WEBVTT - Killer Robots and AI on the Battlefield: the Pentagon vs Anthropic 0:00:00.560 --> 0:00:04.360 Dario Amida runs the three hundred and fifty billion US 0:00:04.400 --> 0:00:08.200 dollar tech company Anthropic. In January, he wrote an essay 0:00:08.560 --> 0:00:12.520 calling it the adolescence of Technology, a line he lifted 0:00:12.600 --> 0:00:14.720 from the Jodie Foster film Contact. 0:00:15.200 --> 0:00:19.280 If you should meet these vegans and wel delmented only 0:00:19.600 --> 0:00:22.919 one question to ask of them, What would it be? 0:00:24.040 --> 0:00:26.960 Well, I suppose it would be how did you do it? 0:00:28.320 --> 0:00:32.200 How did you evolve? How did you survive this technological 0:00:32.240 --> 0:00:34.479 adolescence without destroying yourself? 0:00:35.120 --> 0:00:37.600 I was always a science fiction fan as a kid, 0:00:37.680 --> 0:00:40.120 so you know, I watched this movie when when I 0:00:40.200 --> 0:00:42.560 was growing up, and just the you know, the idea 0:00:42.720 --> 0:00:45.720 pose really really seemed to fit the situation we're in 0:00:45.760 --> 0:00:49.600 with AI, where you know, we have this we're starting 0:00:49.600 --> 0:00:51.680 to get these immense powers with AI. 0:00:53.440 --> 0:00:56.120 Dario warns, the world isn't ready for AI. 0:00:56.640 --> 0:00:59.040 You know, it's a bit like a you know, a 0:00:59.120 --> 0:01:02.600 kind of teenager. You have all these new powers and abilities, 0:01:02.600 --> 0:01:06.119 mental and physical, but you know you haven't necessarily adapted 0:01:06.160 --> 0:01:08.600 to them yet, and so you know, this is I 0:01:08.640 --> 0:01:10.640 feel like this is something that we're going through. 0:01:11.560 --> 0:01:14.039 This year, Anthropic has been locked in a battle with 0:01:14.120 --> 0:01:18.000 the Pentagon over whether it's clawed chatbot can be used 0:01:18.040 --> 0:01:22.880 for mass surveillance and fully automated weapons. Now it's rival 0:01:23.280 --> 0:01:26.479 Open AI has stepped into the ring and cut its 0:01:26.480 --> 0:01:32.720 own deal with the government. I'm Nicole Johnston and you're 0:01:32.760 --> 0:01:37.839 listening to seven AM today David Rowe from the Australian 0:01:37.880 --> 0:01:43.440 Strategic Policy Institute on Tech titans, robodogs and should AI 0:01:43.760 --> 0:01:56.040 help the Pentagon fight wars. It's Monday, March nine. David, 0:01:56.040 --> 0:01:58.360 could you start by giving us a bit of background 0:01:58.400 --> 0:02:02.000 about this AI company, Anthropic, and how it ended up 0:02:02.000 --> 0:02:04.120 in such a bust up with the Pentagon. 0:02:04.640 --> 0:02:08.800 So Anthropic is one of the big three AI labs, 0:02:08.880 --> 0:02:12.240 the leading AI labs in the United States and hence 0:02:12.280 --> 0:02:18.400 the world. Anthropici, which is building very very capable AI systems, 0:02:18.840 --> 0:02:22.680 was of interest to the Pentagon. The Pentagon signed a 0:02:22.720 --> 0:02:27.519 contract under Joe Biden's administration with Anthropic for Anthropic to 0:02:27.560 --> 0:02:30.959 provide AI at a classified level, so you know, the 0:02:31.240 --> 0:02:33.840 sort of the most secret, the most elite, the most 0:02:33.919 --> 0:02:38.440 high end type of work. It's key founders, Dario Miday 0:02:38.480 --> 0:02:42.680 and his sister Daniella Amiday, put a couple of hard 0:02:42.800 --> 0:02:46.320 conditions on its arrangement with the Pentagon. 0:02:46.120 --> 0:02:48.840 And so we have said to the Department of War 0:02:49.400 --> 0:02:53.240 that we are okay with all use cases basically ninety 0:02:53.280 --> 0:02:55.120 eight or ninety nine percent of the use cases they 0:02:55.160 --> 0:02:58.120 want to do, except for two that we're concerned about. 