Have politicians missed how important general purpose AI systems will be to the EU’s AI success and competitiveness? Companies in a wide range of sectors are pointing out how important these AI solutions will be for innovation and for improving the products and services they can offer to customers and society. Regulating these systems disproportionately – on the basis of the ChatGPT storm – will be deeply unfortunate for all other uses of General Purpose AI systems and so-called ‘Foundation Models’.
AI systems that can be used for several different purposes, so-called ‘General Purpose AI Systems’ (GPAIS) and ‘Foundational Models’ (FM), are rapidly evolving. Those kinds of systems are general and may be used for many different purposes. Generative AI-systems are expected to generate an increased productivity of at least 15-40 per cent depending on the use-case (The economic potential of generative AI: The next productivity frontier, McKinsey, June 2023).
During the ongoing AI Act trialogue between decision makers at EU level they now discuss in detail how to regulate this kind of AI-systems, also called generative AI. It is essential that policymakers do not rush into hasty conclusions, but instead take the time to assess and understand the implications of regulatory interventions specifically on those systems. We hear from too many companies they will not be able to deal with the proposed requirements for developing and using generative AI-systems. The main argument is it is impossible for a company to assess risks the system will potentially pose later, as it is depending on the use case.
There are many arguments against the regulation of GPAIS and FMs. Neither the European Commission nor the political institutions have carried out an impact assessment on how each proposal would affect the use of AI, the likely costs, or the impact on competitiveness. Any far-reaching requirements are likely to place SMEs and European companies in general at a disadvantage. The largest US companies will have better financial opportunities and greater knowledge of how to follow the extensive and costly proposals. As regards the Council’s proposal for the regulation of GPAIS (Article 4a-c), this must be supplemented by a risk-based approach and only apply to significant high-risk applications according to Article 6, Annex II and III. The GPAIS should only be regulated on the basis of a concrete application with a risk-based approach, comparable to all other kinds of use that falls under the AI Act.
It is of the utmost importance that legislators recall the principles of legislation that strengthen the investment climate, improve innovation opportunities and strengthen competitiveness
Businesses question the European Parliament’s technology-specific proposal to expressly regulate foundational models in the AI Act. Instead, they see significant benefits in dealing with this specific type of generative AI-system through codes of conducts, as per Article 69. Legislation needs to be made as technology neutral as possible, in order to provide predictability and to remain relevant over time.
It is vitally important for the uptake of AI that any provisions on GPAIS, if they are to be included in the AI act at all, are dealt with in a balanced fashion. It is of the utmost importance that legislators recall the principles of legislation that strengthen the investment climate, improve innovation opportunities and strengthen competitiveness. Technology neutrality, proportionality and a risk-based approach are essential principles and prerequisites for future-proof rules and for compliance to be achievable. This way, we can create a level playing field and ensure increased competitiveness.