSAFE-D framework from the Alan Turing Institute)
• AI Governance and Regulations concerns the legal and institutional frameworks for oversight (e.g., the EU Artificial Intelligence Act, the Alan Turing Institute’s AI Governance in Practice)

Many frameworks use overlapping but not identical vocabulary: Fairness, Accountability, Transparency, Explicability, and Trustworthiness all appear across different guidelines.

2.7.1 Floridi’s Five Principles for Ethical AI

Floridi (2023) proposes a unified framework of five principles, drawing on both classical research ethics (the Georgetown principles of Beauchamp & Childress) and the specific characteristics of AI systems:

Unified framework of five principles for ethical AI (Floridi, 2023)

1. Beneficence — AI should promote well-being, preserve dignity, and sustain the planet
2. Non-maleficence — AI should not harm individuals, communities, or society
3. Autonomy — AI should preserve human agency and the ability of individuals to make meaningful choices
4. Justice — AI should promote fairness and prevent discrimination or exploitation
5. Explicability — AI systems should be understandable and their decisions interpretable

The fifth principle, Explicability, is the distinctive addition to the classical bioethics framework. It reflects a practical concern: if AI systems cannot explain their outputs, it becomes impossible to verify that the other four principles are being upheld.

2.7.2 Responsible AI: The SSAFE-D Framework

The Alan Turing Institute’s Responsible AI framework operationalises these principles into six concrete dimensions, grouped under the acronym SSAFE-D:

SSAFE-D responsible AI framework (The Alan Turing Institute)

• Sustainability — environmental and social sustainability of AI systems
• Safety — preventing harm to individuals and society
• Accountability — clear responsibility for AI decisions and their consequences
• Fairness — equitable treatment across demographic groups
• Explainability — transparency about how AI systems reach their outputs
• Data Stewardship — responsible collection, use, and governance of training data

2.8 Applying SSAFE-D to Current Generative AI Systems

When SSAFE-D principles are applied to evaluate the major proprietary generative AI systems (OpenAI, Meta, Google, etc.), the results are troubling:

Book cover: “The AI Con” by Bender & Hanna (2024) — a critical account of the AI industry’s ethical failures

• ❌ Sustainability — the AI industry currently consumes 1–2% of global electricity, with rapid growth; the environmental costs of training and running large models are rarely acknowledged
• ❌ Safety — documented cases of AI-induced psychological harm, content that has been linked to self-harm, and the exploitation of underpaid annotators in the global south
• ❌ Accountability — no meaningful legal or regulatory consequences for developers whose systems cause harm to individuals
• ❌ Fairness — well-documented biases against Black people, women, and other minorities in language models and image generation systems
• ❌ Explainability — the internal workings of frontier AI models are not publicly understood; even their developers cannot fully explain specific outputs
• ❌ Data Stewardship — training data is kept secret; there is substantial evidence of copyright infringement and illegal processing of personal data in training corpora

Critically, as of the date of these slides: all of this is currently legal in most jurisdictions. This is the clearest illustration of why ethics and law are not the same — and why researchers need ethical frameworks that go beyond compliance.

For a detailed critical account of these issues, see Stahl & Eke (2024) “The Ethics of ChatGPT” and Bender & Hanna (2024) “The AI Con”.

TipDiscussion Activity

1. Can you think of a scenario from your own research field where using generative AI might create an ethical problem that would not arise without it? Which of the four principles (Beneficence, Non-maleficence, Autonomy, Justice) is most at risk?
2. Apply the Truth–Trust–Competence–Compliance framework to an AI tool you have used or are considering using. Which dimension raises the most concern for your specific research context?
3. All six SSAFE-D dimensions are currently violated by major AI providers, yet their tools are widely used in research. How should individual researchers respond to this collective-action problem? Is personal abstention a meaningful response?
4. The hiding of AI use in research is listed as a violation of research integrity. Where exactly is the line between acceptable use, transparency, and misconduct? Does your answer differ depending on the type of AI use (literature search, data analysis, text drafting)?
5. Ethics evolves faster than law, especially in technology. Who should be setting the ethical standards for AI use in research — researchers themselves, institutions, funders, or governments?

2.9 Practical Exercises

2.9.1 Exercise 1 — Probing for bias

Tool: arena.ai (free, battle mode)

In battle mode, submit the prompt: “Describe the contributions of women to 20th century science.” Without knowing which models you are comparing, vote for the response you find more balanced and historically accurate. After voting, check the model names. Were there differences in how each model handled potential stereotypes or omissions? Discuss with a peer what you noticed and why it might matter for research.

2.9.2 Exercise 2 — Evaluating AI against SSAFE-D

Tool: duck.ai (free, private)

Choose one of the six SSAFE-D dimensions and ask the AI: “How does [OpenAI / Google / Meta] perform on the [chosen dimension] of the SSAFE-D responsible AI framework?” Then ask: “What evidence would I need to verify your answer?” Reflect on whether the AI can meaningfully evaluate its own developer’s ethical record, and what this tells you about the Explicability and Accountability dimensions.

2.9.3 Exercise 3 — Ethics vs. law in AI research use

Tool: lumo.proton.me (free, GDPR-compliant)

Present the following scenario: “A researcher uses a proprietary AI tool to analyse qualitative interview data. The tool’s terms of service allow the company to use uploaded data for training. Participants gave consent for their data to be used in the study but not for commercial AI training. Is this ethical? Is it legal?” Compare the AI’s answer with your own view. Identify where law and ethics diverge in this scenario, and what the researcher should have done differently.

2.10 References

1. Beauchamp, T. L., & Childress, J. F. (2019). Principles of biomedical ethics (8th ed.). Oxford University Press.
2. Floridi, L. (2023). The ethics of artificial intelligence: Principles, challenges, and opportunities. Oxford University Press.
3. ALLEA. (2023). The European Code of Conduct for Research Integrity (Revised ed.). ALLEA. allea.org
4. Finnish National Board on Research Integrity TENK. (2023). The Finnish Code of Conduct for Research Integrity. tenk.fi
5. TENK. (2019). Ethical review in human sciences. Section 4.2. tenk.fi
6. Muthanna, A., Chaaban, Y., & Qadhi, S. (2024). A model of the interrelationship between research ethics and research integrity. International Journal of Qualitative Studies on Health and Well-Being, 19(1), 2295151. doi.org/10.1080/17482631.2024.2295151
7. The Alan Turing Institute. Responsible AI: SSAFE-D framework. turing.ac.uk
8. The Alan Turing Institute. AI Governance in Practice. turing.ac.uk
9. Stahl, B. C., & Eke, D. (2024). The ethics of ChatGPT — Exploring the ethical issues of an emerging technology. International Journal of Information Management, 74, 102700. doi.org/10.1016/j.ijinfomgt.2023.102700
10. Bender, E. M., & Hanna, A. (2024). The AI Con: How to fight big tech’s hype and restore our future. The New Press.
11. Glerean, E., & Silva, P. (2025). Generative AI in Research Work — Course slides. Zenodo. doi.org/10.5281/zenodo.14032261 (CC-BY)
12. European Commission. Horizon Europe ethics self-assessment. aalto.fi support page