0:02:58.480 --> 0:03:02.840 One is domestic now surveillance. Case Number two is fully 0:03:02.880 --> 0:03:06.480 autonomous weapons. This is not the partially autonomous weapons that 0:03:06.520 --> 0:03:10.200 are used in Ukraine or you know, could potentially be 0:03:10.320 --> 0:03:13.919 used in Taiwan today. This is the idea of making 0:03:13.960 --> 0:03:16.320 weapons that fire without any human involvement. 0:03:17.240 --> 0:03:20.680 When Trump took over, the Pentagon said no, we're not 0:03:20.800 --> 0:03:23.960 happy with that. We want you to agree that we 0:03:24.080 --> 0:03:27.880 can use your systems for quote, all lawful purposes. 0:03:28.080 --> 0:03:31.840 Defense Secretary Pete Hegseth gave CEO Dario Onmaday until the 0:03:31.960 --> 0:03:34.760 end of the week to sign a document ensuring the 0:03:34.800 --> 0:03:37.800 military would have full access to the company's AI model. 0:03:38.160 --> 0:03:41.320 Officials are considering invoking the Defense Production Act to force 0:03:41.360 --> 0:03:43.600 Anthropic to do what the military wants. 0:03:44.120 --> 0:03:47.320 Anthropic thought about this for a while and then said no, sorry, 0:03:47.400 --> 0:03:51.400 we're not comfortable with this. We insist on our two conditions, 0:03:51.800 --> 0:03:59.280 and then everything went pair shape from there. One other 0:03:59.360 --> 0:04:03.040 thing that is worth noting is that Anthropic is also 0:04:03.200 --> 0:04:08.000 the least favorite of the Trump administration's AI companies. 0:04:08.520 --> 0:04:11.840 President Trump didn't wait for the five pm deadline to pass, 0:04:12.080 --> 0:04:15.760 announcing that three forty seven pm he's cutting off AI 0:04:15.840 --> 0:04:19.880 giant Anthropic, posting on truth social the United States of 0:04:19.880 --> 0:04:23.680 America will never allow a radical left walk company to 0:04:23.839 --> 0:04:26.760 dictate how our great military fights and wins wars. 0:04:27.320 --> 0:04:31.560 Donald Trump himself also announced that Anthropic could not be 0:04:31.760 --> 0:04:35.760 used by any US government agency. So we're now in 0:04:35.760 --> 0:04:38.960 a position where arguably the most capable of the US 0:04:39.040 --> 0:04:42.120 companies doing this kind of work is now effectively cut 0:04:42.160 --> 0:04:45.800 out from providing those sorts of support to US war fighters. 0:04:46.040 --> 0:04:48.320 So we then had Sam Altman from the Open AI 0:04:48.520 --> 0:04:51.360 or chat GPT step in and he kind of deal 0:04:51.400 --> 0:04:52.120 with the Pedagon. 0:04:52.760 --> 0:04:55.479 What happened after that, Yeah, So it was only a 0:04:55.480 --> 0:04:58.320 matter of days, a couple of days before OpenAI announced 0:04:58.360 --> 0:05:02.599 that it was signing this agreement the US military. What 0:05:02.760 --> 0:05:07.640 is confusing everyone is that Sam Altman says that those 0:05:07.680 --> 0:05:11.680 two conditions that Anthropic was insisting on are part of 0:05:11.680 --> 0:05:16.400 their contract with the Pentagon. So the Pentagon, having refused 0:05:16.560 --> 0:05:21.520 to agree to making exceptions for autonomous lethal strike and 0:05:21.760 --> 0:05:25.960 domestic mass surveillance. It's now agreed to those same conditions 0:05:26.000 --> 0:05:31.280 with Open AI somewhat suspicious and has not surprisingly led 0:05:31.320 --> 0:05:34.320 to a lot of skepticism in the AI community and 0:05:34.360 --> 0:05:38.480 people commenting about this. One other thing worth remembering is 0:05:38.520 --> 0:05:43.080 that the personal history kind of goes back between Altman, 0:05:43.320 --> 0:05:46.359 Ammi Day and others. It's a kind of like a 0:05:46.360 --> 0:05:49.719 big dysfunctional family tree. The whole you know, the top 0:05:49.800 --> 0:05:53.360 end of the AI community in Silicon Valley. They all 0:05:53.360 --> 0:05:55.800 know each other, they've all worked together. Well, there's a 0:05:55.839 --> 0:05:57.920 lot of kind of personal rivalry, there's a lot of 0:05:58.520 --> 0:06:02.240 in some cases personal animals between the main players. So 0:06:02.320 --> 0:06:04.479 that just sort of feeds into this sense of, you know, 0:06:04.520 --> 0:06:08.200 you've got two competing companies with potentially conflicting or different 0:06:08.200 --> 0:06:10.880 sets of values, and that this is a real kind of, 0:06:10.960 --> 0:06:12.719 you know, stand off between tech titans. 