The Center for Data Innovation hosted an insightful panel discussion about the European Commission’s new draft proposal for an Artificial Intelligence Act (AIA). The panel brought together voices from the EU’s executive and legislative arms, industry, and small businesses. While the Commissions’ proposal was well-received on the whole, panelists raised a number of issues that co-legislators now have the chance to address as the law works its way through the European Parliament.
From the outset, Kilian Gross from DG CNECT, who is one of the lead authors of the law, made clear that the Commission recognises the huge potential AI offers and wants to encourage its development in Europe. In the eyes of the Commission, Mr. Gross stressed, “about 80 to 85 percent of AI does not need significant regulation.” However, he argued that certain features of AI systems—opacity, difficulty to predict outcomes, challenges in explaining decisions, and data intensity—require a law that addresses and mitigates potential violations to the fundamental rights and the safety of European citizens. The purpose of the AIA is to safeguard citizens’ remedies against rights violations and ensure that AI offers the same level of protection of rights that the analogue world does. The Commission produced a set of horizontal rules applying to all AI systems in order to avoid a plethora of national or industry-specific rules from emerging, which would lead to the kind of market fragmentation issues that have plagued the Digital Single Market for decades.
Cornelia Kutterer from Microsoft welcomed the AIA’s focus on high-risk AI uses, and the clear breakdown of responsibilities between developers and deployers of AI. She had concerns around some of the law’s technical requirements, specifically, the obligation to use “error-free” and “complete” training and validation data. Ms. Kutterer questioned whether any dataset can ever meet this bar, whether error-free datasets guarantee bias-free outcomes, and if the law’s provisions account for the dynamic nature of the machine learning development lifecycle. Moreover, she raised the point that some of the Commission’s policy objectives are inconsistent. For instance, accuracy and fairness often need to be traded off against each other in AI. The technical provisions of the AIA fail to take some of this nuance into account and demands AI attributes that may not be reconcilable.
Axel Voss (MEP, EPP) noted that the priority of the EU must be to create a harmonized approach to AI that does not over-regulate, in order to spur the widespread adoption of the technology in Europe. The aim should be to reap the benefits that AI offers. After Mr. Gross explained that the law’s aim is to develop a general approach to trustworthy and ethical AI, (rather than a single market issue per se), Mr. Voss argued that this creates a strong case for the Committee on Legal Affairs (JURI committee) to take on primary parliamentary responsibility for the AIA. The other potential committee, the Internal Market and Consumer Protection (IMCO), is already busy with actual single market legislation like DSA and DMA. Moreover, JURI would be able to expedite the process and, in a best-case scenario, could see the AIA adopted in the next 12 months. This is in Europe’s interest, as it would offer the entire AI industry a degree of predictability and stability.
Svenja Hahn (MEP, Renew Europe) argued that the AIA should loosen some of its provisions in order to support SMEs: “European companies are not exactly at the forefront of the digital industry, and we must ensure the AIA does not weaken us further.” She pointed out that European startups find it difficult to scale their businesses, and that SMEs already struggle to compete with large companies in a business environment that is overburdened with red tape. Whilst the AIA is a decent starting point from the Commission, she argued that the law must now be amended so it ends up enabling and spurring on AI innovation. Ms. Hahn stressed that the law should offer SMEs in Europe new opportunities—not just more bureaucracy—in a field that is in its infancy. Ms. Hahn was clear that the sandbox provision of the law can and should be enhanced to ensure that the European startup industry and small businesses can partake in the coming decade of AI commercialization rather than being crushed by the AIA.
At this point, Annika Linck of the European Digital SME Alliance, raised a number of well-received points. SMEs, she explained, already suffer disproportionately from the administrative burden and cost that excessive compliance frameworks in Europe impose on them. The GDPR, whilst much-loved by politicians and civil society groups, was a costly nightmare for small businesses and ended up increasing market concentration in Europe. The AIA must not lead to the same outcome. SMEs need targeted relief from some of the law’s provisions. Moreover, Ms. Linck explained that small businesses do not have the in-house capacity to absorb the reams of new regulation that Brussels regularly produces. As such, the legislation would benefit from being streamlined and reduced in scope. Ms. Linck cited evidence that showed the main barriers to AI adoption faced by SMEs are a lack of capital, lack of skills, and data paucity. The AIA will likely exacerbate this, and thus SMEs need targeted support from the Commission if they are to develop and deploy AI in the future.
Mr. Gross recognized the need to balance the law’s regulatory aims with the need for Europe to innovate in the field of AI. The AIA, he argued, provides streamlined requirements for software developers and clarifies the liability faced by those who deploy AI. This is beneficial for businesses. In addition, the fact that SMEs will be able to declare their AI tools as certifiably trustworthy by the EU will offer them a competitive advantage. Mr. Gross conceded that the Commission does not want to send out the message that AI is too risky and should not be pursued.
Mr. Voss pointed out that in Europe, digital regulation often follows the flawed pattern of, “Here’s the list of what we want from algorithms, now let’s pass a law demanding just that.” The result is a complex web of multi-layered rules that gets ever more complicated for businesses operating in the digital economy. He noted, “We demand that algorithms are human-centric, safe, secure, transparent, explainable, can’t be misused, don’t discriminate and comply with GDPR.” Mr. Voss expressed concern that this long list of demands is not reconciled with the practical realities of how software works, how businesses operate, and the cost that regulations impose on businesses. The AIA is a chance for Europe to enhance its competitiveness, but in combination with the suite of digital legislation coming out of the EU can end up hamstringing Europe’s economy.