Commission's second Alliance Assembly: excellence and trust on the menu

On 9 October, the European Commission held its second AI Alliance Assembly. The event was fully dedicated to the ecosystem of excellence and trust that the Commission wants to develop -as presented in its White Paper.

Please find below a summary of the key moments.  

Introduction

Commissioner Thierry Breton introduced the webinar by stating that our society produces a huge amount of industrial and public data. Thierry Breton wants European businesses to use this data for the benefit of European consumers. Europe has a strong position in research but needs to increase efforts to remain at the cutting-edge of industrial development.

It is necessary to ensure that European companies have access to European data, it would give them a boost and ensure that data is used in line with European values. Thierry Breton mentioned the example of microelectronic components, which help unfold the development of AI. In Europe, it is possible to ensure that components are energy efficient and trustworthy. The Digital Innovation Hubs will accelerate the adoption of AI.

Regarding the need to regulate AI, the Commissioner referred to the recent public consultation on AI where most respondents felt that more rules were needed. The majority supports restrictions for biometric identification (ban, moratorium, other kinds of restrictions).

Ulrich Nußbaum, German (federal) State Secretary, explained that even though the German presidency is focused on COVID-19, on the long run AI (made in Europe) is important to safeguard jobs. Innovation and trust are two sides of the same coin.

Panel on an ecosystem of excellence

Rita Cucchiara, Professor at Università di Modena e Reggio Emilia (Italy), lamented that European students are very much appreciated but unfortunately often leave Europe for better opportunities. She stressed the need to develop common scientific projects, not only application projects.

Fredrik Heintz, Professor at Linköping University (Sweden), underlined that everyone needs to understand the basics of AI because it is everywhere now. Everyone needs to learn the basics of computer science, it should be incoporated at primary education level. We need to educate more people in the field of AI, especially dual degrees (e.g. law and AI). The biggest challenge is that most of the learning will take place after completing the degree. This requires lifelong learning. Engineers need to update their knowledge.

Yasir Khokhar, CEO of Connecterra (Netherlands), explained that it remains difficult for start-ups to work cross-border. Start-ups need more administrative support in Europe.  Overall, the CEO believed that there is an excessive aversion to risk in Europe, there is room for a bigger appetite for risks and to be okay to fail.

Christel Fiorina, Senior Project Leader for the French Ministry of Economy, explained the state of play in France. A national strategy was published in March 2018, 15 billion euros were invested in AI from 2018 to 2022.

Pekka Ala-Pietilä, Chair of the High-Level Expert Group on Artificial Intelligence, explained that Europe needs to re-skill its workforce. AI literacy needs to be supported. In addition, public procurement could be used strategically.

Panel on the uptake of AI in the public sector

Giancula Misuraca, Associate Research Fellow at Danube University Krems (Austria), presented his study on the uptake of AI in the public sector. Overall, public services are not in such a transformative change. Adoption is low: 33% of AI in general public services, 18% in health, 17% in economic affairs. Most AI cases (57%) are found in national administrations, 30% at local level.

Gatis Ozols, Head of Public Services Department at the Ministry of Regional Development (Latvia), explained that AI is used for translation of official webpage in Latvia. AI enables multilingual communications with the Latvian government. They also have a chatbot to reduce the workload of civil servants. They plan to improve access to institutions by having automatic transcriptions of parliamentary sessions. In addition, a pre-commercial procurement procedure has been successfully implemented to use AI in the judicial system. Several teams competed in creating a model that predicts the length of Court procedures.

Lucia Velasco, Chief of Staff of the Secretary of State for Digitalisation (Spain), added that Spain launched a programme with several hospitals to predict the evolution of COVID-19 patients. Next month, Spain will issue its national strategy on AI.

Marieke Van Putten, Team manager of International Affairs Digital Government (the Netherlands), stressed out the importance of pre-commercial procurements to stimulate SMEs to come up with solutions to societal challenges. In a pre-commercial procurement, the contracting authority is not seeking finalised products but innovative ideas that could become products.

