From ChatGPT to image processing and universal translators, artificial intelligence (AI) algorithms are everywhere—shaping our relationship with the world and society more than ever. In the scientific realm, AI has already been widely adopted for years across medical, social, human, and natural sciences. But what about geosciences?
We spoke with eight researchers from the FGSE, who share how AI is being integrated into their work, how it’s driving innovation—and the new challenges it brings. Explore their projects or read below for a summary of the benefits and limitations of using AI in their research.
“Beyond their powerful applications for research, AI visualisation tools are very useful for communicating with the general public.”
Guillaume Jouvet
Guillaume Jouvet supercharges his models to simulate and study the evolution of glaciers—past and future. AI enables him to work across increasingly vast timescales and ever finer spatial resolutions. He also applies AI in other research contexts.
Machine learning allows Tom Beucler to predict the formation of tropical cyclones more reliably than traditional models, and to improve the realism of storms in climate simulations. As a researcher involved in developing AI itself, he also warns of its potential pitfalls.
“Care must be taken to avoid potentially biased use of machine learning when it is not governed by scientific or ethical laws.”
Tom Beucler
“AI offers a great opportunity for collaboration with scientists from different fields.”
Marj Tonini
Marj Tonini designs AI-based models to assess the environmental impacts of natural hazards, such as wildfires and landslides. By combining natural hazard databases with environmental data through machine learning algorithms, she can map risk exposure and pinpoint the most vulnerable areas. She also sees other advantages in using this tool.
Many urban governments aim to promote sustainable mobility. But how should they go about it? Which infrastructures are most effective? How do cyclists behave in traffic? Christian Kaiser helps answer these questions by using interactive visualization tools combined with deep learning algorithms.
“Before this research, I didn’t think machine learning could be very useful in the field of geophysics.”
Niklas Linde
Niklas Linde trains AI to recognize and generate geological structures for subsurface modeling. Initially skeptical, he has since embraced image-generation algorithms and adapted them to his research.
Céline Rozenblat studies cities and their development. She sees potential in AI to reveal connections between various aspects of urban life—urban planning, social relations, mobility, and more. Still, she remains pragmatic about the broader use of AI.
“The impact of AI depends on how it is used. It can be applied like a recipe, which does not generate new knowledge, but it can also be used to go beyond known principles.”
Céline Rozenblat
“The current rise of AI is mainly due to the massive influx of digital data and advances in using the computing power of computers to exploit it.”
Grégoire Mariéthoz
Data is a key element in the use of AI. Yet it’s often insufficient or of suboptimal quality—due to factors like infrequent satellite passes or incomplete radar images. Grégoire Mariéthoz develops statistical methods to fill gaps in sparse or fragmented datasets, optimizing their use.
Whether pioneers, occasional users, or committed practitioners of AI, these researchers leverage its capabilities to enhance their models, expand spatial and temporal scales, refine predictions, and uncover new paradigms.
An outside perspective is provided by Gérald Hess, who brings his vision as an epistemologist and ethicist to bear on the rise of artificial intelligence in our daily lives.
Artificial intelligence: yes, but why?
Environmental modeling has long been a cornerstone of geoscience research—used to anticipate climate shifts, trace groundwater flows, or decode the workings of urban environments. Traditional models, built without AI, are typically based on complex mathematical equations and a multitude of variables like temperature, humidity, terrain, elevation, and exposure. These models require substantial computational power and long processing times, and they perform best when all relevant variables are measurable and well-defined. However, when randomness must be accounted for, when modeling areas where data collection is infeasible (due to inaccessibility or lack of instrumentation), or when expanding spatial or temporal scales, the computational complexity increases significantly—often exceeding the capabilities of conventional computing systems.
