Views: 101
Practically valuable, scientifically interesting and already good functioning, existing AI models and EU digital initiatives are becoming an integrating component for numerous EU science and research branches. Most of the AIs – elaborated and/or closely connected to the EU member states – are showing positive perspectives, providing certain science’s impetus in increasing growth and EU’s competitiveness. Such AI models are being used in varied science’s fields, e.g. in health and pharma, agriculture and sustainability, as well in social sciences and humanities.
Background
Artificial Intelligence (AI) is fundamentally transforming the scientific process across all scientific stages, from hypothesis generation and experimental design to data analysis, peer review and dissemination of results.
In many science and research fields, in physical structure predictions, materials discovery and computational humanities, etc. AI accelerates discovery, fosters interdisciplinary collaboration and enhances reproducibility, while improving access to advanced analytical and computational capabilities.
Acknowledging that AIs have become a “transformative, interdisciplinary and international force”, recent Joint Research Centre (JRC) study also noted both AIs integration into almost all scientific disciplines and immense opportunities in accelerating innovation and enhancing research. However, the AIs deployment faces some challenges: i.e. “responsible AI uptake” requires a coordinated and evidence-based approach to address significant technical, interdisciplinary and ethical issues.
Although the JRS analysis was conducted only in three distinct -though vital – research sectors (protein structure prediction, material discovery and computational humanities), the analysis has vital theoretical and practical consequences for the adoption of the European Strategy for AI in
Science.
Reference to: https://euagenda.eu/publications/the-role-of-artificial-intelligence-in-scientific-research-a-science-for-policy-european-perspective
Modern EU AI strategy in science and infrastructure
Recent European Commission new strategy on the EU-wide AIs in science (2025) is aimed at “paving the way” for the resourceful use of European science for the EU’s progressive growth. The so-called RAISE process – resources for AI in science employment – is a background for fruitful and accelerated AI models adoption by the European scientists in all disciplines.
The Joint Research Centre (JRC) report “Role of Artificial Intelligence in Scientific Research: a Science for Policy in the European Perspective” – mentioned above – accompanied the Commission’s AI in science strategy, providing a detailed analysis on the use of AI in the scientific process. The JRC report helps national governance and policymakers developing informed policies to unlock AI’s full potential in EU research and informed investment decisions.
Thus, the JRC is going to lead the established AI EvaluationsHub, announced in the strategy, to monitor and evaluate AI models and systems in strategic scientific fields: the EvaluationsHub provides a centralized Supplier Performance Management (SPM) platform to gain visibility, reduce risk while keeping suppliers accountable, without the administrative overload; hence, it assures a full control of a supplier performance, ESG, and risk assessment in one platform.
More in: https://evaluationshub.com/; and in: https://openai.com/safety/evaluations-hub/
To accelerate the uptake of AI in science, the JRC suggests creating “shared infrastructures and open science platforms” to ensure reproducibility, wider access and trustworthiness: the idea is based on facts that the AI models are becoming more powerful and versatile, they also require significant resources for training and deployment. This requires enormous investment in High-Performance Computing (HPC) and in AI Factories (to foster innovation and collaboration) through the essentially needed open scientific data repositories to secure EU’s position as a leader in AI science and research.
Larger use of AI models in science would generate new requirements for specialised expertise among researchers: the JRC assessment on skills shows the need for “hybrid” teams – multidisciplinary and interdisciplinary – able to combine expertise in engineering, computer sciences and AI with specific domain expertise. Corporate policies should therefore focus on attracting, developing and retaining interdisciplinary talents to ensure that human expertise remains central to the research process. In short, the AI models provide scientists with access to strategic resources, data, tools, computation and data-training so that AI becomes a “co-scientist” in numerous research disciplines.
More in: https://joint-research-centre.ec.europa.eu/jrc-news-and-updates/ai-strategic-tool-improve-scientific-research-2025-10-08_en
AI examples in health and natural science
= BLOOM (Big Science Large Open-science Open-access Multilingual Language Model)
It demonstrates a major pan-European collaborative AI research effort that emphasises openness, transparency, multilingualism and responsible/licensed release. It could be useful in scientific research contexts where multilingual access or open model availability is at stake, like e.g. in humanities, social sciences and cross-lingual scientific texts. As soon as it is an open AI, it enables researchers in Europe (and globally) to build, fine-tune or study large models without relying on proprietary black-box systems. BLOOM is a 176 billion-parameter open-access multilingual large language model developed by a volunteer-driven “big science workshop” initiative, being supported by 46 natural and 13 programming languages.
