00:00I think for robotics, sort of as a tool, a technology, and like all other technology,
00:06it won't really be safe and responsible unless the user of it is safe and responsible.
00:26Some robots have traditionally been used mainly for casks that are very repeatable,
00:29where you have to do the same thing over and over again.
00:32So therefore, they're used a lot in manufacturing, things like manufacturing cars,
00:35also because that kind of manufacturing has very heavy objects that may be dangerous for humans to handle or difficult.
00:41But now we're seeing them move into a bit less repeatable and more dynamic environments.
00:46So things like agriculture is a situation where you have a lot of structure,
00:51like a strawberry field has a very clear structure, but things are also a little bit different.
00:55So one strawberry is never the same as the other one.
00:57You have to always pick them in a little bit different ways.
01:00So a situation like that is much harder for a robot to handle.
01:06One thing that's especially been gaining a lot of progress in the last years
01:09are these humanoid robots that can help people at home.
01:12They have a sort of body, almost like a human, and are physically able to do any task that a human can do.
01:17But it's sort of their intelligence that's not quite where it needs to be for them to re-face a human worker for a given task.
01:25So I think in 2030, we'll see some of the earliest adopters have these kind of robots in their home to help them with data tasks.
01:31But they'll maybe be a bit too expensive, and they'll maybe break a bit too often for the average user to have them at home.
01:38GPT is trained by giving it lots of examples of text on the internet, and then it learns to predict what comes next in that text.
01:48But then a robot doesn't just need language, it also needs vision, sensing, and acting.
01:53And so you can supplement this language understanding with what's called a vision model.
01:59A robot can use the same kind of model to understand what goes on around it.
02:03But what's very special for robots is that they also have actions, so you can also train up what's called a vision-language action model,
02:10where, based on examples, it can learn how do my actions influence the world around me.
02:19Most of our work lives will probably change to some degree because we can rely more on advanced robots and AI systems to help us in our daily tasks.
02:27Some jumps will probably disappear, and it's quite important for countries to also have support for those whose jump doesn't need to be done anymore,
02:35or who can have a different kind of role in their job, maybe more assisted by robots and AI systems.
02:45It's possible to lose control of an AI system because it has found a solution that we can't really fully understand.
02:50So we have to be careful, and I think it's important to invest in research on AI safety.
02:55Like all other technology, it won't really be safe and responsible unless the user of it is safe and responsible.
03:01So one example is this European AI Act that came out a year or two ago, where they made some very clear rules for what we can use AI for,
03:09and depending on the task we use it for, what sort of preconditions we have to take, what we have to be careful about.
03:16So to make these sort of national and multinational agreements is a good way to make robot research and production responsible.
Comments