The Future of Warfare: Anthropic vs OpenAI

0:00Thanks for listening to The Restisted Politics to support the podcast and listen without the adverts and get early access to episodes and live show tickets, go to TheRestistPolitics.com. That's TheRestistPolitics.com. Hi, Rory here with so much happening in the world of AI this week. Both AI powered attacks on a run with Astropics fight with a Pentagon. I've been very lucky to be able to sit down with Matt Clifford again in our exclusive AI mini series where we will be talking about defense, security, geopolitics and even some practical tips on how AI can improve your business. So please sign up at TheRestistPolitics.com to join our member series and listen to the discussion between me and Matt and all the other content that we produce. A lot of our conversations over the last few episodes you have brought back to TheRestistPolitics and in particular they're changing and challenging relationship between Europe and the United States under President Trump. I'm really intrigued by how you think about

1:04the the sort of core principles that are at stake in this anthropic DOD deal and the reason I want to do it is that I think right now because anthropic is sort of the liberal coded slightly European, hence the question that we answered but a lot of people like this is outrageous, you know this is DOD trying to overturn reasonable ethical company drawing red lines around the product. But I want to do a thought experiment for you which is imagine a camel Harris has been elected. Elon Musk says of course you can use styling in defense by the way no one else has anything like it but here are my red lines about how you're going to use it and they're deeply ethically based and in particular the my number one issue is I don't want it using anything that could increase the likelihood of a new clear event in Ukraine and therefore you're not going to use it for blah blah blah blah blah. I think the same people who are protesting now that Hegsat's overstep would be saying why could it should a private company be setting red lines over what a democratic elected government does with the technology? Yeah I think that's right and I think at some

2:06level the philosophical position that Hegsat holds which is that if you bought a bit of kit and you're a military you want to be able to deploy it and you don't want someone stopping you deploying it is absolutely right. But actually as you've pointed out with Musk and as the Saudis found out when they bought some very expensive American airplanes and tried to use them in Yemen and indeed as we found in the Ukraine war where Trump suddenly instructs American companies who we've contracted with not to provide things or as indeed as the US found when it took on China and China suddenly announced it was going to stop all exports of rare arts. Nobody really has sovereignty not even not even the Secretary of War is actually able to instruct without having a lot of friction of various other actors saying what he kind of can't do and that's partly because of all these things that you've talked about the more complicated these systems become the more

3:09dependencies you have I mean if what you've purchased is a hand grenade yeah nobody can stop you using a hand grenade I mean it's basically the explosives and the thing you pull the panel and you chuck it right it's gone and the person who sold it you can't do anything about it but increasingly what we're talking about is defense companies and we saw this in Afghanistan it was absolutely extraordinary what really happened to the Afghan military was not the US decision to withdraw 5,000 troops it was that the US withdrew about 20,000 contractors many of his jobs were about doing the software updates on the helicopters and who sat in these bases and as soon as they were removed all these helicopters and planes are the Afghan Air Force had bought were complete uses because they basically needed a software engineer tinkering them every time they landed and if they didn't get that they can take off again right and these companies have designed their contracts like that because they don't make their money now by selling a hand grenade they they make the money through

4:10the software engineers and maintain them and that's cheaper of course for the Department of War 2 right so part of war hasn't bought from anthropic a right hand grenade what they've bought is anthropic services to run these things yeah they've bought masks styling and all his people who do it right um so short of nationalizing styling yeah or anthropic which they could do and of course this and this is what you probably would do in war right I mean if you really yeah were worried about mask you'd be like okay so sorry styling because now I was yeah right um short of doing that I think you are very vulnerable to companies no longer providing I mean it just as you do when you privatize anything I mean I found this in the prison system where we managed to bring in a company who offered to do all the maintenance in the UK prisons for a quarter of the price the government was spending and then turned around and said uh we can't maintain your prisons anymore and we do much about it yeah well I am I mean I have to say in general I find myself um

5:13partly kids you know I find myself very sympathetic to anthropic and I think they are trying to do the right thing in very difficult circumstances but I do buy the argument which you know has been flowing around the internet the last few days that there is a contradiction in spending three four years saying by the way we're building incredibly dangerous technology that it's equivalent of import to nuclear weapons then being you know upset when uh governments as a war in that case we we really need to control it what does that mean the other thing that's going on here which is so troubling and and the reason you know if the world comes with grinding halt or autonomous weapons are flying around killing everybody without proper human sufficient is competition between these companies because as soon as anthropic refuses to supply Sam Altman turns up and seems to initially say well I'm not too bothered about surveying everyone and killing everyone with autonomous weapons and then a couple of days later I was like oh okay maybe I saw you did that and I'm a backtrack is it yeah someone put us the first live negotiation on Twitter between a private company and the department of defense and that's also another thing that we don't talk about

6:16enough which is you've you've just put your finger on it right this AI development could not be happening at a worst time because it's taking place in the context the Trump administration right who is Pete Hexif he's a a TV news anchor whose greatest claim to fame as Secretary of Wars that he can bench lift two hundred ninety pounds and who conducts sensitive secret negotiations over Twitter and we have a president who's has a weird instinct for deregulation which means he's not going to get involved in really telling anyone what to do in terms of safety but he's also vindictive narcissistic capable of waking up in the morning and randomly punishing anyone against so if you were really thinking what is the ideal political framework in which to develop what could be some of the most dangerous world-changing technology in the world my god we're

7:17unlucky that it's happening in the middle of Trump's America I I actually would put more confidence that the Chinese government has long-term thoughts protocols procedures processes to assess its AI development than I do with the US amendment well I think one of the things that and it is goes back to things that I've been banging the drum on for a while it's you know I I don't think to the first question that we talked about that there is a realistic case where we sort of relocate one of these companies and make them British but I do think this is why it's important that Britain is building domestic compute but also trying to build AI companies that are relevant and important because it strikes me that particular with this administration which you know I think it would itself say has a more transactional approach to Europe we need to have bargaining chips and we need to be relevant and you know we at the moment we don't have the hard assets in AI to have a seat at the table but it sounds to me as though you just told us at the beginning to show that any policymaker in the European Union needs to say okay let's put 450 million people together

8:25let's put the economies of Germany France the European Union together with Canada the UK God knows Australia Japan South Korea and let's solve the problems that Matt's raised let's work out how the hell we do build a two gigawatt center yeah let's work out how the hell we get the advanced chips yeah and let's make it an objective to get one of these Frontier AI models here because everything that you've told me is terrifying thank you so much for listening if you want to hear the full episode as well as the rest of our AI miniseries with Matt Clifford the government's former AI advisor and a man who's leading so much of the policy around the world please sign up at therestaspolitics.com and become a member