Views: 4
Among various directions in European integration concept, some new “unions” constantly appear: recent attempt aims at “integrating” a digital approach to science. It is called “Living guidance on responsible use of generative AI in research”, and is the second version adopted by the ERA Forum this April. The ERA’s “document” is aimed at eliminating risks entailed in using genAI; it also includes recommendations for researchers, the research organisations and funding entities.
Background
Rapid acceleration of AI capabilities, driven by significant advances in widespread data has increased computing power and machine learning techniques in science and research during last couple of years. Most vital achievements took place in the development of digital models (so-called foundation models) through the genAI models trained on extensive amounts of “unlabelled data”. These advancements have given rise to what is presently known as the “General purpose AI, or genAI”, capable of performing a wide array of tasks. For example, they include genAIs which can generate various forms of new content (text, code, data, images, music, voice, videos, etc.), usually based on instructions (also known as prompts) provided by the user. The quality of the output produced by these models is often so high that it is difficult to distinguish it from human-generated content.
As soon as the GPT (aka Generative Pre-training Transformer) in essence is a kind of artificial intelligence, it can be used to assist in computer code generation as a valuable tool for developers who are looking to automate tasks or speed up the development process. This can free up developers/researcher’s time and allow them focusing on more complex and creative tasks. It is to be added that the GPT is a family of AI models built by OpenAI, and ChatGPT is a chatbot that uses GPT.
GPT is basically a description of what the genAI models do and how they work; initially, GPT was made up of only LLMs (large language models); but OpenAI has since expanded this to include a number of models: GPT-4.5, as a large multimodal model (LMM); GPT-4o, as a large multimodal model (LMM) which is smaller and more efficient than GPT-4.5, and GPT-4o mini, as a small language model (SLM).
Source: https://zapier.com/blog/what-is-gpt/
On Forum
The ERA Forum was established in 2022 to support the development and implementation of the European Research Area (ERA) policy agenda. The Forum includes representatives from EU states, countries associated to Horizon Europe, the European Commission, the Committee of the Regions, the Economic and Social Committee, as well as the stakeholders from research and innovation communities.
The main tasks of the ERA Forum are to: – assist the Commission in relation to the implementation of existing Union legislation, programmes and policies; – assist the Commission in the preparation of legislative proposals and policy initiatives; and – coordinate with the EU member states and “exchanging views”.
Source: https://european-research-area.ec.europa.eu/era-forum
Opportunities and risks for researchers
The ERA Forum acknowledged that genAI has already provided numerous opportunities for different research and science sectors. However, it notes, the genAI also “harbors risks, such as the large-scale generation of disinformation, intellectual property and data protection issues and other unethical uses with significant societal and environmental consequences”.
Source and citation from: European Commission, Directorate-General for Research and Innovation Directorate: E-Prosperity, Unit E4 – Industry 5.0 and AI in Science; RTD Publications in www.eu-ai-in-science.Europa.eu.
Alongside various positive aspects, the new digital technologies (including genAI) entail the risk of abuse: i.e. some risks are due to the digital tool’s technical limitations, others have to do with the (intentional or unintentional) use of the genAI in ways that erode sound research practices. Other risks for EU researchers could stem from, e.g. the proprietary nature of some of the tools (for example, lack of openness, fees to access the service, use of input data), as well as concentration of ownership or the undesirable transfer of critical technology and intellectual property.
More in: Council Recommendation of 23 May 2024 on enhancing research security – ST/9097/2024/INIT.
Presently, however, the genAI’s impact on research and various aspects of scientific processes calls for reflection, for example, when working with text (summarizing papers, brainstorming or exploring ideas, drafting or translating). In many respects, these tools could harm research integrity and raise questions about the ability of current models to combat deceptive scientific practices and misinformation.
Recommendations for research organisations
As soon as the main actors in performing research in the EU are institutions and big companies, it’s vital to see what guidelines entail. Among them are four recommendations:
1. = Promoting, guiding and supporting “responsible use” of genAI in research activities. With this in mind, the research organisations shall: a) provide and/or facilitate training for all career levels and disciplines, including for research managers and research support staff, on using genAI, especially (but not exclusively) on verifying output, maintaining privacy, addressing biases and protecting intellectual property rights and sensitive knowledge; b) promote an atmosphere of trust where researchers are encouraged to transparently disclose the use of generative AI without concerns for adverse effects; c) provide support and guidelines to ensure compliance with ethical and legal requirements (EU data protection rules, protection of intellectual property rights, etc.).