The associated AI’s model and training dataset (ROOTS) is openly released under a Responsible AI License (RAIL) v1.0; which was trained on the Jean Zay supercomputer in France (via GENCI/IDRIS) among other European digital resources.
Some caveats/tips: – although it’s open, large-model training and fine-tuning, the AIs still require substantial computing facilities and expertise; – even open models could have limitations regarding domain-specific accuracy, in bias or reliability: hence, it is always better to evaluate physics, chemistry, highly technical biomedical, etc. it is advisable to have a domain-specific fine-tuning model rather than just a general LLM; – finally, the licensing/usage terms shall be checked, specifically in case of strict regulatory compliance and/or commercialisation.
= Owkin, French AI-biotech company is focused on algorithms for drug discovery, diagnostics and biomedical research using data from numerous EU hospitals/academia. Owkin is a pioneering advanced agentic AI that can process vast biological data, uncover hidden causal relationships and generate new actionable insights: this AI is vital for numerous for such science sectors as: the biomedicine science using data across multiple institutions (hospitals, clinics) that cannot easily share basic/raw data due to privacy/regulation, hence complex learning is the optimal solution; – collaboration between academia, hospitals and industry within the European regulatory/data-protection context; for researchers working in bio-medical or pharma-associated science (with a due respect for the EU-wide data governance).
The caveats: if the user’s domain is biomedical/clinical, it is advisable to check the national regulatory system through, e.g. the EU GDPR, medical device regulations and/or ethical boards, as the use of AI models with clinical data has extra constraints. Besides, there is another pharma-partner: the MELLODDY model (Machine Learning Ledger Orchestration for Drug Discovery), which includes several EU’s and global pharmaceutical companies, applying large data base and learning to train AI across data silos.
More in “Understand complex biology through agentic AI”, in: https://www.owkin.com/
= AlphaFold AI model, although non-EU-based, but is heavily used in the EU science: – it is a deep-learning-based AI system developed by DeepMind (UK/Alphabet) that predicts 3-D protein structures from amino acid sequences; – the AI project has widespread uptake in European research labs and has become foundational in structural biology and related fields.
The AlphaFold algorithm (and its predicted structure database) is a vital de-facto digital tool in several occasions: if the scientific work involves molecular biology, biochemistry, structural biology, drug design, etc.; – if research (and/or EU funding, infrastructure, policy, etc. emphasises reuse of large-scale predictions and open resources; – this AI serves often as a show-case in enabling scientific breakthroughs (solving a long-standing “protein folding” problem), and facilitating case-studies in EU science policy contexts.
More in: https://deepmind.google/science/alphafold/
AI in agriculture and sustainability
= AI4Copernicus, AI4Europe and AgriDataSpace AI: these digital frameworks represent deploying AI models for “precision development” in agriculture, soil health monitoring and biodiversity tracking.
= Users of EU’s open data and transparent AI for sustainability policy can address: – PhenoAI (EU Plant Phenotyping Network), which is the AI-based image recognition models running on field phenotyping robots and drones; it used in plant research stations in France, Netherlands and Spain.
AI in social sciences and humanities
= CLARIN and DARIAH: European digital research infrastructures using multilingual transformer-based models fine-tuned on the European Parliament data and information, on various literature archives and historical publications; these AIs enable digital processing and text mining, multilingual sentiment analysis and cultural data annotation, etc.
= AI for Policy Modelling (Europe Direct, SHERPA Project): this AI systems are used to model societal impact, misinformation flows and citizen sentiment across numerous EU languages.
More in “Developing AI policy” and in “policy modeling: https://www.sciencedirect.com/science/article/abs/pii/S0161893823001199
On AI-assisted scientific methods
= AI4EOSC and RAISE are the models providing the so-called AI-as-a-service platforms for European researchers with integrated data governance, reproducibility tools and HPC backing; the models are part of the EU Open Science Cloud and are using the Horizon Europe funding.
= EuroHPC AI Centers of Excellence is the digital platform using the following AI models- EXCELLERAT (in engineering), BioExcel (in life science AI), and HiDALGO (in global challenges modelling); all models can integrate AI directly into simulation and inference workflows.
AI in the EU law, legal profession and governance
Numerous AI models are increasingly used in the EU legal and justice system: the process of adopting AI models is copped with a strong emphasis on ethics, transparency, multilingual access and compliance with the EU law (especially GDPR and AI Act principles).
