ELLIS Position Paper
Executive Summary
ELLIS (European Laboratory for Learning and Intelligent Systems, https://ellis.eu) is a European grassroots initiative in AI with a focus on scientific excellence, innovation and societal impact. It aims to create a European AI Laboratory inspired by models such as EMBL (European Molecular Biology Laboratory) and the Vector Institute (Toronto).
Like Vector, ELLIS builds upon machine learning as driving modern AI — inspired by a model of human intelligence which is not ‘programmed’ but evolved and learned from data. Virtually all of the dramatic recent progress and impact of AI in today’s world is fueled by data-driven machine learning. Machine intelligence will make further progress as modern forms of deep and reinforcement learning as well as causal inference gain traction, enabled by a new generation of the most talented young students attracted by modern AI. However, Europe is facing major barriers when it comes to retaining top talent in European institutions: lack of competitive salaries, high teaching obligations in universities, rigid environments that do not support a fluid relationship with industry and the creation of startups, and the lack of critical mass due to a fragmented situation with separate islands of excellence.
Proposed in 2018, ELLIS tackles these barriers by pursuing a three-pillar strategy to foster European excellence in this highly competitive field: research programs and fellows, a competitive pan-European PhD program and a network of ELLIS units and ELLIS institutes. Openness for participation throughout Europe is key, subject to strict excellence criteria and mechanisms that ensure the highest quality of research leading to clear economic and societal impact.
Canada established a CIFAR network of Fellows studying Learning in Machines and Brains before launching the Pan-Canadian AI Strategy. CIFAR Fellows receive CAN$30k unrestricted funding per year and meet at two workshops where almost all Fellows are present, usually co-located with a leading international conference. This network not only significantly advanced the field, but it also gave a voice to machine learning and modern AI, resulting in the pan-Canadian AI strategy and the fast foundation of three major new institutes for modern AI (Mila in Montreal, Vector in Toronto, and Amii in Edmonton), run as non-profit corporations with links to partnering universities. Inspired by this model, the first pillar of ELLIS is already underway through the establishment of 11 Fellow Programs, dedicated to topics ranging from basic research in theory and algorithms to applications in health and climate sciences and human-centric elements of AI. The program proposals were reviewed by recent Turing award winner Yoshua Bengio with the help of CIFAR.
The second pillar of ELLIS entails inspiring, identifying, nurturing and connecting the best PhD students in Europe in modern AI through the ELLIS PhD program.
The third pillar of ELLIS consists in the creation of a network of ELLIS units located at leading existing institutions or created from scratch performing modern AI research across Europe. According to an open call, an ELLIS unit invests at least 1.5 Mio EUR per year, for at least five years, and requires a commitment to excellence as well as a contribution (at least 300k EUR annually) towards network activities, such as hosting exchange faculty and students, leading research programs and organizing workshops. The response to this call was outstanding, and the set of initial ELLIS units includes many of the strongest European hotspots for modern AI, such as Amsterdam, EPFL, ETH, ISTA, Oxford, and Tübingen, together with new, emerging locations, such as the Valencian region. Jointly, they committed more than 200 Mio EUR of their own funding, a significant part of which is reserved for network activities such as exchange visits and joint projects. They assemble more than 35 ERC grantees, and have already spun out a similar number of startup companies. A second set of ELLIS units has been reviewed in 2020 and includes applications from Berlin, Cambridge, Lisbon, UCL and other sites.
Furthermore, ELLIS aims to create an intergovernmental organization operating a set of fully fledged ELLIS institutes. This is loosely inspired by the EMBL model and should attract outstanding scientists, providing them with the means to generate cutting edge scientific, economic, and societal innovation. All European countries with a strong interest in AI (both inside and outside the EU) should expect economic and societal benefits if they invest in this, and thus not necessarily expect the EU to provide the bulk of the funding. However, EU funding can help in starting the ELLIS network as well as in including scientists in countries that do not have the financial means necessary. It can also ensure ELLIS has a voice at the EU level, where the “High Level Expert Group” for AI is currently biased towards classical rule-based AI. ELLIS submitted a proposal for the EU call ICT-48-2020: Towards a vibrant European network of AI excellence centres. The proposal received top scores among all proposals in this call and will be funded with 12 Million Euros.
Achieving the ELLIS vision entails attracting scientists from the US in cases where even an ETH full professorship or a Max Planck directorship is not sufficiently attractive. This needs not only attractive salaries, but (at least initially) also co-affiliations with the best academic institutions. ELLIS faculty could split their time with universities or existing research institutes. Their personal position would need to be guaranteed for life, even if their full research budget is not (note that in the US, it hardly ever is). ELLIS co-affiliation should imply partial teaching relief, similar to arrangements for instance in Toronto (between Vector Institute and U of T).
While ELLIS is making significant progress in implementing the first two pillars, the path to the establishment of an intergovernmental organization is more complex. There are multiple paths towards realizing this :
- Establishment of an intergovernmental working group of a core group of motivated countries.
- As stepping stones, one can build on existing joint ventures such as the Max-Planck-ETH Center for Learning Systems, or the joint PhD program between MPI for Intelligent Systems and Cambridge, operating since 2014, to create a pan-European ELLIS PhD program.
- There is significant economic interest in machine learning technology among high tech regions in Europe, and those may want to realize the ELLIS vision faster than national governments. Baden-Württemberg has reserved substantial funds for an institute, and its Minister President Kretschmann has explicitly endorsed the points made in the ELLIS open letter, including flexible governance following the EMBL model, the need for guest houses to host international visitors, etc. Other regions in Europe have made similar public commitments, e.g., Bavaria and the region of Valencia in Spain.
- There is philanthropic interest to ensure that Europe does not fall behind in this field which is considered crucial for our future.
ELLIS will explore new ways of fostering innovation. It will provide a well-defined interface to industrial partners and support the creation of start-ups. Following the focus on attracting exceptional talent, ELLIS sites will support co-affiliations, encourage leaves-of-absence, and foster startups (done transparently to ensure societal acceptance). This will counteract brain drain, provide economic stimulation, and ensure European technological sovereignty.
