Introduction
A large network of top European experts in artificial intelligence (AI) and machine learning (ML) is to promote the development and deployment of cutting-edge AI solutions in the future and make Europe the world's lighthouse of AI. The EU is funding the project with around €7.5 million, with a further €2.5 million coming from the United Kingdom and Switzerland. The "European Lighthouse on Secure and Safe AI" (ELSA) project networks excellent researchers from 26 top research institutions and companies in Europe to pool their expertise in the field of AI and machine learning. The growing network builds upon the European Laboratory for Learning and Intelligent Systems (ELLIS) and is coordinated by the CISPA Helmholtz Center for Information Security in Saarbrücken which is member of the ELLIS Unit Saarbrücken.
Explaining the science
"Artificial intelligence has the potential to improve all our lives enormously - be it through better healthcare or completely new possibilities for mobility. But a blessing can quickly turn into a curse if the technology is not based on a secure foundation. I see enormous potential in bringing together Europe's top researchers through ELSA so that we can face the great challenges of AI and ML together. I am sure we can make a lot of difference together," says Mario Fritz, faculty at CISPA.
The new virtual centre of excellence will focus on the further development of ML methods, in particular so-called deep learning, as these methods form the basis for most modern AI applications. The initial three-year project will focus on developing robust technical approaches that are compatible with legal and ethical principles. At the same time, the network of excellence aims to create the necessary structures to promote the development and use of AI technology and to break down barriers. ELSA builds on and extends ELLIS. Founded in 2018, ELLIS is an internationally recognized pan-European AI network of excellence which wants to ensure that the highest level of modern AI research can be performed in the open societies of Europe.
“ELLIS has set out to build a European brand of AI. For this it is crucial not only to have state-of-the-art performance, but also a high level of transparency, accountability, and trust. With its focus on security and safety, ELSA tackles some of the core issues, and with its outstanding team drawing from the extended ELLIS community, it has the technical power to make progress on some of the hard open problems", says ELLIS President Bernhard Schölkopf.
ELLIS and its partner network ELISE (European Network of AI Excellence Centres) have already demonstrated in the past what Europe is capable of in the field of AI and ML. Following their success and structure, ELSA also rests on three pillars: selected research programs, local research units, and a PhD and postdoc program.
The ELSA research programs will underpin the research agenda of the 26 partnering institutions and will focus on technical robustness and security, privacy protection techniques and infrastructures, and human agency and oversight. One of the primary goals of the network of excellence is to detect threats early and mitigate damage. This is important because areas for AI solutions are such sensitive as the healthcare sector or autonomous driving. Artificial intelligence is also used in robotics, cybersecurity, media and document security. All of these application areas are the focus of the network.
To achieve its goals, the network is addressing three major challenges: The development of robustness guarantees and certificates, data-secure and robust collaborative learning, and the development of human control mechanisms for the ethical and secure use of AI. The local research units are designed to bring together the best AI researchers in Europe not only virtually, but also at their sites. In addition, excellent researchers from other institutions can be connected to ELSA units through fellowships. Above all, the network will be open to contributions from industry partners and collaborations in order to significantly advance AI solutions and remain internationally competitive.
This is where the ELSA Innovation Lab will make a decisive contribution. Building on the already existing and highly successful PhD and postdoc programs of ELLIS and ELISE, ELSA will support young researchers from all over the world. They will benefit from the knowledge and shared mentorship of leading scientists in the network, be able to participate in events to network, and have the opportunity to further their education through workshops. ELSA is funded under Horizon Europe, one of the largest research and innovation funding programs in the world. Its content is geared toward important societal issues such as health, sustainable development, and digitalisation. A central goal of the EU framework program is to promote excellence in European science. For more information please visit here.
Project aims
The Turing is working primarily on WP3 'Human Agency and Oversight' T3.3, T3.4 with University of Birmingham in order to deliver T3.5 which the Institute is leading on. The Turing is responsible for delivering D3.6 Papers and accompanying code from Task 3.5 with new methods and open source tools.
Applications
Research progress
The research work has been focused on third party certification for AI systems, logging information from stakeholders during the ML development process and eliciting information from end users to incorporate into the learning procedure. The research team hosted the first Work Package 3 workshop on Monday 18th March 2024 as an AI UK Fringe Event.
Evaluating AI-powered language models for maths: A new open-source platform
While evaluating foundation models is of key importance, it’s not just about the models in isolation. The Turing is proud to have contributed to a key new project helping to enable and understand interactions of humans with models, aiming to advance models which work well with humans. As well as new tools, this project required broad collaborations across multiple disciplines, where the Turing excels.
Learn more about this timely contribution on the University of Cambridge website: New open-source platform allows users to evaluate performance of AI-powered chatbots
Publications
- How can we safeguard equality and fundamental rights against AI-generated violations?
- Evaluating language models for mathematics through interactions | PNAS
- FeedbackLogs: Recording and Incorporating Stakeholder Feedback into Machine Learning Pipelines
- Human Uncertainty in Concept-Based AI Systems
- On the informativeness of supervision signals
- Human-in-the-Loop Mixup
- Perspectives on incorporating expert feedback into model updates
- Evaluating Language Models for Mathematics through Interactions
- Selective Concept Models: Permitting Stakeholder Customisation at Test-Time
Events
Past events
ELSA workshop on generative AI and creative arts
On Monday 18 March 2024, The Alan Turing Institute hosted the first ELSA work package 3 workshop organized as an AI UK fringe event, in collaboration with Lancaster University and University of Birmingham. The workshop consisted of a private session where the team showcased project advancements to a selected number of attendees and a public-facing session where Lord Tim Clement-Jones (House of Lords), Jeffrey Nachmanoff (screenwriter and director), Professor Lilian Edwards (Newcastle University) and Matt Rogerson (The Guardian) discussed the transformative impact of generative AI on creative arts, its legal implications, challenges and opportunities.
ELSA General Assembly
The second ELSA General Assembly is taking place on 9th – 11th Sept 2024 Windermere, UK. As well as project updates, there will be presentations on ‘Insights on Secure and Safe AI’ by the European Commission and ‘AI regulations on Security and Privacy in the UK’ by the ICO. The full agenda and more information is available here: ELSA General Assembly Meeting 2024 – ELSA (elsa-ai.eu)
The first ELSA General Assembly took place in Sestri Levante, Italy September 2023. During the event, the participants from 26 partner countries gave updates on the status of the project and showcased their research. There were active discussions about the research by partners and potential research directions. Overall, the first ELSA General Assembly was a great success, owing to the significant impact of the research achievements presented at the event, and the opportunities that the event provided for partners to further strengthen the ELSA network and refine the project strategy. More information go to: First year anniversary: ELSA gathers for first General Assembly Meeting