0:06:12.880 --> 0:06:15.480 David, if we could have a look at the bigger picture, now, 0:06:15.680 --> 0:06:19.440 Anthropic has these two red lines about how its technology 0:06:19.520 --> 0:06:22.919 is used. But isn't AI already being used on the 0:06:22.920 --> 0:06:27.400 battlefields in Gaza, Ukraine possibly across the Middle East right now? 0:06:28.040 --> 0:06:32.200 So could you describe for us how militaries are employing it? 0:06:32.520 --> 0:06:37.040 So at the moment. Most military uses of AI are 0:06:37.160 --> 0:06:42.279 actually about information processing and turning an ocean of data 0:06:42.320 --> 0:06:47.240 into a comprehensible picture that human operators can use to 0:06:47.320 --> 0:06:52.360 make better decisions in real time, often under huge amounts 0:06:52.400 --> 0:06:56.120 of pressure. 0:06:53.760 --> 0:06:59.200 Chasing us human vatching with your tonk doesn't it work? 0:07:00.080 --> 0:07:02.839 Are about to witness a live operation. 0:07:04.240 --> 0:07:11.760 There is some move towards automation of operations like strike operations, 0:07:11.800 --> 0:07:15.800 so the Russians, for instance, in Ukraine, are thought to 0:07:15.880 --> 0:07:20.080 be introducing more and more automation into dropping a bomb 0:07:20.200 --> 0:07:22.560 or or firing on that target. 0:07:22.920 --> 0:07:27.440 We see a Ukrainian tank moving at speed towards the factory. 0:07:31.480 --> 0:07:36.239 It's deep inside the kill zone and Russian attack drones 0:07:36.520 --> 0:07:37.920 will try to destroy it. 0:07:39.760 --> 0:07:45.400 But for the time being, using machines to decide on 0:07:45.480 --> 0:07:49.920 a target and fire on that target without any human 0:07:50.000 --> 0:07:54.760 involvement is still considered to be an ethical red line 0:07:55.000 --> 0:07:59.320 for most countries around the world for very obvious reasons. 0:07:59.440 --> 0:08:01.760 I mean, it goes to the heart of the idea 0:08:01.800 --> 0:08:04.560 of the sanctity of human life. If we start saying 0:08:05.080 --> 0:08:11.239 that we will allow non sentient, non conscious, morally unaccountable 0:08:11.320 --> 0:08:15.280 machines to make life or death decisions about human beings. 0:08:15.280 --> 0:08:19.880 That is a huge rubicon to cross. That is something 0:08:19.920 --> 0:08:21.440 that is not happening yet. 0:08:22.480 --> 0:08:27.440 So is that essentially the argument that Anthropy has for 0:08:27.520 --> 0:08:30.720 how it's technology is being used. It's all about ethics 0:08:30.760 --> 0:08:31.280 for them. 0:08:31.560 --> 0:08:36.720 Yeah, although interestingly, Anthropic Statement didn't actually say they were 0:08:37.280 --> 0:08:43.200 against using autonomous strike per se. They said that the 0:08:43.320 --> 0:08:48.040 technology is at the moment not good enough to do 0:08:48.080 --> 0:08:52.840 it reliably and without making mistakes. So it was very 0:08:52.840 --> 0:08:56.480 interesting that for a company that is known for its 0:08:56.480 --> 0:09:00.760 safety consciousness with AI seem to be taking a slightly 0:09:00.840 --> 0:09:05.440 flexible position on the idea of killer robots. 0:09:07.200 --> 0:09:23.000 Coming up the Frightening Future of War run by AI David, 0:09:23.000 --> 0:09:25.400 I was actually in China last year and I saw 0:09:25.960 --> 0:09:30.480 the robodogs, for example, that they've developed. Could you tell 0:09:30.559 --> 0:09:35.800 us more about some of this technology killer robots as 0:09:35.840 --> 0:09:38.600 you describe them, What is really out there on the 0:09:38.640 --> 0:09:40.040 battlefield now? 0:09:40.600 --> 0:09:44.080 So, I mean, those robodogs are fascinating because they make 0:09:44.240 --> 0:09:48.640 fantastic videos. They look amazing. How useful they actually are 0:09:49.400 --> 0:09:53.000 in battlefield situations is still pretty dubious. 