Living Guidelines for Responsible Generative AI Use in Research

Generative AI presents numerous opportunities across various sectors, but it also introduces significant risks, including the large-scale dissemination of misinformation, challenges related to intellectual property and data protection, and other unethical applications with far-reaching societal and environmental consequences. Research, in particular, stands to be profoundly disrupted. AI holds immense potential for accelerating scientific discovery, driving new breakthroughs, boosting productivity, and enhancing the efficiency of research and verification processes. For instance, researchers can leverage generative AI to assist non-native speakers, quickly summarize vast amounts of literature, retrieve and contextualize knowledge, and aid in data analysis coding. However, the technology is also susceptible to abuse, stemming from its technical limitations or intentional misuse that undermines sound research practices. Proprietary aspects of some AI tools, such as lack of openness, access fees, and data usage policies, also pose risks to European research, alongside concerns about concentrated ownership and the transfer of critical technologies and intellectual property.

To ensure the responsible application of generative AI, researchers are advised to adhere to the following: 1. **Remain Ultimately Responsible for Scientific Output:** Researchers are accountable for the integrity of all content produced, whether generated by or with the assistance of AI tools. A critical approach to AI output is essential, recognizing limitations such as bias, hallucinations, and inaccuracies. AI systems cannot be considered authors or co-authors; authorship implies human agency and responsibility. * **Training Data Bias:** AI models can reflect biases present in their training data, leading to skewed or inaccurate responses. * **Prompt Bias:** Models may exhibit sycophantic behavior, aligning responses with perceived user preferences, potentially resulting in biased outputs. * **Invented Citations and Incorrect Summaries:** AI can generate plausible but false citations or inaccurate summaries of existing papers. Researchers must meticulously verify all references and content. * **Interpretability:** The 'black box' nature of AI models makes it difficult to understand their reasoning, highlighting the need for cross-validation, especially in data analysis. 2. **Use Generative AI Transparently:** Researchers should clearly detail the substantial use of generative AI tools in their research processes. When AI significantly shapes results, its use should be noted in the methods section, following journal or disciplinary standards. This transparency extends to the use of AI for detecting bad practices, provided human supervision is maintained. Researchers must also account for the stochastic nature of AI tools, aiming for reproducibility and disclosing any limitations, biases, or mitigation measures. 3. **Pay Particular Attention to Privacy, Confidentiality, and Intellectual Property Rights:** Researchers must be aware that input data, prompts, and generated content may be used for AI model training. Sensitive or unpublished work should not be uploaded to external AI systems without assurances against reuse. Personal data should only be provided to AI systems with explicit consent from the data subject and a clear, compliant purpose. Researchers must understand the technical, ethical, and security implications concerning privacy, confidentiality, and IP rights, consulting institutional guidelines and tool-specific privacy options.

Research funding organizations are instrumental in shaping the landscape of generative AI in research. Their role includes: * **Informing Funding Calls and Criteria:** Incorporate considerations for the responsible use of generative AI into funding calls and evaluation criteria. This could involve requiring applicants to outline how they plan to use AI tools ethically and transparently, and how they will address potential risks. * **Supporting Responsible AI Research:** Prioritize funding for research projects that focus on understanding and mitigating the risks associated with generative AI, as well as those that explore its beneficial applications in a responsible manner. * **Promoting Transparency in Funded Projects:** Encourage or mandate that researchers funded by the organization transparently disclose the use of generative AI in their projects and outputs. * **Developing and Disseminating Best Practices:** Collaborate with other stakeholders to develop and disseminate best practices and guidelines for the responsible use of generative AI in research, ensuring these are accessible to all applicants and awardees. * **Monitoring and Evaluation:** Include provisions for monitoring and evaluating the use of generative AI in funded projects to ensure compliance with ethical standards and responsible research practices. * **Adapting to Technological Advancements:** Stay abreast of the rapid advancements in AI technology and adapt funding strategies and policies accordingly to address emerging opportunities and challenges.

Transparency and accountability are cornerstones of responsible research, especially when generative AI is involved. Researchers must be transparent about their use of AI tools, detailing which tools were used and how they substantially shaped the research process, particularly when AI influences results or analysis. This disclosure should follow established disciplinary norms and journal guidelines. The stochastic nature of generative AI means that outputs can vary even with the same input, making reproducibility a key consideration. Researchers should aim for robust and reproducible results, disclosing any limitations imposed by the AI tools used. Ultimately, human researchers remain accountable for all scientific output. AI systems cannot bear authorship or responsibility. This human oversight is critical for maintaining the integrity of the research process and ensuring that AI serves as a tool to augment, not replace, human judgment and ethical responsibility.

The responsible and effective deployment of AI applications in research requires a broad engagement from all stakeholders. Universities, research organizations, funding bodies, publishers, and researchers at all career stages must actively participate in shaping the discourse around AI. Cultivating a responsible use of AI should become an integral part of the research culture, grounded in shared values. Rules and recommendations must be complemented by widespread engagement to develop a collective understanding and practice of using generative AI appropriately and effectively. These guidelines, while non-binding, serve as a valuable supporting tool for researchers, research organizations, and funding bodies, including those involved in European Framework Programmes. Users are encouraged to adapt these guidelines to their specific contexts, always keeping proportionality in mind, to ensure generative AI contributes positively to the advancement of science and innovation.