HeFDI Data Week: Please meet AI, our dear new colleague. In other words: can scientists and machines truly cooperate?

HeFDI Data Week 2025 Abstract: As in many professional fields, the question arises in science as to what a future division of labour between AI and scientists could and should look like. With their 'AI Scientists', Lu et al. (2024) promise a fully automated research process carried out solely by large language models. The AI scientist works faster and more cost-effectively than the human counterpart, but so far also without taking (research) ethical concerns into account. In conclusion, the most likely and desirable outcome does not seem to be the abandonment of humans in research, but rather cooperation between AI and human scientists. AI could take over routine and painstaking work and serve as a brainstorming partner. There would be great potential here for all disciplines; the exact form should be set out in subject-specific guidelines. About the HeFDI Data Week: The HeFDI Data Week 2025 is a multi-day online event series which is offered in the context of the nationwide "Digitaltag”. The series is aimed at researchers, teachers, students and anyone who wants to learn more about data management, FAIR data and code. Over the course of the week, various topics, developments and challenges related to research data will be covered, such as tools and offers for disciplines from NFDI consortia, legal aspects of research data management as well as research data management and artificial intelligence. The HeFDI Data Week is a programme of the federal state initiative HeFDI - Hessian Research Data Infrastructures, which is funded by the Hessian Ministry of Higher Education, Research, Science and the Arts (HMWK). DOI-Link: https://doi.org/10.5281/zenodo.15422379 ; Licence information: Creative Commons Attribution 4.0 International (CC BY 4.0) Date Topic Presenter 27. June 2025 Please meet AI, our dear new colleague. In other words: Can scientists and machines truly cooperate? Dr. Thomas Arnold (Technical University Darmstadt) gefördert durch Please meet AI, our dear new colleague. In other words: can scientists and machines truly cooperate? Dr. Thomas Arnold Disclaimer: the talk uses (re-phrased) citations from the literature The AI Scientist: Towards Fully Automated Open-Ended Scientific Discovery 13.08.2024 ● Fully automatic scientific discovery system ● Perform research independently ● Collaborations with leading universities (Oxford, British Columbia) The AI Scientist: Towards Fully Automated Open-Ended Scientific Discovery 13.08.2024 Key Claims Fully AI-driven system Automated research lifecycle Automated peer review Imitating human science community Diverse research areas Cost Efficiency of the AI Scientist Designed to be compute efficient ~$15 for a full paper What do they actually do? Example Prompts: Novelty check You are an ambitious AI PhD student who is looking to publish a paper that will contribute significantly to the field. You have an idea and you want to check if it is novel or not. I.e., not overlapping significantly with existing literature or already well explored. Be a harsh critic for novelty,... Example Prompts: Experiment Runner Your goal is to implement the following idea: {title}. The proposed experiment is as follows: {idea}. You are given a total of up to {max_runs} runs to complete the necessary experiments. You do not need to use all {max_runs}. First, plan the list of experiments you would like to run. ... Closed models vs. open models ● Proprietary models yield the best quality ● Future focus on open models ● Self-improvement ambitions using open models The Evolving Role of Scientists AI will transform science roles Emphasizing AI-human collaboration AI as a new colleague Nature comments Further criticism Papers contain only incremental developments Most probably to be rejected by editors and reviewers “Popularity bias” for references with high citation counts Scientific Research Potential with AI ● Automation of Research tasks ○ automate repetitive tasks ○ allow scientists to focus on critical thinking ● Delegation to AI Scientists ○ “I have 100 ideas that I don’t have time for” ○ Get the AI Scientist to do those Augmenting LLMs with Specialized Tools ● Need for Advanced Techniques beyond LLMs ● Combining LLMs with Symbolic AI ● Logical rules to improve on pure statistics Back to the Technical Roots LLM Revolution in Scientific Discovery Unified models vs. specialized models LLMs show potential in generation, planning, etc. Common training corpora: research papers Metadata connect papers and add paper semantics Future directions: Facilitating trustworthy predictions ● Hallucinations are particularly dangerous in high-stake scientific domains ● Retrieval-augmented generation (RAG) provides LLMs with relevant, up-to-date and trustworthy information ● Cross-modal RAG instead of only text-based What about social sciences? LLMs can also significantly impact social sciences by achieving remarkable performance in representative tasks (Ziems et al., 2024) and serving as agents for social simulation experiments (Horton, 2023). Open questions ●Training data: dataset scale & quality, lack of cross-modal datasets ●Architectures and Learning Objectives: longer sequences, structural information, something better than next word prediction? ●Evaluation: LLMs evaluating themselves? ● Ethics! Ethics ●Data privacy and consent ●Potential misuse of information ●Bias in algorithmic decision-making ●Equitable access -> A collaborative approach involving ethicists, scientists, policymakers, and other stakeholders to develop robust guidelines that ensure responsible and beneficial use of LLMs in science is needed References ● Sakana.AI (2024). https://sakana.ai/ai-scientist/ Blogpost ● Castelvecchi (2024). Researchers built an ‘AI Scientist’ — what can it do? Nature. https://www.nature.com/articles/d41586-024-02842-3 ● Yu Zhang, Xiusi Chen, Bowen Jin, Sheng Wang, Shuiwang Ji, Wei Wang, Jiawei Han (2024). A Comprehensive Survey of Scientific Large Language Models and Their Applications in Scientific Discovery. CoRR abs/2406.10833 ● Qiang Zhang, Keyang Ding, Tianwen Lyv, Xinda Wang, Qingyu Yin, Yiwen Zhang, Jing Yu, Yuhao Wang, Xiaotong Li, Zhuoyi Xiang, Xiang Zhuang, Zeyuan Wang, Ming Qin, Mengyao Zhang, Jinlu Zhang, Jiyu Cui, Renjun Xu, Hongyang Chen, Xiaohui Fan, Huabin Xing, Huajun Chen (2024). Scientific Large Language Models. A Survey on Biological & Chemical Domains. CoRR abs/2401.14656 Thanks! Questions?