0:09:53.240 --> 0:09:57.800 London Mail, the soldiers scan the skies and use handheld 0:09:57.880 --> 0:09:58.880 drone detectors. 0:10:00.840 --> 0:10:04.319 I mean, the Big revolution in warfighting, which we've seen, 0:10:04.600 --> 0:10:09.920 particularly through Russia's invasion of Ukraine, is towards aerial drones. 0:10:10.160 --> 0:10:16.840 The radio alerts us it's tracking back. The drone detector 0:10:16.880 --> 0:10:17.960 doesn't stop bleeping. 0:10:19.360 --> 0:10:23.600 So that's the big area where some form of robot 0:10:23.760 --> 0:10:27.199 is being used. Those sorts of things like land based 0:10:28.000 --> 0:10:32.440 vehicles like robotic dogs, for instance, they're not a big 0:10:32.480 --> 0:10:36.679 part of war fighting at this point. However, we're seeing 0:10:36.760 --> 0:10:42.560 less remote controlled type activities where we actually have a 0:10:42.679 --> 0:10:45.439 human operator sitting there essentially with a kind of a joystick, 0:10:45.640 --> 0:10:50.439 and we're giving more independence and autonomy to machines on 0:10:50.480 --> 0:10:54.600 the ground. Worth also noting that China is very very 0:10:54.640 --> 0:10:58.000 capable at this They are leading the world when it 0:10:58.080 --> 0:11:01.120 comes to robotics. The Russians also have learned a lot 0:11:01.160 --> 0:11:05.080 from the Ukraine War and have been making improvements on 0:11:05.200 --> 0:11:08.880 that front as well. There's also a concern that they 0:11:08.920 --> 0:11:14.360 will have lower ethical standards than democracies. So Australia, I'm 0:11:14.400 --> 0:11:16.679 pretty confident we'll stick to some of those red lines 0:11:16.720 --> 0:11:20.840 around things like automation of lethal force. I'm much less 0:11:20.840 --> 0:11:23.720 confident that China will, and I have next to no 0:11:23.800 --> 0:11:30.040 confidence that Russia will, so I guess the big strategic 0:11:30.120 --> 0:11:34.760 peril here is that our standards are forced to start 0:11:34.760 --> 0:11:39.760 to slip, even in democracies because there is pressure from 0:11:39.840 --> 0:11:43.600 our adversaries. That means that we risk not being able 0:11:43.640 --> 0:11:48.400 to keep up in a battlefield scenario. If you take 0:11:48.559 --> 0:11:51.760 humans out of the loop entirely, you can operate at 0:11:51.800 --> 0:11:54.880 a much, much faster pace, and one of the big 0:11:54.920 --> 0:11:57.320 concerns is that there will just be a slippery slope 0:11:57.360 --> 0:12:00.880 that we all gradually have to go down in order 0:12:01.600 --> 0:12:03.679 to keep up in a war situation. 0:12:04.120 --> 0:12:09.920 David, it's not only Ukraine and China, it's also Iran. 0:12:10.640 --> 0:12:13.720 We're seeing how effective their drones are at the moment. 0:12:14.320 --> 0:12:16.959 But over the weekend we had Sam Oltman. He called 0:12:16.960 --> 0:12:20.480 for a debate about who should have more power over 0:12:20.760 --> 0:12:23.400 advanced to AI, whether it should be in the hands 0:12:23.400 --> 0:12:27.920 of governmental big tech. What do you think how should 0:12:27.920 --> 0:12:30.079 that debate be discussed. 0:12:30.679 --> 0:12:35.560 I would describe the Pentagon anthropic blow up as a 0:12:35.640 --> 0:12:38.760 kind of first salvo in a conversation that I suspect 0:12:38.760 --> 0:12:41.160 that we're going to be having for the rest of 0:12:41.200 --> 0:12:44.959 our lives. The difference with AI companies today is that 0:12:45.000 --> 0:12:48.680 they are making things that are entering every aspect of 0:12:48.800 --> 0:12:52.320 our lives. So we have to ask the question, do 0:12:52.360 --> 0:12:55.160 we want that kind of influence over our lives to 0:12:55.200 --> 0:13:02.560 be entirely generated by the private sector, where the government 0:13:02.640 --> 0:13:06.480 then either acts as a large customer like the military 0:13:06.520 --> 0:13:10.200 does in buying products from the private sector, or as 0:13:10.240 --> 0:13:13.840 a regulator from time to time, where the government says, Okay, 0:13:13.880 --> 0:13:16.440 we're going to set limits on how those products can 0:13:16.440 --> 0:13:20.320 be sold and used. Or do we think that this 0:13:20.480 --> 0:13:24.520 is such a fundamental transformation to our lives that we 0:13:24.640 --> 0:13:29.400 actually need some kind of partnership between the state and 0:13:29.440 --> 0:13:32.880 the private sector in the way this stuff is actually built, 0:13:33.200 --> 0:13:36.760 the way it's sold, the way it's integrated, and the 0:13:36.800 --> 0:13:40.160 way ultimately that is going to influence how we live 0:13:40.200 --> 0:13:44.920 our lives. My view is that AI is sufficiently transformative 0:13:45.720 --> 0:13:48.120 that we all need to have a say in how 0:13:48.160 --> 0:13:50.640 it's going to affect our lives. It can't just be 0:13:50.679 --> 0:13:53.480 a private sector product in which the government then sets 0:13:53.520 --> 0:13:56.280 some kind of laws or regulations around it. It's much 0:13:56.320 --> 0:13:56.920 bigger than that. 0:13:57.360 --> 0:14:00.839 And finally, David, how do you envisage the battlefields of 0:14:00.880 --> 0:14:01.440 the future. 0:14:02.160 --> 0:14:06.240 I envisage one in which machines and automation are going 0:14:06.280 --> 0:14:14.240 to play at the overwhelming majority of the roles. The 0:14:14.360 --> 0:14:18.559 question is to what extent humans remain in the loop. 0:14:19.320 --> 0:14:24.120 If it's just about technical proficiency, then I reckon within 0:14:24.280 --> 0:14:27.840 ten to twenty years there wouldn't be humans involved anymore. 0:14:28.680 --> 0:14:30.920 If we decide that it is a moral and ethical 0:14:31.000 --> 0:14:35.280 question that humans need to remain involved, then we can 0:14:35.400 --> 0:14:37.480 ensure that that's the case. But it will take an 0:14:37.560 --> 0:14:44.360 active decision on our part. Without active ethical discussion, warfare 0:14:44.400 --> 0:14:49.120 will end up being a fight between two lots of machines, 0:14:49.840 --> 0:14:52.760 and only when those machines are fully destroyed will people 0:14:52.800 --> 0:14:56.000 actually start fighting. And then follow that to its logical 0:14:56.040 --> 0:14:58.680 conclusion and we'll get back to sticks and stones. But 0:14:59.280 --> 0:15:02.600 in a nutshell, well, it will be machines fighting machines 0:15:03.160 --> 0:15:08.680 unless we insist on human involvement for moral and ethical reasons. 0:15:10.160 --> 0:15:12.320 David, thank you so much for talking to us. 0:15:12.960 --> 0:15:13.600 Thanks so much. 0:15:24.320 --> 0:15:27.000 Also in the news, people in Katherine are assessing the 0:15:27.080 --> 0:15:30.080 damage after the town's worst flooding in twenty eight years. 0:15:30.720 --> 0:15:34.200 Homes and businesses have been inundated with water which reached 0:15:34.200 --> 0:15:37.440 as high as roof level in some areas. Locals have 0:15:37.520 --> 0:15:41.120 been evacuated to Darwin with the Bureau of Meteorology warning 0:15:41.320 --> 0:15:44.600 more rainfall is possible over the next few days, which 0:15:44.640 --> 0:15:50.040 could see the Catherine River rise even further, and the 0:15:50.080 --> 0:15:53.080 May budget will have no cash for wind or solar farms. 0:15:53.160 --> 0:15:56.800 According to reports, new spending on the renewable rollout has 0:15:56.800 --> 0:15:59.560 been halted in an attempt by the federal government to 0:15:59.600 --> 0:16:03.680 claw back savings. That's despite concerns Australia is behind on 0:16:03.760 --> 0:16:06.880 its target to reach eighty two percent green energy by 0:16:06.960 --> 0:16:11.320 twenty thirty. I'm Nicole Johnston. This is seven am. Thanks 0:16:11.360 --> 0:16:11.880 for listening.