Dealing with generative AI tools in scientific work
With the following information, BTU Cottbus-Senftenberg provides academic staff with guidance on the use of generative AI tools in the context of academic work. BTU Cottbus-Senftenberg does not fundamentally reject the use of AI tools in the university context, but supports their sensible, critical, reflective and responsible use without neglecting problematic implications such as copyright and data protection law and research ethics issues. It follows the DFG's guidelines for the use of generative models for text and image creation as well as the general principles of good scientific practice, in particular the principle of transparency as a principle of scientific honesty.
The use of generative AI tools should be based on professional and methodological reflection and risks as well as ethical and legal aspects should be adequately considered. Due to the current highly dynamic nature of the topic, the following information will be updated regularly and the creation of a university-specific guideline with rigid specifications and regulations will be avoided for the time being. BTU is also currently reviewing the legally compliant provision of generative AI tools for BTU teaching staff and employees.
Generative artificial intelligence (AI) is a special form of AI and includes technology that can independently generate new content, for example in the form of text, audio and video files, images or code. It does not actually create new data, but merely reassembles or applies existing information from the underlying data.
AI tools can be used in a variety of ways in scientific work. For example, they can help to optimize scientific work processes, solve complex problems more efficiently, gain new insights, develop innovative solutions and improve research results. In addition to commercial tools, there are now more and more open source offerings (e.g. locally usable). The choice of an AI tool also depends on the intended use. Sites such as https://www.futurepedia.io/,https://www.hcilab.org/ai-tools-directory/,https://www.advanced-innovation.io/ki-tools and https://theresanaiforthat.com/ provide an overview of AI tools . However, errors, superficialities, bias and so-called hallucinations can never be ruled out with the AI tools presented, so the results should always be checked critically. In addition, the field is developing very quickly, which means that new AI tools and possibilities are constantly being introduced. Before using the non-binding AI tools presented, it is also advisable to check the costs, terms of use and data protection.
Possibilities for using AI tools in scientific work:
- Literature research and analysis: help with identifying, collecting and analyzing relevant scientific literature to present the state of research on a specific topic and formulate new research questions
Examples of AI tools: Elicit, Research Rabbit, Perplexity, Semantic Scholar, SciSpace, Consensus, Iris.ai, Keenious, ChatPDF - Experimental design and execution: AI-supported simulations and modeling tools can help to plan, conduct and analyze experiments in order to test hypotheses and gain new insights
Examples: Google DeepMind's "AlphaFold" algorithm; deep learning models to predict the activity and toxicity of new compounds in drug research - Data analysis and interpretation: analyzing large amounts of data, automating repetitive tasks, recognizing patterns and trends and deriving insights, helping to test hypotheses and uncover correlations
Examples of AI tools: Formula Bot, IBM Watson Analytics, DataRobot, RapidMiner, Julius, TensorFlow, H2O.ai, KNIME - Text generation and editing: Help with writing, revising, formatting and translating scientific texts to communicate research results clearly and precisely and to convert complex topics into easily understandable formats.
Examples of AI tools for text generation: ChatGPT, Claude, Gemini (formerly Bard), Perplexity, Neuroflash, AlephAlpha, Mindverse, Copycockpit, Writesonic, h2oGPT
Examples of AI tools for text correction: Grammarly, DeepL Write, Trinka, Plag, Quillbot
Examples of AI tools for translating and transcribing: DeepL, Google Translate, Sonix, Buzz Captions, Trint.
Examples of AI tools for paraphrasing: QuillBot, PolitePost, ChatPDF
Other examples of image generation tools are: Adobe Firefly, Stable Diffusion, DALL-E and Midjourney. Audio and videos can be created with the following AI tools, for example: Soundraw, D-ID, Synthesia, Runway.
When handling personal data of third parties in connection with AI-based tools, compliance with data protection law, in particular the General Data Protection Regulation (GDPR), the Federal Data Protection Act (BDSG) and the Brandenburg Data Protection Act (BbgDSG), must be ensured. This applies both to the upload of information to AI-based tools and to the content generated by these tools. Personal data may only be entered into generative AI tools if the software operators do not make this data accessible to third parties or use it as training data. In addition to data protection, the input of confidential information, sensitive research data or internal documents is also not permitted.
It is recommended to preferably use data-saving AI tools and, if possible, to adjust the data protection settings so that, for example, chats are not saved and chat histories are not used as training data.
If researchers use the results of AI-based tools in their own work after checking the content, the users are responsible for any incorrect or distorted content generated by the AI, incorrect references, copyright infringements or plagiarism.
Authorship of AI-generated output
AI-supported programs for text production cannot be considered authors or creators of the text they generate within the meaning of the German Copyright and Related Rights Act (UrhG); users of such programs can. The decisive factor here is a significant degree of intellectual contribution.1
In addition, automated image or music generation technologies do not generate copyrightable creations, meaning that images and music created in this way are "in the public domain" and not protected by copyright. Copyright protection only arises for a natural person if a person plays a significant role in the production of the work of art (so-called "level of creation"). Technology cannot be protected by copyright.2
Uploading copyright-protected materials to AI-based tools can constitute a copyright-relevant act, meaning that legal regulations must be observed in this respect.
