Paul Nemitz’s idea for International Law on AI

At the AI International Accord Conference on April 28, 2021, organized by Boston Global Forum, Paul Nemitz presented his ideas:

Is it right to take the multilateral agreement on the law of the sea which is a United Nations agreement as a model for our work on AI? The multilateral agreement on the law of the sea is the biggest binding convention of the United Nations. I would say having in mind the comments which we heard from our first speaker, namely the ambition to have as many countries on board as possible as a precondition for such an accord working. I think that makes the law of the sea a great model. Second, the law of the sea at the time was extremely innovative in dealing with a space which was largely unexplored out there beyond the site from the coast 70 percent of the space of the world and this was exactly the reason that states got together and agreed on it. If you look at them today, they are amazing principles of a binding nature and because everybody understood that this new frontier, you know the ocean being at the beginning of exploitation deep sea mining and so on, it needs common rules to avoid conflict in the future and the United Nations and states went very far in this, even set up a court and a dispute settlement system which today works fairly well. So I think we should have our ambitions high. We should go beyond just repeating non-binding types of principles which we already find in the OECD code and in many ethics’ codes. We should focus on the specific nature of the technology which provides us with the opportunity to set out principles on which agreement is possible. We should aim for binding rules because we are dealing with power relations here. Power relations between states and private corporations. Let’s not forget that those corporations which are at the forefront of operationalizing AI, in terms of financial capacity, are bigger than many states of this world. So, I can imagine that many states actually have an interest in binding rules for this reason alone. We’re dealing with power relations between the states and individuals not only individuals of the own country and power relations between corporations which use AI and people. Since these power relations are one of the key reasons why we are dealing in a binding way with AI in legislation in the European Union, this would be my encouragement to our work, that we recognize that this element of managing power from technology and controlling power is one of the key challenges of the future and underlies our willingness to come together in this group and look for global principles which help us to give a frame to this new power over people, over states, over resources, over security. To close I recommend to everyone the new report from Bruce Schneier at the Belfer Center at Harvard University came out in late April 2021 on AI as a hacking tool in the future and he describes a world in which all binding principles, all decency, all systems of security which we may have in technical devices will be systematically undermined at a by AI systems which have been built for this purpose. So I do not believe that we need an agreement which becomes a marketing tool for AI. I do believe we underestimate the risks. And I do believe here we have another principle which we will find agreement on by many. Namely, that AI which has this purpose as described by Bruce Schneier should be out ruled namely AI which serves as a tool for criminal purpose.