Von der Leyen sets out AI risk management plan

UK Prime Minister Rishi Sunak filled most of the headlines during and after the recent Bletchley Park AI Safety summit as he sought to place the UK at the forefront of the AI revolution and the largely unknown risks and opportunities that come with it. But less dramatic and therefore less reported was the typically thoughtful and measured contribution made by EC President Ursula von der Leyen, who mapped out the EU’s risk-based approach to AI at the highest level, which sounded like it came straight from a risk management handbook. Speaking at the birthplace of computers and home to Alan Turing, the father of modern computing, during the war as he and his team worked tirelessly to crack Germany’s enigma code machine, von der Leyen said we are entering a completely different era on the back of AI, and the risk needs to be properly managed through robust corporate governance. “We are now at the dawn of an era where machines can act intelligently. My wish for the next five years is that we learn from the past and act fast!” she said. “Given the complexity of these intelligent machines, AI safety is correspondingly much more complex. Experience from other technologies can therefore be our guide. Take the history of atomic energy and the nuclear bomb. Scientists discovered quantum physics, which led to nuclear energy – good – but also with societal risks, and also to the atomic bomb,” pointed out the EC President. Von der Leyen said this experience teaches the first important lesson: the independence of the scientific community is essential. “We need a system of objective scientific checks and balances. We need to nurture a community of outstanding, independent scientists. Scientists with access to resources to evaluate the risks of AI, and free to call out those risks,” she said. Second, the EC President said that the EU needs to establish AI safety standards that are accepted worldwide. Aviation is a good source of inspiration, she said. “Air travel has become extremely safe because we learned systematically from mistakes. Any error can lead to a catastrophic result. The aviation community made it a standard practice that any incident or error is made public and followed up. It is not seen as a failure but as responsible and appropriate to flag an error. Any error is investigated and the results are publicly available and recommendations are always followed up. This approach shows the value of shared standards and procedures,” said von der Leyen. Third, the EC President said that we can also learn from more recent experience in developing a culture of cybersecurity. “Organisations are better prepared against cyberattacks when there is effective information sharing. Widely shared security alerts prevent the worst viral spread. It is a matter of time. AI systems also evolve and learn,” she said. “Complex algorithms can never be exhaustively tested. So, above all else, we must make sure that developers act swiftly when problems occur, both before and after their models are put on the market. In short, I hope that in five years’ time we will all have systems in place that implement those lessons. Doing all of this is the key to unlock the huge benefits of AI,” said von der Leyen. Last summer, the EC President said a global approach to understanding the impact of AI is needed. At that time she proposed something like the IPCC for climate. Since then, the debate has been so intense, for example in their G7 Hiroshima process, that they see even more clearly what is needed, she said. Von der Leyen said that she believes a framework to understand and mitigate the risks of very complex AI systems should be built on four pillars. First, she said a thriving and independent scientific community, equipped with the means to evaluate AI systems is needed, with public funding and access to the best supercomputers. Second, she said internationally accepted procedures and standards for testing AI safety need to be developed. Third, it has to be standard procedure that every significant incident caused by errors or misuse of AI is reported and followed up. And fourth, an international system of alerts fed by trusted flaggers is needed. Von der Leyen said that these four pillars should constitute an effective system of governance. But at the heart of it all, there needs to be a culture of responsibility, she added. “For private actors, this means one general principle: the greater the AI model's capability and attendant risks, the greater the responsibility. That means solid, traceable corporate responsibility, embedded within your business model. But this goes beyond purely corporate responsibility,” said the EC President. “Public authorities are ultimately responsible for the safety and security of citizens. So, we must put in place binding principled rules. And public authorities must have powers of intervention, as a complement and backstop to self-regulation. These are the guardrails,” she said. “And just as on a motorway: guardrails are not barriers – they allow the traffic to keep to the road and proceed safely. That is why we proposed an AI act. Its basic principles are to support innovation, harvest the benefits of AI and to focus regulation only on the high risks,” explained von der Leyen. The EU’s AI act is in the final stages of the legislative process and, as part of that, the foundation of a European AI Office is under discussion. “This office could deal with the most advanced AI models, with responsibility for oversight, in the logic of the four-pillar framework I have outlined. The European AI Office should work with the scientific community at large. It could contribute to fostering standards and testing practices for frontier AI systems. It could complement the private sector in investigation and testing. It should be able to act on the basis of alerts and make sure that developers take responsibility,” she said. And finally, a European AI Office would enforce the common rules in all 27 member states for those most advanced models, added von der Leyen. “That is good news for those businesses and for safety in Europe. But the European AI Office should also have a global vocation. It should be open to cooperate with similar entities around the world. Dear colleagues, history is watching. As Mark Twain famously remarked: ‘The secret of getting ahead, is getting started’,” concluded the EC President.

Von der Leyen sets out AI risk management plan

Want to read this article?

Read Next

Swerma hands out graduate awards

Swedish navy issues Russia shadow fleet warning

Ransomware attack affects Swedish alcohol sales

Eeckman wins Swerma risk manager award

Risk managers discuss impact of Nato membership

Swerma hands out graduate awards

Swedish navy issues Russia shadow fleet warning

Ransomware attack affects Swedish alcohol sales

Eeckman wins Swerma risk manager award

Risk managers discuss impact of Nato membership

Want to read this article?

Read Next

Swerma hands out graduate awards

Swedish navy issues Russia shadow fleet warning

Ransomware attack affects Swedish alcohol sales

Eeckman wins Swerma risk manager award

Risk managers discuss impact of Nato membership

Related Articles