Labelling requirements for AI-generated text in an academic context
An obligation to label AI-generated text may arise from the terms of use of a software as well as from applicable examination regulations and framework regulations of a university (e.g. if the indication of any aids is prescribed).1
In order to make the type of use of AI tools transparent, passages of the work that are based on the use of AI tools must be marked accordingly in accordance with the framework conditions, stating the source and the type of use (e.g. use for brainstorming, for creating the outline, for developing/optimizing software source texts, for linguistic optimization, for creating text passages, etc.).
OER licensing
AI-generated text that is created without significant human influence is considered to be in the public domain under the Copyright Act. If users of AI software can claim copyright for an AI-generated text, licensing as an Open Educational Resource (OER) is possible. However, it must be ensured that the AI-generated text does not contain any copyright-protected content.1
AI output in the public domain is inherently open: copyright does not stand in the way of sharing, copying and remixing content in the public domain.3 If there is a copyright on the adaptation of AI-generated output by a personal intellectual creation, the adaptation should be licensed as openly as possible for further use as OER. The Creative Commons Zero (CC0) license, which is based on the public domain, or the license with attribution (CC BY) are suitable here.2, 3
Good scientific practice
As a rule, the marked adoption of AI-generated text will not formally constitute a violation of the rules of good scientific practice.1
The use of AI-based tools must not violate the principles of good scientific practice, in particular the requirement for transparency as a principle of scientific honesty. It must be recognizable to third parties which parts of a work were generated by an AI and to what extent. As a result, there is a documentation obligation, i.e. the AI-based tools used to create the work must be listed in accordance with the specifications, e.g. under "Methods" or "References" or in the acknowledgements or documented in the appendix of the work as follows:
Name of the AI tool, software version, date of retrieval if applicable, URL if applicable, prompt used and result if applicable (transcription or screenshots).
Publications:
1 Peter Salden, Jonas Leschke (eds.): "Didaktische und rechtliche Perspektiven auf KI-gestütztes Schreiben in der Hochschulbildung", Center for Science Didactics at Ruhr-Universität Bochum, 2023, https://doi.org/10.13154/294-9734
The first section of the document provides an introduction to the topic, briefly outlines the technical background and then focuses on didactic aspects. The second section - "Legal opinion on the use of AI software in the university context" - answers the legal questions raised in the introduction (including authorship, labeling obligations, scientific misconduct).
2 "AI and OER: How well do they fit together?", Georg Fischer, April 24, 2023, iRights.info, https://irights.info/artikel/kuenstliche-intelligenz-und-open-educational-resources/31872
3 "OER and CC licenses for generative AI", Fabian Rack, 15 November 2023, iRights.info, https://irights.info/artikel/oer-cc-lizenzen-generative-ki/32090
Declarations of independence, the rules of good scientific practice and examination regulations usually already contain provisions that are applicable to the use of AI tools. Existing declarations can be supplemented on a subject-related basis to specify whether and, if so, to what extent and under what conditions AI tools can be used.
When using and entering data, persons working in science as employees of BTU must comply with the General Data Protection Regulation (GDPR) and the existing confidentiality regulations and non-disclosure agreements. This also applies to contractual confidentiality obligations (non-disclosure agreement (NDA), confidentiality agreement (GHV), confidentiality/confidentiality regulations in cooperations, R&D contracts, etc.). Users are responsible for the conscientious, critical and reflective use of AI tools. To this end, they should:
- Obtain information on departmental requirements,
- not enter any third-party personal or sensitive data or confidential or BTU-internal information and adjust data protection settings (if possible),
- inform themselves about the data storage and use of each individual AI tool,
- anonymize or pseudonymize data used to train an AI tool,
- critically review all results of an AI tool (also with regard to intellectual property),
- reference and document the use of AI tools in accordance with the guidelines,
- generally follow the DFG guidelines for dealing with generative models for text and image creation as well as the principles of good scientific practice.
As a rule, the following minimum documentation should be provided when using AI-based tools: Title (for text, image and multimedia generation tools, the prompt (input of the person using the tool) is considered the title), name and version of the tool, provider, date of generation of the content, address (URL of the tool).
Various AI tools are being tested at the BTU's IKMZ as part of the KI@MINT project; a separate list of recommendations is not (yet) available. There is also an internal IKMZ working group on the topic of AI tools and licenses, which is working on the question of a central provision of tools at the BTU. If you have any questions, pleasecontact the head of the Multimedia Center, Mr. Boguslaw Malys(contact details).
As part of the BTU's scientific continuing education program, regular events are offered on the use of generative AI in a scientific context. Information on current offers can be found on the Research Department's intranet page, the Graduate Research School (GRS) website and the ZWW website.
In September 2023, the German Research Foundation (DFG) published initial guidelines for dealing with generative models for text and image creation. As a starting point for continuous monitoring, the paper is intended to provide orientation for researchers in their work as well as for applicants to the DFG and those involved in the review, evaluation and decision-making process.
