00:00 Well, let's get more on this now with Virginia Dignam, a member of the European Union's high-level
00:04 expert group on AI and a professor of social and ethical artificial intelligence at Sweden's
00:11 University of Umeå.
00:12 Thank you very much indeed for being with us.
00:14 So we heard the EU Commissioner Thierry Breton describing the plans as historic, saying that
00:20 it sets clear rules for the use of AI.
00:23 Overall, how do you think, how significant do you think it is, this provisional deal
00:27 that's been called the AI Act, and how necessary is it?
00:32 It is very necessary, and it is very important.
00:36 I agree with the Commissioner that it's a historical moment.
00:40 It's the very first in the world set of rules established specifically for AI, and it's
00:48 based on the protection of fundamental rights and European values.
00:52 It is an example, and it is, I think, something which will be a flagship or a pointer for
01:00 all other developments in the world.
01:03 I actually, I'm a member of the United Nations Advisory Board on AI, and we were having a
01:09 meeting the last two days in New York.
01:11 I just arrived from New York, and all of us there from all over the world, we are looking
01:15 at the developments in Europe as something which is going to be really influential for
01:20 the rest of our other countries and other regions are going to deal with the AI regulation.
01:26 Okay, so the world is looking at what the Europeans are doing about this and calling
01:30 it a flagship.
01:32 And so, looking at the details of this, what key elements stand out to you, and do they
01:38 go some way to regulating the AI used within systems like CHAT, GPT, and facial recognition?
01:45 The first we need to understand is that it is a provisional agreement.
01:49 We need still to wait some time to read the final text, so there will be some work still
01:55 in the coming weeks towards the final text.
01:58 So I cannot really speak with certitude what will come there, but from what we have seen,
02:03 there are definitely some big steps forward towards strong prohibitions, and there is
02:10 a list of banned applications, including, for instance, remote biometric identification,
02:17 and also other types of manipulative techniques.
02:20 So those, I think, are very strong bans and no-goes, red lines for how AI should be used
02:29 within Europe.
02:31 And at the same time, with respect to general purpose models or general purpose AI, as the
02:39 Act talks about them, there are some clear indications that those systems need to comply
02:48 with at least some transparency requirements, and if they are of systemical level, then
02:56 there needs to be some extra steps and extra requirements to those that develop and those
03:03 that use the systems.
03:05 And could this rulebook help, or might it hinder researchers and start-ups in the AI
03:11 industry?
03:13 Research is explicitly set out of the regulations, so while things are in research level, there
03:20 is not many requirements.
03:24 Once things start being used in practice, then of course we need to take into account
03:30 the regulations, again, following this risk-based approach and the different levels of risk.
03:37 But that's not different than in other areas in which research, when we move from research
03:44 to application, there are extra rules.
03:47 I don't think that it's a hinder for start-ups.
03:50 I think that it gives start-ups a direction in which to go.
03:54 It gives also a kind of a level playing field in which everybody knows what are the rules
04:00 of the game.
04:01 A game without rules is a game where no one can win, no one can play.
04:05 So I think it's important for also, especially for start-ups, to have this type of opportunities
04:12 to also innovate within a specific set of rules, and also innovate in the new standards
04:19 and the new requirements that are coming with this law.
04:25 Virginia Dignam from the University of Umeå in Sweden, thank you very much indeed for
04:29 joining us.
Comments