Why Nearly All Deepfakes Should Be Illegal: Max Tegmark Speaks To Forbes At Davos About AI

[Music] hi everybody I'm Diane Brady I'm here with Max tegmark who is a Swedish American physicist professor at MIT you're the co-founder of the future of Life Institute and I know you as the one who wrote this famous letter um earlier this earlier last year excuse me calling for a pause in AI I know you're here talking about deep fa fakes what is the state of your mind here at Davos from what you're seeing are people getting the message Yes actually so seven years ago I decided to shift my MIT research group from physics to AI because it's just so exciting and um it's become increasingly clear to me that uh AI is either going to become the best thing ever to happen to humanity or the worst and I'm very much pushing for steering it towards good and um what that the way we need to do that is to treat AI just the same way we treat other powerful Tech you know when you flew here for for example did you lose any sleep about worrying your plane was going to crash because of so basically I'm from the government and I'm here to help you think that the policy makers are the ones who are going to put the breaks on for well the policy makers are going to put safety standards in place the FAA has such good safety standards that the companies have a huge incentive to meet them as quickly as possible so they can maximize their profits right and that private public partnership works great that's why aircraft engines are just insanely reliable they can go a million hours without thing and and and in AI we unfortunately don't have anything like that right now the sooner we get it the sooner we unleash all the creativity of industry and academics to figure out how to make things even more safe I think one of the challenges when you talk about the FAA of course you know we know Aviation as a deep history we know um how the technology works when I hear people like Sam Alman talk about their own technology they don't seem to even know the fullness of what it can do so how do you regulate that exactly that's precisely the CR such a good question the reason jet engines are so reliable is because of mechanistic interpretability we know exactly how it works so we can really model it whereas the powerful AI systems like chat pt4 just giant black boxes we have very little clue on how it works so what my MIT research group works on is actually opening up the black box and so we can understand better how it works how much we should trust it and how to make it more transparency almost what Mozilla AI is trying to do as well right but I think most people here are too pessimistic about how well we can do because they're stuck in this oldfashioned idea that figuring out how it works and making it more robust has to be know by people this is actually something AI can also do so we're developing AI tools where an AI looks at another Ai and tries to figure out how it works sees if it can um simplify it make it more more robust more trustworthy and uh you know as soon as there are safety standards that companies have to meet they have an incentive to go that extra step and I think it can totally be done can I ask you a question have you talk to many people in China since we've got these parallel if not you know multiple tracks of development are they talking about these issues too in the same way you are yeah I actually I've been I've been uh doing my best to make sure that the Chinese worry just as much about AI safety as we do in the west and uh this is definitely beginning to happen in fact the Chinese government was the first government that started putting in real regulations last year because they uh start they became clear to the Chinese government that um they didn't want to lose control over China to some unelect some tech company or something like that and um this is of of course the way we get China to adopt all sort standards out of self-interest and I really am very optimistic about this if you ask you know why is it that China has an equivalent of the FDA for medicines it's not because I persuaded them or the US persuaded them it's a good thing to do out of self-interest they realize a lot of Chinese people were getting injured by by sloppy manufacturing similarly China never persuaded Us in the US to do the FDA we did it to protect their own and it's much faster to spin up national regulation first but once you have that you know people from the FDA in the US love talking to the European FDA and the British true and that's almost a can I ask you one question Max which is we are coming into an election year if you saw the Edelman trust barometer there's a 40 Point difference between Republicans and Democrats with Republicans really distrusting technology which is and so when you look at the candidates when you look at the discussions that are taking place in Washington on the campaign Trail are you optimistic that that's being Incorporated because really it has to start first with the lawmakers well I I think we technologists have a huge responsibility also for trust Frankl because this is going to be the year of fake that's why afterwards I want to make a deep fake of you so you can see just how easy it is I am the real thing and we're already seeing there was a big deep fake in Singapore of the Prime Minister and we're going to get so much stuff like this which is really harmful do you want more of a self-regulation among comp that's not enough what what we need to do is just go after the whole supply chain so everybody has responsibility we need laws Banning non-consensual deep fakes that can be mistaken for real I love satire satday Night Live should be able to do all the Deep fakes they want but if it looks like the real thing and the person hasn't given their consent it should just be illegal and anyone who facilitates it should be liable to this is how we battle for example child pornography where where it's not just illegal to make it but also to distribute it and even have it and it's not just for having democracy either by the way you know 90% of well deep fakes are non-consensual of course I'm in the media industry we don't like deep fakes but um just to when you Kay let I'm going to put on my policy maker hat you're talking to me um what is the actual action that should be taken now especially at a time when there's not a lot of discussion taking place across the floor so there is like 90% bipartisan support in the polls Banning non-consensual defects I don't know one other issue where Republicans so right this is a this is a great easy victory for people Reaching Across the aisle just Banning potential deep fakes the sooner the better before the election will be better than after the election and that's also going to really help crack down on on deep fake fraud which is like massive yeah which has had a huge impact anything else on your radar you want to put on ours because I know that you've been thinking about this for many years you're here where it's a tsunami of AI what what do you want us to think about I think that we both need to keep our eyes on the long game that in a few years we might get AI that's smaller than all of us and which at which point we really have to have artificial general intelligence yeah and even super intelligence and at that point of course we really have to have very good governance to make sure we keep it under control but we shouldn't just gaze into the distance we have to look at okay what's the next play and that that's why I'm talking about deep fakes heree everybody agrees on it the abuses are terrible you you probably saw this story of this California woman the mother who gets called up by who she thinks is her daughter panicking saying she's been kidnapped you know we just we should stop scam new tools and the difference is four years ago that was super expensive you'd practically need a movie studio right now anyone can do it in five minutes I'll def pick you in a few minutes you know so I can't wait we can win this one very easily and and and I think politicians also deserve to get a just a quick win showing that they can do something bipartisan they're going to get a lot of popularity from winning that one and they're going to help safe election and then then they can build on that momentum take on the next thing and next thing and not just obsessed about gazing into the distance the whole time and that the foundation of trust thanks for joining us yeah exactly

Channel: Forbes Breaking News

Views: 2,657

Keywords:

Id: yMWrLdkg9o0

Length: 8min 15sec (495 seconds)

Published: Fri Jan 19 2024