Framework conditions to ensure good scientific practice and the quality of scientific results:
- Transparency and traceability of the research process and the knowledge gained for third parties
- Scientists themselves are responsible for adhering to the basic principles of scientific integrity. The use of generative models cannot release scientists from this responsibility in terms of content and form.
- When making their results publicly available, scientists should disclose whether and which generative models were used , for what purpose and to what extent .
- Only the responsible natural persons can appear as authors in scientific publications. They must ensure that the use of generative models does not infringe any third-party intellectual property and that no scientific misconduct occurs, for example in the form of plagiarism.
- The use of generative models in the preparation of examiners' reports is not permitted with regard to the confidentiality of the review process. Documents provided for review are confidential and in particular may not be used as input for generative models.
On March 20, 2024, the European Commission published guidelines on the use of AI in research and innovation . Although generative AI tools offer speed and convenience in the creation of text, images and code, researchers must also consider the limitations of the technology, including possible plagiarism, disclosure of sensitive information or inherent biases in the models. Building on the principles of research integrity, the recommendations address the key opportunities and challenges and provide guidance to researchers, organizations and funders for a common approach across Europe. As generative AI is constantly evolving, these guidelines will be updated with regular feedback from the scientific community and stakeholders.
Living Guidelines for the responsible use of generative AI in research are provided (Go to factsheet).
"Nature "and the other Springer Nature journals have added two principles to the publication guidelines:1
- ChatGPT and other Large Language Models (LLM Tools) must not be listed as authors, as they cannot take responsibility for the text.
- LLM tools should be documented under Methods or in the Acknowledgments. If a publication does not contain these sections, the introduction or another appropriate section can be used to document the use of LLM tools.
In addition, the scientific journal has announced in an editorial that it will not publish images and videos created or enhanced using an AI generator, at least for the foreseeable future. The only exceptions are articles that explicitly deal with artificial intelligence.2
1 "Tools such as ChatGPT threaten transparent science; here are our ground rules for their use", Nature 613, 612 (2023), doi: https://doi.org/10.1038/d41586-023-00191-1
2 "Why Nature will not allow the use of generative AI in images and video", Nature 618, 214 (2023), doi: https://doi.org/10.1038/d41586-023-01546-4
Elsevier has also published a policy on the use of generative AI and AI-supported technologies in the writing process.
- AI and AI-enhanced technologies should only be used to improve the readability and language of the paper and not to replace important authoring tasks.
- The entire paper should be carefully reviewed and edited. Ultimately, authors are responsible and accountable for the content of their work.
- AI and AI-enabled technologies should not be listed as an author or co-author, nor should AI be cited as an author.
- The use of AI and AI-assisted technologies should be disclosed in the manuscript and a statement to this effect entitled "Declaration of AI and AI-assisted technologies in the writing process" should appear at the end of the manuscript immediately above the references or bibliography in the published paper.
- Elsevier does not permit the use of generative AI or AI-assisted tools to create or modify images in submitted manuscripts. The only exception is the use of AI or AI-assisted tools as part of the research design or research methods. If this is the case, such use must be described in a reproducible manner in the methods section. This should include an explanation of how the AI or AI-assisted tools were used in image creation or modification, the name of the model or tool, the version and extension numbers, and the manufacturer. Authors should adhere to the specific usage guidelines of the AI software and ensure correct attribution of content.
- The use of generative AI or AI-supported tools in the creation of graphics, e.g. for book or commissioned covers or graphic summaries, is not permitted.
For examiners and editors, the use of AI is not permitted, as it violates the confidentiality of authors' and property rights and may violate data protection rights.
"The use of generative AI and AI-assisted technologies in writing for Elsevier", August 18, 2023, https://www.elsevier.com/about/policies-and-standards/the-use-of-generative-ai-and-ai-assisted-technologies-in-writing-for-elsevier
The editorial policies for science journals with comparable requirements can be found here: https://www.science.org/content/page/science-journals-editorial-policies (section "Image and Text Integrity").
- Literature and further articles on good scientific practice (GWP) and artificial intelligence (AI) on the website of the Ombudsman for Science: https://ombudsman-fuer-die-wissenschaft.de/12365/gute-wissenschaftliche-praxis-und-kuenstliche-intelligenz/
- "Proposals for declarations of independence for the possible use of AI tools" by the German Association for University Didactics (dghd), Annette Glathe, Ass. Jur. Jan Hansen (TU Darmstadt) and Martina Mörth, Anja Riedel (BZHL) as part of the dghd topic series "KI in der Hochschullehre" (2023), status: 25.08.2023
The suggestions may be passed on and used provided the authors are named. The PDF file linked above is also available as a PowerPoint presentation. - "Overview: Contributions in the HFD to ChatGPT in studying and teaching", Lisa Hoffmann, Hochschulforum Digitalisierung, blog entry from 20.03.2024, https://hochschulforumdigitalisierung.de/chatgpt-in-studium-und-lehre/
Contact
PD Dr. rer. oec. habil. Ines Brusch
Head of Department
T +49 (0) 355 69-5500
ines.brusch(at)b-tu.de
Dr. rer. nat. Katrin Weise
DFG officer
T +49 (0) 355 69-2716
katrin.weise(at)b-tu.de