“Regulation Should Not Hurt Innovation” – Interview with Françoise Soulié-Fogelman about Europe’s Position in AI

What is the current state of artificial intelligence in Europe?

In the work at the AI High level Expert Group[1], four enablers for successful deployment of AI in Europe were identified: Data and Infrastructure for AI (develop open-source AI software libraries and data management and sharing initiatives), Generating appropriate Skills and Education for AI (redesign education systems, develop and retain talent and upskill and reskill the current workforce), Establishing an appropriate governance and regulatory framework (ensure appropriate policy-making based on a risk-based and multi-stakeholder approach, and establish governance mechanisms for a Single Market for Trustworthy AI in Europe), Raising Funding and Investment (address the investment challenges of the market and enable an open and lucrative climate of investment). The European Commission have since acted on these different enablers, in particular by pushing a very significant regulation and financing agenda.

Where are the strengths of Europe concerning AI technologies?

Europe has many strong points: a very well-educated population, with good scientists and engineers (this reflects in Europe’s first rank in terms of publication level); a good education system with well-established AI programs, a strong segment of B2B companies, which represent 2/3 of AI-created value², and a vibrant community of AI start-ups[2] with 106 new creations in 2021 compared to 119 in China and 299 in the US.

Which barriers do companies in Europe face adopting AI?

Many European companies still have not adopted AI, especially among SMEs, yet the number of those considering AI is growing. The main barrier is always the fragmented European market, which has many consequences: difficulty to access or share datasets, in particular for “European” topics (for example speech recognition data sets for French) difficulty for start-ups to secure sufficient investments and scaleup, difficulty for talent to find enough high-level jobs and for companies to offer high enough salaries compared to those GAFA routinely proposes in European locations; difficulty for SMEs to identify trusted partners to help them embark on their AI transformation; difficulty to find sovereign solutions made-in-Europe for sensitive applications.

What could politics, economy and society do about it?

We need regulation for AI so that the European society trusts AI is fair and beneficial. However, regulation should not hurt innovation and negatively impact the economy. The coordinated Plan on AI and AI Act[3] proposed by the European Commission still have to be enacted: the impact on European economy will be very significant. We need to follow-up carefully how it is implemented to make sure that AI indeed serves for the good of European citizens.

How do we secure a human-centered use of AI, fair and free of bias?

Europe is very active in producing regulation to ensure that AI deployed in Europe is trustworthy. The topic of trustworthy AI is generating lots of interest and activities among European companies, with many questions discussed, among which that of making ethics recommendations operational while preserving innovation certainly is top on companies’ agenda.

Thank you very much for the Interview!

Further information provided by Françoise Soulié-Fogelman:

The last AI Index from Stanford[4] shows an increase in all indicators measuring the global activity in AI worldwide (from 2020 to 2021): AI Journal Publications (+13%), AI Patent Applications (+69%), Total AI Private Investment (+101% to $B 90.83), legislative records on AI (from 1 in 2016 to 18 in 2021 in 25 countries) etc. while cost for training AI system and training time decrease (for example -63.6% for cost and -94.4% for time in image classification since 2018 with best systems achieving better than human performance, for example, 99,02% Top-5 accuracy of an AI system on the ImageNet classification challenge compared to the 94,90% human baseline). However, when comparing these indicators between US, China and Europe, the respective situations appear very different: for example, if Europe maintains a strong publication level with respect to the US (19,05% of total world journal publications for Europe to 13,67% for the US), they’re both distanced by China (31,04%); Europe is completely left behind the US and China for AI patent applications (3,89%, 16,92% and 51,69%) and for total AI Private Investment (6,42%, 52,87% and 17,21%). Also, most pre-trained AI models come from the US (for example the best ImageNet classifier is from Microsoft or language model GPT-3 from Open AI) as well as most GitHub libraries (TensorFlow has the most cumulated number of GitHub stars 160,70 k, compared to scikit-learn originally from France which comes 5^th with 48k).

[1] https://digital-strategy.ec.europa.eu/en/library/policy-and-investment-recommendations-trustworthy-artificial-intelligence

[2] https://www.ai-startups-europe.eu/

[3] https://digital-strategy.ec.europa.eu/en/library/proposal-regulation-laying-down-harmonised-rules-artificial-intelligence, https://digital-strategy.ec.europa.eu/en/library/coordinated-plan-artificial-intelligence-2021-review

[4] https://aiindex.stanford.edu/report/