Gaining precision and overcoming computational limitations
This is where artificial intelligence enters the picture. ‘With AI, we can harness the computational power of graphics cards, which enables an extraordinary number of parallel operations. It’s like having 10,000 Citroëns instead of six Ferraris,’ explains Guillaume Jouvet. Moreover, machine learning and deep learning improve computational efficiency by retaining intermediate results—building on previous calculations to perform new ones. By leveraging the full informational content of the data, machine learning enhances predictive accuracy, clarifies relationships between variables, and reduces model complexity. Several researchers also point out that ‘AI makes it much easier to integrate and test variables of different types within their models compared to traditional approaches.’
Marj Tonini took advantage of these strengths to develop a fire risk measurement model, which was subsequently adopted as a European standard: “We tested models with and without AI for fire risk predictions in areas for which no data is available. The AI-based models proved to be significantly more reliable than those obtained with “traditional” models.”
Faced with the challenges of environmental modelling, AI offers an innovative solution by enabling models to be explored more thoroughly on extended spatial and temporal scales, or by revealing potentially unexpected cause-and-effect relationships.
Artificial Intelligence: A Powerful Tool Demanding Prudence and Thoughtful Use
Artificial intelligence is undeniably a technological revolution, but its use requires a thoughtful approach. Experts in the field agree on one key point: basic principles must be respected when applying AI. This includes the need for large and relevant datasets to train models, the design of efficient algorithms, and consistency with theoretical knowledge of the processes being studied. ‘AI should not be used indiscriminately. We must maintain a critical perspective on both the data and the models we apply. Clearly identifying the problems and determining which data are relevant to solving them remains essential.’
Grégoire Mariéthoz points out that ‘the decision to use AI must be made at the design stage of the model.’ It is important to assess the time savings offered by AI in terms of data calculation and exploitation, compared to the time required to collect large amounts of data and train algorithms on it. The decision to integrate AI into a model depends as much on this balance of time investment as on the expected results.
In summary, AI is a powerful tool, but its use requires a balance between the innovative potential it offers and the precautions necessary to avoid misuse. Researchers emphasise ‘the importance of ethical reflection, respect for basic principles and continuous discernment when integrating AI into scientific models. This is how we can fully leverage this technology while minimising potential abuses.’
Pushing the boundaries of environmental knowledge
AI could help push the boundaries of environmental knowledge. ‘It opens up new horizons and offers a tool that will allow us to reach new milestones,’ says Guillaume Jouvet enthusiastically. Tom Beucler even envisions a potential shift in the scientific process in certain contexts. ‘Instead of starting with theoretical hypotheses and then testing them against data, AI allows us to begin with the data itself to generate new hypotheses—with the potential to refine or even challenge existing models. This dynamic could ultimately lead us to uncover entirely new physical processes.’
Will such discoveries become a reality? In any case, the constant development of new algorithms and the integration of AI into the training of future scientists will undoubtedly contribute to its growth in geoscience research in the years to come.
Some definitions
AI: Artificial intelligence (AI) refers to the ability of a machine, computer program, or computer system to perform tasks that typically require human intelligence. This includes activities such as problem solving, learning, pattern recognition, natural language understanding, visual perception, and many others. The goal of artificial intelligence is to develop systems capable of making autonomous decisions, learning from experience, and adapting to changing environments.
Machine learning: a subcategory of AI that aims to give computers the ability to ‘learn’ from data using a mathematical approach. This enables them to solve tasks for which they have not been specifically programmed and to make decisions or perform tasks without direct human intervention.
Deep learning: a subcategory of machine learning based on a multi-level neural network structure. Deep learning has seen significant advances thanks to increased computing power and the availability of large amounts of data, making it a key technique for applications such as computer vision, speech recognition and machine translation.
Stochastic model: A stochastic model is a mathematical model that takes into account uncertainty or randomness in its components. Unlike deterministic models, which produce identical results for given initial conditions, stochastic models incorporate elements of chance into their formulations. These models are often used to represent phenomena influenced by random variables or random processes.
In a stochastic model, results are not predetermined, but rather characterised by probability distributions. This allows for a more accurate representation of real-world phenomena that may be subject to unpredictable variations or random fluctuations. Stochastic models are commonly used in various fields such as finance, meteorology, particle physics, and other areas where uncertainty and chance play a significant role.
