AI for All: Two Suggestions

Mohammed Sijelmassi, CTO of Sopra Steria, a European IT company, stresses the importance of making AI accessible and transparent to and for all. He underlines the trustworthiness of AI not only as a value but also as a competitive advantage. He proposes two measures: first, in line with his previous article on the Digital Decade, to introduce AI-specific curricula into educational and adult learning systems to a much higher degree. Second, the EU and Member States should pool their resources to build AI superclusters, enabling European researchers and companies to train their models. This would be a measurable contribution to digital sovereignty. 

AI is a powerful and versatile technology. Everyone already uses or is exposed to AI-powered services and products in one way or the other. Whether it’s real-time translation, the recommendation to viewers, automatic driving, electricity grid control, traffic management, the interpretation of radiological images, or more. Therefore, making AI trustworthy is essential: it is crucial to protect our society and mitigate risks – AI made in Europe could become a competitive advantage.

Trustworthiness means that AI should be technically protected against cyber-attacks, respect ethical rules (fairness, privacy, non-discrimination, etc.), and be explicable and transparent to the best extent possible.

Trustworthiness and competitiveness are not trade-offs. The French initiative ‘confiance.ai,’ in which Sopra Steria takes part, together with leading companies such as Airbus, Atos, Renault, Thales, and others along with academics and start-ups, attempts to show precisely this. In this alliance, we are developing the cooperation and development platform needed.

We aim to validate trust in AI solutions for industrial partners that comply with forthcoming EU Regulations, for instance, regarding plant demand prediction, aerial photo interpretation, visual industrial control, or airborne collision avoidance for unmanned aircraft. To do that, we will develop tools and methods for standards.

Our approach is to add domain-specific knowledge to better detect anomalies, understand decisions, and establish quality management. We want to provide platforms, methods, and tools where SMEs can evaluate the trustworthiness of their models. This could lead to the establishment of a ‘sandbox’ to evaluate compliance as envisaged by the EU Regulation on AI. This Regulation should balance obligations and not discourage creativity and innovation, as bureaucratic implementation can lead to complexity and unnecessary compliance costs.

In this context, I propose two policy initiatives. Certainly, there is more to be done, but I would like to highlight two, which, I believe, must be pursued quickly. The first is to scale up awareness, education, and training programmes for specialists and the public. The second one is to create an infrastructure, i.e., specialised computing facilities, which allows European researchers and companies to train their models.

Scaling up AI knowledge and competences

AI is difficult to comprehend. People must understand that we are not talking about intelligent robots – but about clever algorithms optimised to solve specific problems. The availability of data is crucial for AI, but it can also raise privacy issues. People should be able to weigh risks and benefits based on evidence and understanding. 

We need to develop curricula for education and adult learning. Such a curriculum does not need to be country-specific, as we are talking about ‘technical’ and ‘factual’ information. The ‘Elements of AI’ course at the University of Helsinki is a good example. The curricula should explain and not judge, should provide information and not opinions.

We need to foster AI teaching and university courses. Europe has top universities such as Paris Saclay, ETH Zürich, Linz, Helsinki, and Tübingen, but it could have more. We need to widen university-industry cooperation beyond simple projects to establish common institutions. 

Europe has talent, excellent research, and top education. The point is scaling up: we need more researchers, AI practitioners, and an open mindset. The ‘European Institute of Technology’ (EIT) could provide the coordination centre for such an initiative.

AI Super Clusters

AI works because of powerful computers and big data. Several AI techniques were already developed in the 90s, for instance, LSTM by two German researchers (Hochreiter and Schmidthuber) and the French researcher Yann LeCun (now chief scientist at Meta) who was instrumental to design the architecture of ‘Convolutional Neural Networks, CNN’. Today’s so-called transformer models are massively scalable and require huge parallel computing power. European researchers and companies need access to such computer power, also known as superclusters. 

The EU has launched a high-performance computing initiative and one on quantum computing, but neither of them has the capacity to train big AI models (with more than one hundred billion parameters). The new language model ‘Bloom’ (which was trained using a supercomputer at the ‘French National Centre for Scientific Research’ with the participation of more than 1,000 volunteer researchers) has however shown that it is possible: it is open and supports more languages than the well-known US versions. This is the way to go.

However, having such facilities is one thing, but using them effectively is another. We need to support SMEs to use such models for their purposes. They do not need to train very big models, but they should have access to sufficient computer power and pre-trained models. This is not to speak against Google, Meta, or Microsoft, but Europe needs to build up resilience, i.e., infrastructure and expertise with European companies. 

Implementation

We need the EU and Member States to agree on big initiatives and realise them by putting resources together. A way to implement these initiatives could be the ‘multi-country projects’ proposed in the Decision ‘Path to the Digital Decade’. The Decision sets out a framework that public and private actors can use to make such initiatives a reality. The more Member States collaborate in such joint initiatives the better. European companies should get more involved in bringing AI closer to businesses and consumers, as a contribution to ‘AI for All’. This would be ‘digital sovereignty’ in practice.