Safety as top priority at AI summit

At the international AI summit in Dublin on October 14, tech executives and European officials focused on ensuring the safety of artificial intelligence, especially as cybersecurity incidents are becoming more frequent. The discussions highlighted the growing concern that without increased security efforts, major incidents could occur within one to two years.

Call for cross-sector safety practices

Google Deepmind's president, Demis Hassabis, suggested drawing on best practices from other industries that manage critical systems, such as biology or nuclear energy. He also mentioned his intention to establish a regulatory authority for the most powerful AI models, possibly modeled after the Financial Industry Regulatory Authority (FINRA).

Challenges of AI misalignment

Hassabis emphasized the issue of misalignment in AI systems, which involves aligning human values and objectives with AI models to ensure they are safe and reliable. He warned that this challenge is becoming more complex with increasingly automated systems, and recent incidents have shown the risks involved.

Need for emergency controls in AI

Microsoft's chief, Brad Smith, supported the idea of implementing "emergency brakes" or mandatory circuit breakers in AI systems. These would allow users to limit damage in case of loss of control, especially for AI managing critical infrastructure or automated systems. He pointed out that AI currently lacks the kind of regulation that everyday objects are subject to.

EU's push for control and transparency

Henna Virkunnen, vice-president of the Commission for Digital Affairs, criticized the global rush to develop AI quickly, arguing that safety is being overlooked. She noted that companies often lack awareness of their AI agents’ actions and stressed the need for control. She also highlighted the EU's AI Act, which requires companies to maintain control over their systems and has requested information from major firms about their safety measures.

Balancing innovation and regulation

Sarah Friar, chief financial officer of OpenAI, supported stronger security measures but cautioned against regulations that could slow down the sector or allow malicious actors to gain an advantage. She emphasized the need for a balance between innovation and regulation, allowing proper evaluation of AI models without hindering technological progress.