The European Commission introduced the second AI regulation framework in the EU in April 2021. Following extensive discussions, EU politicians reached an agreement on the details of the policy on Friday. The democracies within the region have undertaken a significant effort to regulate a technology with profound social and economic implications through the AI Act. While the structure of the AI Act has been defined, it still needs to undergo final legal procedures before implementation.
Justin Hendrix interviewed Luca Bertuzzi, the technology editor at EURACTIV, who closely followed the negotiations in Brussels. The conversation below has been lightly edited for clarity.
Luca Bertuzzi
I am Luca Bertuzzi, a technology journalist at EURACTIV, focusing on German affairs.
Justin Hendrix:
You’ve been covering the trilogue discussions in Brussels regarding the AI Act for the past three days. The negotiations were intense, lasting for at least 22 hours from Wednesday to late Friday. What are your thoughts on this?
Luca Bertuzzi
Exhausted from the rigorous discussions. Since the inception of the AI Act, I have closely followed its progress. The journey has been quite remarkable due to the heightened interest in AI. Initially met with skepticism, the policy now garners more positive attention. The recent 36-hour negotiations underscored the commitment of EU policymakers to address this critical technical issue.
Justin Hendrix:
The conclusion seemed celebratory, with policymakers appearing relieved. Was there a sense of camaraderie at the end?
Luca Bertuzzi
While politicians often vie for PR wins, this occasion warranted celebration. The AI Act marks a milestone as the first AI legislation in the Western world. It is poised to set a global standard for data security akin to GDPR. Governments worldwide are engaging with the Commission to understand and potentially adopt similar regulations.
Justin Hendrix:
There were concerns about the AI Act potentially failing due to disagreements, particularly regarding the regulation of foundational models. How were these issues resolved in the final discussions?
Luca Bertuzzi
Indeed, there were valid concerns. The negotiations began in April 2021, preceding the emergence of technologies like ChatGPT. The European Parliament established a robust foundational framework, incorporating two tiers and specific requirements for high-risk models. Nations like France and Germany, home to unique AI companies challenging big tech, expressed reservations.
Justin Hendrix:
The details of the agreement are still pending release. What can we expect in terms of the timeline for implementation?
Luca Bertuzzi
A consolidated version may be available by January, followed by a period of professional review lasting several weeks. The official publication and enforcement could occur in the spring, with full implementation expected by 2026.
Justin Hendrix:
The AI Act introduces penalties and fines for non-compliance. What repercussions might businesses face for violations?
Luca Bertuzzi
Violations, such as using prohibited applications, could result in fines ranging from 35 million euros to 6.5% of the global turnover. The Act aims to ensure compliance through significant penalties.
Justin Hendrix:
The Act also addresses high-risk use cases. How are these categorized, and what measures are in place to manage them?
Luca Bertuzzi
High-risk applications posing threats to fundamental rights and safety require adherence to stringent risk management protocols. The Act defines criteria for evaluating the risk level of AI systems, emphasizing due diligence in vulnerable areas like recruitment and justice administration.
Justin Hendrix:
The AI Act includes provisions for national security and law enforcement. How do these aspects interact within the regulatory framework?
Luca Bertuzzi
National security exemptions are crucial, particularly for law enforcement applications. The Act mandates transparency and oversight for public entities using high-risk systems, ensuring accountability and privacy protections.
Justin Hendrix:
In conclusion, the AI Act represents a significant step towards regulating AI technologies in the EU. The collaboration and negotiations involved in shaping this legislation underscore the complexities and importance of addressing AI’s societal impacts.
Luca Bertuzzi
Indeed, the governance model established by the AI Act will be pivotal in navigating the evolving landscape of AI technologies. Industry influence and lobbying efforts have played a role in shaping the final language of the Act, with policymakers striving to address the core challenges posed by AI advancements.
Justin Hendrix:
Thank you for sharing your insights and expertise on this critical topic, Luca. Your perspective sheds light on the intricate process of developing AI regulations in the EU.
Luca Bertuzzi
It was my pleasure. Thank you for the engaging discussion.
