The EU AI Act: A Double-Edged Sword For Europe’s AI Innovation Future

Luboslava Uram is CTO and managing director at Solvd Group, a subsidiary of Allianz Group. Transforming the claim management experience.

As the CTO of a group of companies at the forefront of innovation in claims processing across Europe, including the application of AI, I have a unique perspective on the EU AI Act. While the Act aims to foster safe and ethical AI development, it presents significant challenges for startups and small-scale providers like us who are driving innovation in the insurance sector.

Our companies have been pioneering the use of AI in claims processing, developing solutions that streamline operations, reduce fraud and improve customer experience. We've seen firsthand how AI can transform the insurance industry, making it more efficient and customer-centric. However, as we prepare to address the requirements of the EU AI Act, I'm concerned that the very innovation we're striving for might be stifled by overly burdensome regulations.

Concerns Over The Classification Of Risk

One of the main problems of the EU AI Act is its risk-based approach to AI regulation. Under the Act, AI systems are classified by their risk level, and high-risk systems will be subject to the most strict requirements.

This classification system is problematic for several reasons:

1. Overclassification Of High-Risk Systems: Our survey of AI startups suggests that between 33% to 50% of AI systems that companies like these are developing are potentially high-risk. This is much higher than the initial guess of the EU Commission of 5% to 15%. The share of high-risk classifications is so high that it could make the European AI market pretty much impossible to innovate in.

2. Definitions Or Lack Thereof: Definitions of various risk categories are often very vague (potentially misleading). The obscurity of this could result in overly cautious interpretations through which companies have to comply with unnecessarily high-risk requirements.

3. Frequent Mistakes: The existing definition of AI in the Act is so broad that it would seem like almost all software can fall under it. This unnecessarily broad scope could impose excessive burdens across diverse technology sectors.

Compliance Challenges

Young AI companies face real challenges in coming into compliance with the EU AI Act:

1. High Resource Requirements: Compliance requirements for high-risk AI systems are significant. These measures include risk management system implementation, data quality assurance, documentation and human oversight. These demands can be especially taxing for startups.

2. Higher Compliance Costs: The monetary burden of compliance is a significant issue. The initial implementation costs and compliance costs can put a burden on the budgets of relatively younger companies, making certain AI projects economically unfeasible.

3. Time to Market Delays: The new administrative and technical requirements are probably going to take longer to get AI products to market. In the fast changing AI field, these delays could cost startups dearly.

Innovation And Competitiveness Worries

A key question is the effect of the Act on innovation and competitiveness. Here are a few areas I have an eye on:

1. Innovation Derailment: According to the research linked above, 50% of AI startups surveyed think the AI Act will slow down AI innovation in Europe. Such a perception, by itself, would be enough to discourage investment and entrepreneurship in the European AI sector through the self-fulfilling prophecy effect.

2. Appeal To Move: Even more concerning, 16% of startups may stop developing AI or move out of the EU. Such a talent brain drain could hit Europe's AI ecosystem badly.

3. Competitive Disadvantage: In the short term, companies located in countries without strict AI regulations may have a comparative cost and timing advantage, being able to create AI solutions faster and at a lower cost. This could put European AI firms at a major global competitive disadvantage.

Delivery Challenges

The Act's requirements are, from a technical perspective, a series of challenges:

1. Technical Challenges In Meeting Explainability Requirements: In case of complex AI systems, especially in deep learning or neural networks, providing clear explanations for decision-making processes is not only a challenge in itself but requires high computational and human resources.

2. Monitoring And Revision: The Act also requires the continuous reviewing and revising of AI systems. For small teams, being able to sustain this degree of continuous monitoring in parallel with product development is often very taxing.

3. Data Governance: Ensuring quality, unbiased data sets for the AI to be trained on is a challenge, especially for the startups with limited access to data.

Solutions And Recommendations

To respond to those issues, I would recommend the following:

1. Refined Risk Classification: The EU could refine its risk classification to more appropriately align with real-world risk of individual AI applications. This could be in the form of more specific categories or guidelines by industry.

2. Proportionate Compliance: Implement compliance obligations that correspond to company size and resources. This may include relaxed processes or longer timelines for startups and SMEs.

3. Innovation Sandboxes: Refine and broaden the use of these sandboxes by giving startups more freedom to test and develop AI systems in a controlled environment and with less regulatory burden.

4. Support With Implementation: To make it easy for startups and SMEs to apply the different obligations of the Act, the government needs to provide clear, practical guidance to support them in this first stage of implementation.

5. Phased Implementation: If the Act is passed, it should be phased in over some time so smaller companies will have time to adapt and comply with the most resource-intensive requirements.

6. Alignment With Global Standards: The EU should work to align EU AI regulation with global standards to avoid placing EU companies in the position of facing multiple, potentially conflicting regulatory systems. However, the EU AI Act can benefit from getting the approach out of the one-size-fits-all-suit and to suit the size of young AI companies, which are often the most innovative and disruptive.

Striking a balance between responsible AI development and innovation is essential. The EU can foster responsible AI innovation with more sophisticated, scalable regulations that avoid strangling nascent AI companies by addressing these concerns.

Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?