To spark ideas and offer inspiration, we have gathered a handful of reading suggestions related to the breakout session themes. These are by no means comprehensive — just a starting point for anyone curious to explore further. Reading the literature is absolutely not required to participate; these suggestions are simply there for those who would like to dive in a bit deeper.
If you have ideas for other resources we should include, we warmly welcome your suggestions.
Quick Read:
- Popular Science Overview:
Quanta Magazine’s “Science, Promise and Peril in the Age of AI” series examines how AI transforms scientific research by improving hypothesis generation, experiment design, and data analysis, while questioning the scientist’s role. Split into Input, Black Box, and Output, it covers AI’s origins, neural networks, and impacts on science and math through articles, expert insights, and interactive explainers. Readers can explore in-depth reporting, a video profile, and an AI glossary to grasp AI’s role in reshaping science. Follow the link:
https://www.quantamagazine.org/series/science-in-the-age-of-ai/ - Why is science slowing down?
The paper "Artificial Intelligence in Science: Promises or Perils for Creativity?" attributes the slowdown to the exhaustion of easy discoveries, the growing burden of complex knowledge, and inefficiencies in modern science systems (Bianchini et al., 2025). It explores how AI, increasingly adopted across 80 scientific fields since the 2010s, enhances creativity by boosting novelty and impact, particularly in fragmented knowledge spaces, though its effects vary significantly by field and region.
Citation: Bianchini, S., Di Girolamo, V., Ravet, J., & Arranz, D. (2025). Artificial Intelligence in Science: Promises or Perils for Creativity? European Commission, Directorate-General for Research and Innovation, Publications Office of the European Union.
https://data.europa.eu/doi/10.2777/6693925
Deeper Read
Breakout Session 1: Scientific Reasoning and Hypothesis Generation with AI
- A Survey on AI-assisted Scientific Discovery: The following paper explores how advanced AI language models are transforming scientific research. These models support researchers in several key areas, including enhancing literature searches, generating novel research ideas, and improving the efficiency of experimental workflows. They also assist in creating both textual and visual content—such as scientific diagrams—and can even contribute to automated peer review processes.
Citation: Eger, S., Cao, Y., D’Souza, J., Geiger, A., Greisinger, C., Gross, S., Hou, Y., Krenn, B., Lauscher, A., Li, Y., Lin, C., Moosavi, N. S., Zhao, W., & Miller, T. (2025). Transforming Science with Large Language Models: A Survey on AI-assisted Scientific Discovery, Experimentation, Content Generation, and Evaluation. arXiv. arXiv:2502.05151.
https://arxiv.org/abs/2502.05151
Breakout Session 2: AI for Experimental Design and Laboratory Automation
- Autonomous experimentation: Self-driving laboratories combine automated hardware with machine learning to autonomously design, run, and analyze experiments. By optimizing experimental decisions in real time, they dramatically reduce the cost and duration of discovery in fields like chemistry and materials science. These platforms enable high-throughput, reproducible research while minimizing human intervention. Their data-driven workflows accelerate innovation and open new possibilities for tackling complex scientific problems.
Citation: Epps, R. W., Volk, A. A., Ibrahim, M. Y. S., & Abolhasani, M. (2021). Universal self-driving laboratory for accelerated discovery of materials and molecules. Chem, 7(10), 2541–2545.
https://doi.org/10.1016/j.chempr.2021.09.004 -
AI-driven instrument design: Artificial intelligence is transforming the development of scientific sensors and instrumentation by autonomously exploring design spaces that extend beyond conventional human approaches. Traditional methods are often constrained by established paradigms, whereas AI can uncover novel configurations that substantially improve measurement performance. In gravitational wave detection, for instance, AI has proposed interferometer designs that surpass next-generation baselines by several factors in sensitivity and efficiency. This paradigm shift toward AI-assisted design opens the door to next-generation instruments across the natural sciences—from physics and chemistry to biology, earth science, and astronomy.
Citation: Krenn, M., Drori, Y., & Adhikari, R. X. (2023). Digital Discovery of interferometric Gravitational Wave Detectors. arXiv. arXiv:2312.04258v1
https://arxiv.org/abs/2312.04258
Breakout Session 3: AI-Enhanced Modeling and Simulation
- Emulating complex simulations: Many scientific fields rely on computationally expensive simulations to model complex systems, limiting their use in tasks that require speed, such as real-time analysis, parameter exploration, or uncertainty quantification. Deep learning–based emulators offer a powerful alternative by approximating these simulations at a fraction of the cost. Recent methods can achieve high accuracy even with limited training data, making them broadly applicable across domains. By replacing slow simulations with fast surrogates, emulators open the door to more efficient experimentation, analysis, and discovery.
Citation: Kasim, M. F., Watson-Parris, D., Deaconu, L., Oliver, S., Hatfield, P., Froula, D. H., Gregori, G., Jarvis, M., Khatiwala, S., Korenaga, J., Topp-Mugglestone, J., Viezzer, E., & Vinko, S. M. (2021). Building high accuracy emulators for scientific simulations with deep neural architecture search. arXiv. arXiv:2001.08055v2
https://arxiv.org/abs/2001.08055 - Simulation-based inference: While sophisticated computer simulations can generate realistic data, their intractable likelihood functions make statistical inference extremely difficult. Recent machine learning advances, particularly neural density estimation and differentiable programming, are revolutionizing the field by enabling direct work with high-dimensional data and dramatically improving inference quality and efficiency. Active learning techniques that guide simulator runs and tighter simulator-inference integration are further expanding possibilities. This transformation allows scientists across disciplines to extract richer statistical insights from complex mechanistic models.
Citation: Cranmer, K., Brehmer, J., & Louppe, G. (2020). The frontier of simulation-based inference. arXiv. arXiv:1911.01429v3
https://arxiv.org/abs/1911.01429
Breakout Session 4: Data Acquisition, Processing, and Analysis
- Data-driven discovery: Modern experiments generate data volumes that exceed the capacity of traditional analysis methods. Machine learning enables scientists to extract patterns, make predictions, and gain insights from these complex datasets. It is especially powerful for modeling natural systems without oversimplifying their behavior. Breakthroughs like AlphaFold and ML-aided drug discovery show its transformative impact. Embracing these tools allows researchers to ask deeper questions and uncover hidden structure in their data.
Citation: Vinuesa, R., Cinnella, P., Rabault, J., Azizpour, H., Bauer, S., Brunton, B. W., Elofsson, A., Jarlebring, E., Kjellström, H., Markidis, S., Marlevi, D., García-Martínez, J., & Brunton, S. L. (2025). Decoding complexity: how machine learning is redefining scientific discovery. arXiv. arXiv:2405.04161v2
https://arxiv.org/abs/2405.04161
Breakout Session 5: Teaching, Training, and Skills Development in Scientific AI
- The AI Fluency: Framework & Foundations course, developed by Anthropic with Professors Joseph Feller and Rick Dakan, is a free, self-paced program. It introduces the AI Fluency Framework—centered on four core skills: Delegation, Description, Discernment, and Diligence—to help learners determine what to delegate to AI, how to communicate effectively with it, critically assess its outputs, and ensure responsible use. Through hands-on practice with large language models (LLMs) and chatbots, the course explores AI’s roles in automation, augmentation, and decision-making. It is already in use in university courses and professional training programs to promote essential AI literacy.
https://www.anthropic.com/ai-fluency
