Responsible use of Generative Artificial Intelligence in the research process
These guidelines are a concise version of "The Living guidelines on the responsible use of generative AI in research", prepared by the European Research Area Forum, composed of European countries and research and innovation stakeholders. We also advise how to utilize Aalto AI Assistant to overcome some of the risks.
Key Principles
The key principles for the responsible use of generative AI in the research process are:
• Reliability in ensuring the quality of research, reflected in the design, methodology, analysis and use of resources. This includes aspects related to verifying and reproducing the information produced by the AI for research. It also involves being aware of possible equality and non-discrimination issues in relation to bias and inaccuracies.
• Honesty in developing, carrying out, reviewing, reporting and communicating on research transparently, fairly, thoroughly and impartially. This principle includes disclosing that generative AI has been used.
• Respect for colleagues, research participants, research subjects, society, ecosystems, cultural heritage and the environment. Responsible use of generative AI should consider the limitations of the technology, its environmental impact16 and its societal effects (bias, diversity, non-discrimination, fairness and prevention of harm). This includes the proper management of information, respect for privacy, confidentiality and intellectual property rights, and proper citation.
• Accountability for the research from idea to publication, for its management and organisation, for training, supervision and mentoring, and for its wider societal impacts. This includes responsibility for all output that a researcher produces, underpinned by the notion of human agency and oversight.
Recommendations for researchers
1. Remain ultimately responsible for scientific output
For generative AI to be used in a responsible manner, researchers should remain ultimately responsible for scientific output.
• Researchers are accountable for the integrity of the content they produce generated by or with the support of AI tools.
• AI systems are neither authors nor co-authors. Authorship implies agency and responsibility, so it lies with human researchers.
• Researchers maintain a critical approach to using the output produced by generative AI and are aware of the tools’ limitations, such as hallucinations, inaccuracies and bias.
The most common types of bias are:
• Training Data Bias: Biases in the data used to train generative AI models can lead to skewed responses, reflecting inaccuracies or systemic biases within the source material.
• Prompt Bias: Models may align their answers with the perceived beliefs or preferences of the user, a phenomenon known as sycophantic behaviour, potentially leading to misleadingly agreeable or biased outputs.
Hallucinations and inaccuracies may include:
• Invented Citations and Incorrect Summaries: Generative AI models may generate plausible sounding but incorrect citations, which can mislead users who rely on those sources for research or verification/n. Even when the citations are correct (the paper exists, has the proper title...), the AI generated summary of those papers may be incorrect. Therefore, researchers are responsible for checking all the references and content of the citations and the summaries.
• Interpretability: generative AI models operate as "black boxes," making it difficult to understand how specific responses are generated. This opacity underscores the importance of cross validation, especially in automated data analysis, where model responses can significantly impact conclusions.
2. Transparency
Use generative AI transparently:
• Researchers, to be transparent, detail which generative AI tools have been used substantially in their research processes. When generative AI meaningfully shapes results, researchers transparently note its use according to the guidelines of their journal or standards in their discipline in the methods section (or equivalent) responsibly evaluating the extent of the contribution. This transparency principle also applies to the use of AI to detect or screen bad practices (i.e. detect hallucinations or plagiarism) while keeping human supervision.
• Researchers disclose or discuss the limitations of generative AI tools used, including possible biases in the generated content, as well as possible mitigation measures. Researchers take into account the stochastic (random) nature of generative AI tools, which is the tendency to produce different output from the same input. Researchers aim for reproducibility and robustness in their results and conclusions.
3. Pay attention to privacy, confidentiality and intellectual property rights
Aalto AI Assistant v. external tools:
• Researchers understand the technical, ethical and security implications regarding privacy, confidentiality and intellectual property rights. Aalto University researchers use Aalto AI Assistant according to its guidance. For personal data or confidential data, choose the Aalto AI Assistant Private Chat Option. The Private Chat Option enables researchers to engage in a private conversation, ensuring the communication remains confidential.
As the Aalto AI Assistant tool is a tool managed by Aalto University and running in Aalto University servers, many types of data that could not be used in external tools can be included in Aalto AI Assistant, see guidance.
• Researchers remain mindful that in external tools, the generated or uploaded input (text, data, prompts, images, etc.) could be used for other purposes, such as the training of AI models. Therefore, they protect unpublished or sensitive work (such as their own or others’ unpublished work) by taking care not to upload it into an external AI system unless there are assurances that the data will not be re-used, e.g., to train future language models or to the untraceable and unverifiable reuse of data.
Read more on AI and copyright
4. The output produced by generative AI can be especially sensitive in relation to the protection of intellectual property rights and personal data
• Researchers pay attention to the potential for plagiarism (text, code, images, etc.) when using outputs from generative AI. Researchers respect others’ authorship and cite their work where appropriate. The output of a generative AI (such as a large language model) may be based on someone else’s results and may require proper recognition and citation.
Read more on AI and copyright
• The output produced by generative AI can contain personal data. If this becomes apparent, researchers are responsible for handling any personal data output responsibly and appropriately, and EU data protection rules are to be followed.
5. Continuously learn how to use generative AI tools properly to maximise their benefits, including by undertaking training
• Generative AI tools are evolving quickly, and new ways to use them are regularly discovered26. Researchers stay up to date on the best practices and share them with colleagues and other stakeholders.
• Researchers aim at minimising the environmental impact of generative AI, by evaluating whether and which AI tool is best suited for the intended task and by using the most effective prompting techniques.
6. Refrain from using generative AI tools substantially in sensitive activities
Refrain from using generative AI tools substantially in sensitive activities that could impact other researchers or organisations for example in peer review and evaluation of research proposals, etc.
• Avoiding the use of generative AI tools eliminates the potential risks of unfair treatment or assessment that may arise from these tools’ limitations (such as hallucinations and bias).
• Safeguard the original unpublished work of fellow researchers from potential exposure or inclusion in an AI model (under the conditions detailed above in the recommendation 3.).
The Finnish National Board on Research Integrity TENK has published guidance on AI and Research Integrity:
Contact persons
