The AI Act and IP – will Europe’s AI prospects be improved?
The EU’s reached a provisional agreement on the Artificial Intelligence Act on 9th December following extended last-minute talks between MEPs and Council negotiators. Although the text still has to be formally adopted and may be subject to some modification, the basic structure of the first global regulation of AI is established. Its pages pay testimony to an effort at balancing safeguards without compromising innovation. OxFirst has assessed the trends global trends in AI development from a patent perspective, and found Europe to be lagging behind the USA and China.
The regulation ascribes restrictions on the basis of perceived risk, with unacceptably risk applications being banned outright. These include, for example, biometric categorisation systems using sensitive characteristics, scraping of facial images to create facial recognition databases, social scoring systems and others. In the case of general-purpose AI (GPAI) systems transparency systems are to be introduced which require technical documentation and dissemination of “detailed summaries about the content used for training” the models. ‘High impact’ foundation models will be subject to a fundamental rights impact assessment and some will require registration in an EU database of high-risk AI systems.
Yet despite these restrictions, the European Commission believes this Act will be “innovation friendly”. Allowances are made for ‘sandboxes’ – controlled environments which mimic those of the end user without exposure to them – and for testing in real-world conditions outside of sandboxes subject to specific safeguards.
These measures have not convinced all, including President Macron who expressed concerns that the French national AI sector will be rendered uncompetitive against China and the USA. A former minister of Mr. Macron who now works for Mistral, a French AI company building large language models, has echoed this sentiment, and over 150 companies wrote to the European Commission stating the AI Act would “jeopardise Europe’s competitiveness”.
Whilst these concerns may be legitimate, the problems around AI development in Europe also have a longer history. OxFirst previously assessed data on patent registrations in AI and found that Europe was significantly overshadowed on AI development by China and America. Analysis of the patent landscape between 2010 and 2020 indicated an acceleration in the number of AI patents being filed, with the greatest increase in 2017/18 of 68%, but that Europe did not feature in a significant way.
Amongst the patent owners, Samsung held the greatest AI patent count with just over 5000 AI patents, more than double the next nearest competitors, IBM and Tencent. It is notable that of the top ten companies by patent count, four are headquartered in the USA and China each, two are headquartered in the Republic of Korea and none are European.
Source: OxFirst Internal Analysis, 2020
Similar trends are revealed when AI patents are assessed by nation, with China possessing by far the largest patent count, over 75% greater than the USA and over five times greater than the Republic of Korea. Patents filed at the European Patent Office (EPO) numbered less than 10% that of the USA and less than 5% that of China. Some of this difference can be ascribed to differing legal limitations. In the EU the European Patent Convention imposed restrictions on software patents, whilst the 2014 US Supreme Court decision in Alice Corp v CLS Bank also established restrictions in the same area. Nonetheless, if patents are used as a proxy for R&D, this data would indicate a comparative lack of innovative output on AI in Europe during the studied period.
Source: OxFirst Internal Analysis, 2020
This matters because patents are the fundamental unit of the knowledge-based economy. They provide innovators with the means to bring their inventions to market, thereby paving the way to commercial success and continued innovation. Whilst there is room for change and some success stories – Mistral AI for example, despite its protestations at the AI Act, looks set to close a fundraising round worth just under $500m and bring its valuation to $2 billion – OxFirst’s analysis makes evident that European AI development has not kept pace with that of other regions of the world with inevitable economic consequence.
So whilst Mr. Macron and others may be right to point to a risk that Europe will lack global AI competitiveness, OxFirst’s data would indicate a longer history to be considered. The question that remains to be answered is whether the AI Act will help or hinder European AI development? Not in question, however, are the consequences. The global value of the AI market in 2022 was just over $450 billion and that is projected to grow to over $1,800 billion by 2030; it is in Europe’s interest to ensure it does not continue to lag behind the pace of AI development.