Panel on an ecosystem of trust

Roberto Viola, Director General of DG CNECT (European Commission), emphasised that people need to own their future. In a human-centric system, they must enjoy equal rights. Rights cannot be jeopardised by the choices of a machine.

Katharina Zweig, Professor at University of Kaiserslautern (Germany), explained what can go wrong with AI.

AI decisions can:

-Be based on the wrong data, erroneous data

-They can rely on correlations rather than causal relationships

-They can be too bad to be used in a specific circumstance

-They can be trained wrongly for a given application

-They can wrongly discriminate

There are possible countermeasures:

-Information about input data

-Information about quality of the resulting decisions

-Access to data to be checked

-Access to ADM system to validate quality of resulting decisions

-Access to code (as a last resort)

If done well, none of this will reveal trade secrets.

It is crucial to control ADM systems if they contain a learning component and if that component can inflict some damage. The degree of control depends on the potential societal damage in a given social context. 60% of ADM systems do not need additional regulation. For many of them, they already have a sectorial regulation.

She gave two recommendations:

1)     We need to build a market for trustworthy ADM systems in which the extra-effort is paid for.

2)      We need to enable lawsuits for people affected.

Ursula Pachl, Deputy Director-General at BEUC, does not support the binary approach (high risk vs non-high risks). She wants a multi-differentiated approach. In her views, biometrics are problematic, they should be avoided for commercial purposes following a precautionary approach.

Panel on conformity assessment, standards and high-risk AI applications

Killian Gross, Head of Unit at DG CNECT (European Commission), moderated this panel.

Kathrin Watson, Head of EU Affairs at Robert Bosch, explained that Bosch has a positive view on conformity assessment and a long experience in obtaining the CE marking. The higher the risk the more costly it is to be assessed but it also brings trust in products. For a simple AI usage, self-assessment can be enough. There are already some standards covering AI products, but they could be extended. For high-risk, it should be third-party assessments. Kathrin Watson strongly recommended to the European Commission to adopt a horizontal approach for AI.

Patrick Bezombes, AFNOR (France), underlined that standards are a matter of digital sovereignty; an AI system can harm our democracies. For instance, you might need an AI system supervising an AI trading system; and you need standards for all of this. The risk is to have a proliferation of standards, sometimes without coherence. The issue of vertical vs horizontal has not been sorted out. There is clearly a need for better coordination.

Patrick Bezombes added that the trustworthy assessment list of the High-Level Expert Group could easily be transferred into a standard. However, for most of the other issues, developing standards will take a lot of time. Europe also needs to work on a new concept of conformity assessment, for instance assessment including simulation. Europe might need a continuous assessment approach, which is ground-breaking.

On the question of the length needed to develop a standard, the expert underlines that it can take up to 8 years to develop a standard on a market which is not mature (e.g. eMobility). Length depends on the maturity of the subject. Standardisation organisations need resources.

Clara Neppel, Senior Director Europe at IEEE (Austria), stressed that three pillars should support the ecosystem of trust:

-standards: socio-technical (P7000Tm series, IEEE 7010-2020, governance) and technical standards, including sectorial.

-certification: IEE sets up a system for ethical certification.

-governance including legislation.

Clara Neppel called on the European Commission to focus on public services first because people cannot choose another service.

Hendrik Schäbe, TÜV Rheinland InterTraffic (Germany), highlighted that notified bodies need to take into account these new developments and they are already doing it. Certified bodies will follow the pace.

Andreas Steinhorst, Executive Secretary at the European co-operation for Accreditation (France), explained that both ex-ante and ex-post assessments are needed, depending on the risks posed by the product. He recommended to the Commission to use existing systems for AI applications, i.e. the New Legislative Framework, conformity assessment bodies etc. They only need to be adjusted.