The Artificial Intelligence Act

We are still grappling with just how profoundly artificial intelligence (AI) will impact some of our economic, social, and political basic notions. Although much remains to be fully captured about its power, limits, benefits, and risks, the time to make choices has arrived, after the publication of the European Commission´s Artificial Intelligence Act (AIA). The landmark endeavor of setting the world-first attempt at a horizontal regulation of AI anticipates far-reaching consequences and implications for organizations developing and using technologies and products incorporating AI. In other words, for all organizations.

One of the fundamental choices to be made now for the future is related to the legal landscape for startups and innovative SMEs to continue innovating in Europe. A yet nascent form of technology such as AI entails an outstanding level of ambition, calibrated by putting in place the necessary risk-mitigation measures. How AI will contribute to solving some of our society’s most enduring challenges is deeply connected to how we deliver on a regulatory model which brings into focus and actively supports AI innovation in Europe. But will the Artificial Intelligence ActA drive more entrepreneurs to want to build an AI startup in Europe or to want to develop their AI products in Europe and not someplace else?

In this position paper, the Portugal Tech League provides a comprehensive analysis and a set of recommendations from the perspective of startups and innovative SMEs communities: in our understanding, the current drafting of the AIA misses the opportunity to effectively spur innovation in Europe, by disincentivizing AI entrepreneurs from developing AI systems in Europe.

To improve the framework, startups, and innovative SMEs outlined the need for a clearer and simpler pathway to scale their AI-enabled products and services both for low and high-risk AI systems, through a more tailored AI sandboxes model, a clearer definition of high-risk AI systems, with a more achievable high-risk list of requirements and a more proportionate set of obligations, underpinned by easy to implement and globally oriented standards.

Available for download here, startups and innovative SMEs outlined the following key improvements:

• Clarify and simplify the pathway for European startups to scale up their AI-enabled products and services both for low and high-risk AI systems, through a leeway for them to have greater flexibility in achieving the risk-mitigation goals of the AIA;
• Clarify the interplay between the AIA and neighboring legal instruments, such as those in Annex II, GDPR, DGA and DSA, and the future Data Act, thus creating legal certainty;
• Redesign regulatory sandboxes model, by harmonizing implementation across the EU Member States, setting clear parameters for participation and criteria that outline the processes within the sandbox and introduce a maximum cap of liability or a different scheme to provide compensation in case of damages;
• Narrow down the scope of Annex I, and detail expressions such as “wide variety of methods” in order to make the AIA simpler;
• Clarify the definition of high-risk AI systems with a clear risk taxonomy and improve the flexibility of the EC’s annual assessment and update of the list of high-risk uses processes. The Portugal Tech League suggests an approach like the Data Protection Impact Assessment introduced in the GDPR to ensure legal certainty;
• Clarify responsibilities and roles of AI providers, users, and operators, ensuring they reflect the startups’ innovation model, and due diligence obligations when using a third-party dataset, to avoid hampering innovation;
• Make sure high-risk requirements are in line with industry best practices and attuned with the reality of the tech industry, by nuancing terms such as “error-free data” and “complete”. In the case of access to reliable, representative, and unbiased databases, the Portugal Tech League believes that the Digital Europe Program line should take into account the needs of startups and innovative SMEs;
• Clarify what an appropriate level of accuracy, robustness, and cybersecurity of AI systems could be and how this would be assessed;
• Making sure conformity assessment procedures in high-risk and low-risk areas take into account the startup innovation model, to avoid double compliance situations or forum shopping, and long times for approval. The Portugal tech League also believes that setting a publicly funded period review conducted by third parties could benefit smaller players;
• The standardization and implementation processes would benefit from a longer transition period, and these standards should start being developed right away, with a view towards international convergence;
• To ensure dissemination and uptake of AI by startups and SMEs, the Portugal Tech League believes that it is crucial to have a Digital Innovation Hub specialized in AI per EU Member State.