Making Europe the best place for start-ups in trustworthy AI
We reached two important milestones in our endeavor to turn Europe into the global hub for trustworthy AI:
- Today, February 2nd, 2024, EU Member States unanimously endorsed the political agreement that we reached in December on the AI Act. The agreement resulted in a balanced and futureproof text, promoting trust and innovation in trustworthy AI.
- Last week, we adopted a wide range of measures to support Europe’s AI start-ups, complementing the regulatory framework.
Both milestones are equally important for European innovators in AI. They reflect our comprehensive approach to AI: promoting both trust and excellence in AI.
Our vision: a thriving European ecosystem of AI start-ups with talented researchers and engineers, developing large language models in all European languages, based on large amounts of easily accessible high-quality data, training them on the world’s fastest supercomputers, and working with industrial partners to turn them into innovative applications, with access to a large Single Market of 450 million people.
And this vision is already becoming reality. For example, in Italy, we see an AI start-up training its large language model in Italian language on the Leonardo supercomputer (part of the EuroHPC Joint Undertaking). In Finland, another start-up works with leading academics on open and trustworthy large language models, training them on another EuroHPC supercomputer -Lumi- with data in all European languages.
With last week’s package of AI innovation measures, we are supporting European AI start-ups in three important ways:
1. First, we give start-ups access to the key “raw materials” of AI: data, computing power, algorithms and talent.
2. Second, we bring together AI start-ups with industrial users, fostering trustworthy AI applications across Europe’s industrial ecosystems.
3. And third, we provide a clear, stable and predictable regulatory environment with harmonized rules in all 27 Member States.
In terms of the “raw materials” of AI, computing power (needed to train advanced AI models) is particularly costly, often making start-ups dependent on partnerships with big tech companies. In Europe, start-ups will now have an alternative. We are giving start-ups (developing trustworthy AI) access to Europe’s high-performance computing infrastructure (EuroHPC). We already have some of the world’s fastest supercomputers (so-called pre-exascale and exascale), but on top of this, together with Member States, we will fund the acquisition and upgrade of AI-dedicated supercomputers.
Revolving around these AI-dedicated supercomputers, we are bringing together the different “raw materials” of AI in new “AI Factories”. These will serve as a one-stop shop for start-ups, offering computing power combined with access to data centers, access to talent through academic partnerships and additional services. Overall, the Commission and Member States will invest an estimated 2.1 billion euros in the AI Factories and AI-dedicated supercomputers until 2027.
In addition, thanks to our common European data spaces and our new data legislation (Data Governance Act and Data Act, which entered into force earlier this month), data will be interoperable and easily available to start-ups to develop new AI models. And through the new ALT-EDIC initiative, the Commission boosts the availability of high-quality language data, which will be key for the development of large language models in different European languages.
Secondly, Europe’s industrial ecosystems, from manufacturing to automotive, will be key drivers of innovation in generative AI. With the new GenAI4EU initiative (with an estimated 500 million euros of additional funding until 2027), we help European start-ups partner with industrial users to transform their AI models into ground-breaking applications. In several promising application areas, such as healthcare and smart cities, we recently launched AI Testing and Experimentation Facilities, helping start-ups validate their applications in real-world conditions.
Lastly, Europe’s AI start-ups need legal certainty. And they need businesses and consumers to trust AI applications. This is where the EU AI Act comes in. It sets effective and proportionate guardrails and ensures that start-ups only need to observe one single set of rules across all 27 EU Member States. And it makes trustworthy AI synonymous to “AI made in Europe”.
The AI Act is based on in-depth stakeholder consultations by the European Commission and meticulous revisions by the co-legislators, culminating in a 38-hour trilogue negotiation and a historic agreement between the co-legislators on 8 December 2023.
This has resulted in a balanced and futureproof text. Rules are purely focused on uses of AI where our fundamental rights, health and safety are at high risk. Research and development are exempted. Transparency measures apply to “general-purpose” models (the large models that are capable of performing a wide range of tasks), while only the most advanced AI models with systemic risks around Europe (and the world) are subject to additional requirements. The AI Pact will help developers anticipate and prepare for the new requirements.
Today’s unanimous endorsement by all Member States is a testament to this balanced approach, as well as its importance for European citizens.
For all these three priorities, the one does not go without the other. Without legal certainty (through the adoption of the AI Act), computing power, data, algorithms and talent will not be enough for a start-up to bring an AI innovation to market. And vice versa. Therefore, we are going full-steam ahead on all fronts. Start-ups – it’s time to come on board!
Article by Thierry Breton, European commissioner for Internal market