However, it is possible to note some available AI developments, tools and initiatives specific to law and the legal profession in the EU:
= In AI tools and models used in European legal practice: ROSS Intelligence AI (EU adaptation and local clones), while originally US-based, European legal-tech vendors have adapted similar NLP-powered legal assistant models tailored to EU case law and directives.
= Multilingual and trained on EUR-Lex, Curia (CJEU database) aimed at national court rulings.
OpenJustice is an open-source platform designed to enable legal professionals to effortlessly create and deploy customized AI models using plain language. The platform’s mission is to democratize access to AI models for the legal profession, enhancing accessibility, transparency and reliability of AI for legal applications. OpenJustice is currently trained on legal data for the US and Canada (with limited coverage). Legal professionals and academics are invited to join and contribute to the OpenJustice community, facilitating the growth of and access to AI resources within the legal field; the AI is open to all lawyers and legal professionals with an institutional email. More in: https://openjustice.ai/
Generally, the EU digital projects use AI to classify and summarize court decisions, especially in countries like France, Germany and the Netherlands, where open case law access is improving.
Thus, AI-powered search engines, like Doctrine.fr, Jurivisions, Predictice, etc. are offering case outcome prediction and legal argument mapping.
= “Justice.ai” (pilot projects in Estonia and Austria): Estonian Ministry of Justice experiments with AI-based pre-trial assistance and automated drafting of procedural documents. Austria is using AI to detect inconsistencies in regulatory texts before publication.
= AI for legal drafting and compliance: the EU-wide LexisNexis AI Modules and Wolters Kluwer’s European AI Suite are the legal publishers that presently offering automated GDPR compliance checkers, AI contract clause risk scoring and directive’s alignment; they are integrated with the EU law-change monitoring system, e.g., when the Commission drafts and/or updates a directive, lawyers receive the “AI-curated” impact analysis.
= ClauseBase (Belgium) and ContractMill (Finland): these are the digital tools for the AI-assisted policy drafting and document automation aligned with EU legislative templates.
= AI in public law and administration: the EU e-Justice Portal – with the AI-enhanced search – provides the ongoing upgrades to integrate semantic AI search so that lawyers/judges can retrieve relevant directives, regulations and cross-border case law via natural language queries.
= AI in public procurement legislation (basically in Spain and Portugal): these AI systems are used to detect fraud patterns and “red flags” in bidding processes; they are trained on EU procurement datasets, such as TED and SIMAP.
= AI in judicial analytics and the predictive justice (mainly, in France). For example, AIs in predictive justice ban and controlled Use, which are the tools like Predictice that can estimate the probability of a court decision, although the French law restricts profiling judges; instead, it reflects a broader EU stance, i.e. the AI may assist lawyers but must not undermine judicial independence. In Italy and the Netherlands, the sentencing pattern AIs as research-partners assistants are using machine learning to detect inconsistencies or bias in sentencing, i.e. to highlight inequalities rather than “automate” sentencing.
Reference to: https://www.ibm.com/think/topics/generative-ai-vs-predictive-ai-whats-the-difference
=Multilingual legal AI models trained on EU law texts like EUR-LexBERT and Legal-BERT-EU, including GitHub access and datasets for the “technical aspects”.
On the EUR-LexBERT portal in: https://eur-lex.europa.eu/homepage.html?locale=en, and https://eur-lex.europa.eu/content/online-learning/personalise-your-experience/eur-lex-preferences.html
LEGAL-BERT is a family of BERT models for the legal domain, intended to assist legal NLP research, computational law and legal technology applications. The Legal-BERT platform was pretrained on a large corpus of legal documents using Google’s original BRET code:
= 116,062 documents of EU legislation, publicly available from EURLEX (http://eur-lex.europa.eu), the repository of EU Law running under the EU Publication Office.
= 61,826 documents of UK legislation, publicly available from the UK legislation portal (http://www.legislation.gov.uk).
= 19,867 cases from European Court of Justice (ECJ), also available from EURLEX.
= 12,554 cases from HUDOC, the repository of the European Court of Human Rights (ECHR) (http://hudoc.echr.coe.int/eng).
= 164,141 cases from various courts across the USA, hosted in the Case Law Access Project portal (https://case.law).
= 76,366 US contracts from EDGAR, the database of US Securities and Exchange Commission (SECOM) (https://www.sec.gov/edgar.shtml).
Note. Additionally, some notable issues, can also be researched, such as:
= Specific AI startups in EU law and regulation technology with user-cases.
= Overview of EU AI Act impact on legal AI (important for law professionals).
=The AI used in the European Patent Office and intellectual property (IP) law.