Causal-AIM AI-Powered Causal Analysis of Scientific Experiments
- Coordinator
- Institute of Science and Technology Austria, Austria
- Runs
- 1 September 2026 to 31 August 2031
- Status
- Running
- EU funding
- €1.5m
- Total cost
- €1.5m
- Partners
- 1 organisations in 1 countries
- Type
- European Research Council grant
- Grant agreement
- 101302384
What the project is doing
In the project's own words, from its summary on CORDIS.
The number of storks and newborn babies in a country are correlated, but neither causes the other. Uncovering which correlations signal an underlying causal relationship is a fundamental step in science, requiring the analysis of ever increasing amounts of raw, high-dimensional, temporal data. This creates an urgent need for computational aids by machine learning systems. Unfortunately, off-the-shelf models fail to draw causal conclusions, as they make predictions only using correlations.
Even in well-designed experiments, the correlations they detect can be spurious, leading to mistakes and hallucinations that systematically affect conclusions when fed into causal reasoning pipelines. My research project, Causal-AIM, tackles the challenge of turning pre-trained foundation models into scalable data-driven measurement devices for scientific experiments. My core hypothesis is that foundation models can be post-trained with suitable causal principles to become causally valid by design: capitalizing on the thriving ecosystem of open-source models, we build AI systems that yield causal rather than merely correlational interpretations.
To detect and quantify causality, thereby supporting scientists in their analysis workflow of experimental data, I propose methodologies for: (1) static, high-dimensional, raw data; (2) structured temporal data; and (3) generalization to new experiments without further training. These capabilities are integrated to support correct causal inference on temporal, multi-modal datasets arising from new scientific experiments.
Alongside, I will release the first real-world benchmark for causal inference from high-dimensional data across scientific disciplines. Overall, Causal-AIM addresses key challenges in using AI models to support causal inferences, and its results will serve as a basis for groundbreaking interdisciplinary applications, including research on neurodevelopmental diseases, ecology, virtual cells, and extreme climate events.
Topics
Who is involved
| Organisation | City | Country | Role |
|---|---|---|---|
| Institute of Science and Technology Austria | Klosterneuburg | Austria | Coordinator |
Source: CORDIS, European Commission. Data of 5 Nov 2026, reused under the Commission's reuse policy. View on CORDIS · Grant DOI