Stay up to this point with free updates
Simply register for Artificial intelligence myFT Digest – delivered straight to your inbox.
The three foremost Western countries developing artificial intelligence technologies are near signing the primary legally binding international treaty on using AI, but corporations fear that a patchwork of national regulations could hamper innovation.
The US, EU and UK are expected to sign the Council of Europe Convention on Artificial Intelligence on Thursday. In its approach to regulating private and non-private systems, the convention emphasizes human rights and democratic values.
The convention was drafted over two years by greater than 50 countries, including Canada, Israel, Japan and Australia. It requires signatories to be held accountable for any harmful and discriminatory consequences of AI systems. It also requires that the outputs of such systems respect equality and data protection rights and that victims of AI-related violations can take legal motion.
“With innovations moving as fast as AI, it's really essential that we take this primary step globally,” said Peter Kyle, Britain's Minister for Science, Innovation and Technology. “It's the primary (agreement) with real impact globally and it also brings together a really diverse group of countries.”
“The proven fact that we hope that such a various group of countries will sign this treaty shows that we as a world community are indeed as much as the challenges of AI,” he added.
Although the treaty is touted as “legally enforceable,” critics indicate that it doesn’t provide for sanctions equivalent to fines. Compliance is measured primarily through monitoring, which is a comparatively weak type of enforcement.
Hanne Juncher, the council's director in control of negotiations, said 10 participants are expected to be among the many first to approve the agreement when it’s opened for signature on Thursday.
She said: “This is a confirmation that (the agreement) goes beyond Europe and that the signatories have put a whole lot of effort into the negotiations and … are satisfied with the final result.”
A senior Biden administration official told the FT that the US was “committed to making sure that AI technologies support respect for human rights and democratic values” and saw “the critical added value of the Council of Europe on this area”.
The deal comes as governments are developing a bunch of recent regulations, commitments and agreements to oversee rapidly evolving AI software. These include the European AI Act, the G7 agreement agreed last October and the Bletchley Declaration signed by 28 countries, including the US and China, last November.
While the U.S. Congress has yet to pass a comprehensive framework for regulating AI, lawmakers in California, where many AI startups are based, did so last week. That bill, which has drawn mixed opinions within the industry, is awaiting the signature of the state's governor.
The EU regulation, which got here into force last month, is the primary major regional law, but Kyle, a Briton, points out that disagreements proceed amongst corporations developing AI software.
“Companies like Meta, for instance, are refusing to launch their latest Llama product within the EU, so it's really good to have a base that goes beyond individual territories,” he said.
Although the EU AI law was seen as an try and set a precedent for other countries, the signing of the brand new treaty demonstrates a more coherent, international approach and doesn’t depend on the so-called Brussels effect.
Věra Jourová, European Commission Vice-President for Values and Transparency, said: “I’m delighted that so many international partners are able to sign the Convention on AI. The latest framework sets out essential steps for the design, development and use of AI applications that can construct trust and assurance that AI innovations respect our values – and protect and promote human rights, democracy and the rule of law.”
“This was the founding principle of the … European AI law and now it serves as a blueprint world wide,” she added.
