The rapid speed of AI development and the rise of AI agents that can act autonomously is reshaping the administration of justice and making the need for regulation more urgent, IBA president Claudio Visco told delegates at the organisation's annual conference, held this week in Denmark.
Speaking at Monday’s morning keynote address – featuring Pharos Futures Research Group founder Dr Nicklas Lundblad and the IBA’s Artificial Intelligence Institute (AII) founding director Dr Farzana Dudhwala – Visco said this growing governance gap is what spurred the creation of the AII in June. It aims to examine how the legal profession can help “civil society to put together the tools necessary to manage this new and potentially useful but at the same time dangerous tool”.
“[The AII’s] purpose is to bring the expertise of the legal profession into the global conversation about artificial intelligence, to help shape governance frameworks now being written, to press for systems to be developed consistently with the rule of law, with human rights, international order and democratic integrity,” Visco said.
One of the legal challenges with which the profession is currently grappling is what should count as evidence of what an AI agent has done, as well as whether it would survive a court room. Dudhwala said she had spent the past few months in conversation with people at the forefront of AI technology – frontier labs, safety institutes, governments and regulators – about this question and whether concepts of the law created over centuries for people acting on behalf of someone else can be applied to machines doing the same thing.
For Lundblad, this means understanding what agency is and why AI agents don’t yet have pure agency – or free will. Instead AI can only capture human intention and act on humans’ agency, hence the recent Hugging Face incident, in which OpenAI agents escaped their sandbox environment and hacked Hugging Face’s IT infrastructure, was not an example of a ‘rogue agent’ but an unexpected consequence in the way the machine tried to solve the problem it had been set, Lundblad said. In that case, humans still remained the actors, he added. “For that to change, we have to figure out something that’s much, much more complicated, and that’s real agency,” said Lundblad.
However, another challenge that regulators and policymakers will need to think about is the concept of intentionality and how systems built on scarce intentionality will break down if humans can delegate agents to, say, flood a government department with endless volumes of appeals. “A queue works because you have to pay attention to it and you have to spend time in it,” Lundblad said.
“So if you have an infinite amount of intention, there’s no sense in saying you should line up because people just send 200 agents and it won’t work. A lot of our institutions are… built on the premise of scarcity of intention. So, we have to think about how we then deal with an enormous amount of artificial intention being directed at them.”
The Global Legal Post is the official publisher of the IBA Daily News at the IBA Annual Conference in Copenhagen. Follow our coverage throughout the week for the latest news, interviews and insights from the conference. Click here to download a PDF version of Tuesday’s issue.
Email your news and story ideas to: [email protected]