*) Disclosure of the use of genAI for assistance when writing may lead to a lower quality score during assessment; hence, research organisations should try to avoid situations like this one. Source: Li, et al. EMNLP 2024. How Does the Disclosure of AI Assistance Affect the Perceptions of Writing? – Proceedings of the Conference on Empirical Methods in Natural Language Processing (2024); in: https://aclanthology.org/2024.emnlp-main.279/.
2. = Actively keep track of the evolution and use of generative AI systems within research organisations. Thus, research organisations remain mindful of the research activities and processes for which they use generative AI to better support its future use*). This “knowledge” can be used: a) to provide further guidance on using generative AI, help identify training needs and understand what kind of support could be most beneficial; b) to help anticipate and guard against possible misuse and abuse of AI tools; c) be published and shared with the scientific community; d) to analyse the limitations of the technology and tools and provide feedback and recommendations to their researchers; and e) keep track of the environmental impact of generative AI within their organisations and promote awareness raising initiatives to help their staff pick the most sustainable option.
*) Nogueira L. A., et.al. (2025). Cutting through the noise: Assessing tools that employ artificial intelligence. – IFLA Journal, 0(0). https://doi.org/10.1177/03400352241304121
3. = Integrating genAI guidelines into the research organisations general research guidelines for good research practices and ethics by: a) using these guidelines as a basis for discussion, research organisations openly consult their research staff and stakeholders on the use of generative AI and related policies; b) applying these guidelines whenever possible; if needed, they could be complemented with specific additional recommendations and/or exceptions that should be published for transparency.
4. = Whenever possible and necessary, implement locally hosted or cloud-based generative AI tools that they govern themselves (or governed by trustworthy third parties, e.g. partner research organizations, the EU or trusted countries). This enables their employees to feed their scientific data into current activity that ensures data protection and confidentiality. Besides, organisations have to ensure the appropriate level of cybersecurity of these systems, especially those connected to the internet.
Conclusion
As the ERA-Forum “document” confirms, “these guidelines intend to set out common directions on the responsible use of generative AI. While non-binding, they should be considered as a supporting tool for researchers, research organisations and research funding bodies”. However, a guided opinion is that the “users of these guidelines are encouraged to adapt them to their specific contexts and situations, keeping proportionality in mind”.
The guidelines also note that they “complement and build on the EU AI policy framework, including the AI Act as well as other policy activities on the impact of AI in science, including the opinion of the Scientific Advice Mechanism (SAM) on AI and a policy brief published by the European Commission’s Directorate-General for Research and Innovation, framing challenges and opportunities.
References to the following three links: = https://digital-strategy.ec.europa.eu/en/policies/regulatory-framework-ai; = https://scientificadvice.eu/advice/artificial-intelligence-in-science; and = https://research-and-innovation.ec.europa.eu/document/download/1e2a4c9c-d3f1-43e9-9488-c8152aabf25f_en .
Besides, there is another vital stakeholder, i.e. All European Academies (ALLEA), which prepared the European Code of Conduct for Research Integrity with a set of principles to produce sound research, including ethical aspects. ALLEA seeks to improve the conditions for research, to provide the best independent and interdisciplinary science advice available, and to strengthen the role of science in society. In doing so, ALLEA channels the expertise of European academies for the benefit of the research community, decision-makers, and the public. Outputs include science-based advice in response to topics that are critical to society as well as activities to encourage scientific cooperation, reasoning, and values through public engagement.
These principles include: a) reliability in ensuring the quality of research, reflected in the design, methodology, analysis and use of resources; b) honesty in developing, carrying out, reviewing, reporting and communicating on research transparently, fairly, thoroughly and impartially; c) respect for colleagues, research participants, research subjects, society, ecosystems, cultural heritage and the environment; and d) accountability for the research from idea to publication, for its management and organisation, for training, supervision, and mentoring, and for its wider societal impacts.
Source: Code of Conduct in https://www.alleageneralassembly.org/wp-content/uploads/2023/06/European-Code-of-Conduct-Revised-Edition-2023.pdf
It is quite interesting that the ALEA has also depicted in 2023 that the research misconduct traditionally defined as fabrication, falsification and/or plagiarism (the so-called FFP) in proposing, performing, reviewing and/or in reporting research results. As a result, there are the following clarifications: a) fabrication – as making up data or results and recording them as if they were real; b) falsification –as manipulating research materials, equipment, images, processes or changing, omitting and suppressing data or results without justification; plagiarism –as using other people’s work or ideas without giving proper credit to the original source. These “misconduct” issues can be easily used in applying genAI to research.