The European Commission’s proposal for an AI Act has now been fully discussed by the Member States. Several improvements have been proposed, some of which have had a partial impact. Meanwhile, the European Parliament’s deliberations are still ongoing, and these have raised a number of concerns over how innovation and the use of AI will be affected if several of these proposals are implemented.
The European Commission’s proposal for an AI law has now been finalised by the Council and will be formally approved by the TTE-council meeting on the 6th of December. Several good intentions have been discussed to clarify and specify how different AI solutions should be classified according to risk, based on the area of use and the applicable sector regulation; in part, these have had a positive impact. Meanwhile, the European Parliament’s deliberations are still ongoing.
The question for the Council and Parliament’s politicians is “What do you want to achieve with the AI Act? Is it innovation, development, and the increased use of AI?”. The future for innovation in AI and its use is perhaps principally about improving people’s quality of life, making companies and authorities more efficient and improving both the environment and the climate. In order not to overly restrict the potential for good offered by AI, legislators need to invest time and effort in developing clear and delimited regulation that is suitable for differentiated AI use.
The AI Act poses some challenges, not least for developers and business users. First, it is not easy or straightforward to understand due to the heavy technical and extensive content. And, as the drafts of the AI Act currently look, before the legislators’ negotiations in the so-called ‘trilogues’, there are many requirements that are overly unclear and disproportionate.
According to the Commission’s impact assessment, around 10-15% of AI systems will be classified as ‘high risk’. For high-risk AI, legislators have suggested that specific requirements must be met and that certain measures must be taken before putting the AI system into service or placing it on the market.
Currently, AI use and the consequences of the outcome already must comply with current legislation such as GDPR and discrimination laws. A major change for AI producers is the product safety requirements in the AI Act, taken from the New Legislative Framework for Products. AI Act contains requirements for, among other things, conformity assessment, market surveillance and CE marking. This is a new development for intangible products.
In addition, there are whole areas where it is being proposed that AI should be classified as high risk without adopting a risk-based approach. For example, this would cover AI used as safety components in road traffic, water and electricity supply, or AI-based systems intended for use in education, employment, and law enforcement.
In addition to the ability to meet regulatory requirements, there are concerns over costs and their likely impact. Will these prevent start-up companies and SMEs from being able to contribute with solutions and innovations in the field of high-risk AI? In the main, the costs are linked to the associated far-reaching administrative, data storage and documentation burdens.
The EU has fallen behind in the development, research and investment in AI. It needs to step up its efforts, as AI will be crucial to delivering the EU’s digital transformation and will continue to have an increasing impact on the economy and daily life. This has been shown in the report from the European Parliament’s ‘Special Committee on Artificial Intelligence in the Digital Age’, which was adopted in May of this year.
One could argue that regulations must be future proofed and technology neutral if one wants to achieve a welcoming investment and R&D climate. Investments and innovation efforts become more uncertain when Delegated Acts, Implementing Acts and common specifications are used; despite this, legislators are choosing to make use of these in the AI Act.
There is a pressing need to find an appropriate balance in the wording of the AI Act, so that trustworthy and competitive AI enters the market, AI that has been developed by large companies and SMEs alike. The Parliament still has an opportunity to put forward a clear and unambiguous proposal. After that, it will be up to the legislators in the spring trilogue to agree on the best AI regulation for Europe.