YouTube:
Trump’s Mammoth Arabian AI Deals | Robert Wright & Paul Triolo
0:00 How significant are the Middle East data center deals?
12:03 Why did AI safety hawks quit talking about international cooperation?
18:38 Why build data centers in the Middle East?
27:20 Microchips and security risks
30:46 The unexamined assumptions of Sinophobia
42:00 What ever happened to peace, love, and understanding?
47:46 AI Czar David Sacks takes on his critics
51:11 Heading to Overtime
The whole discussion is worthwhile, but I want to ook at the material on our relationship with China, starting at roughly 30:46. I've taken the AI-generated transcript, put it below the asterisks, and put in some breaks dividing it roughly into segments spoken by our two interlocutors.
* * * * *
R. Wright: there's such a consensus on on the China issue broadly that you don't hear people's arguments challenged very oft they're not forced to get back to first principles and so that's why
P. Triolo: they change but the principles change over time yeah no I
R. Wright: I you've written a lot about this how the goalpost change and particularly for industry and then yes a lot of these the justifications for some of these things you know don't aren't really very well examined um I would argue that the AI one is is falls falls into that because the the idea for example that that um we're going to get to some inflection point and and uh and and and you know China would would you would would could get ahead and develop advanced AI and then use it against us you know again I I I'm a little skeptical that that sort of um dynamic is would would necessarily play out in the way some of the the the the some some people think think it would and and interestingly um you know the the the CEO of Anthropic Dario Amodei has if you haven't read and his his his um
R. Wright: I have
P. Triolo: essay um his
R. Wright: machines of loving grace which has this section on why we have to squash China you know well
P. Triolo: yeah he doesn't he's careful not to say China but it's authoritarian every time he says authoritarian
R. Wright: he refers to an authoritarian block i think we know who he's thinking of as a leader of that and he says he says we have to uh secure and maintain dominance including at one point he used the phrase military dominance I think until they like abandon their anti-democratic ways or something and he doesn't specify what that would mean
P. Triolo: for me I just read it reread it today and his basic argument is we need to dominate so we can show China that we're dominant and then they will change they will change their system which again I I mean I'm sorry but it seems to me a little bit naive um to to think that this is the case and then of course you know then there's the the whole Taiwan issue too which is which is at play here so but but again that's that's the kind of and and if you read that essay you're you know any kind of geopolitical you know great power competition strategist is going to read that and go okay wow you know somebody's in for a rude awakening here if they think that that China is going to you know is going to sort oh go you know bow down to US AI AI dominance and say something you know and and change the system i mean again I'm I just I don't see how we get to that situation
R. Wright: yeah well I think I I think part of his stance may be defensive he's worried he wants to make sure that they can't dominate us with AI and again people really trace out what they're afraid of
P. Triolo: why are they dominating us with 5G right
R. Wright: i mean well I mean this would be different AI is going to be a more important technology than 5G I'd say anyway just sketching out their fear I think i mean one of the unex one of the if I just one of the unexamined assumptions I think is that China would like to convert all other nations to its form of government or something that
P. Triolo: there's no evidence of that
R. Wright: i I know there there isn't but I really think that's behind a lot of this and so I think from from Daario's point of view it's like well okay even if they never bow you know buckle and whatever he wants them to do uh by way of uh paying their respects to democracy even if they never do that at least they won't come uh you know turn us into their slaves which I think is is what he fears again I agree with you there's no evidence that that's their goal but I think that's
P. Triolo: and there's no evidence that that that a Chinese government that that realized that one of its companies had some advantage in the AI space would sort of immediately use that to sort of you know try to blackmail the US to become communist i mean that that's essentially what what um what the US side is asserting here right i mean so anyway I'm I'm just skeptical that that we thought through what it means to get to to that everybody knows that we're at the state where one side is very dominant and then what the the political leadership of either country would do would do knowing they have this capability because most people talk about that as being temporary too it would be so many months then the Chinese would catch up and so the the the debate around that though there's there is no there isn't a whole lot of debate around that it's sort of in certain circles within the US government and think tanks that's just sort of accepted as yes we have to stop China because if they get their first their version of AI whatever or their their decision to use AI will be bad and therefore you know we can't take that chance so that that's sort of what's driving a lot of this again I I really think and and I agree with your your view here that that if that's your if I'm characterizing you correctly is you know that's there's a lot of um there's sort of a lot of unknowns there and a lot of um what I would call sort of almost cavalier treatment of how governments serious governments and serious lead political leadership would deal with a situation like this right i mean that this is not trivial there's a very good if you haven't seen it also this this website called AI 2027 which um is a very and interesting read but it plays out this scenario up to 2027 where essentially you get down to two companies one on each side supported by their governments that are pursuing you know that and trying to get to that to a AGI and again I think there the portrayal there of of of the how the governments might react to that is is it's well done i mean it's a very good attempt at this but it really is sort of not again not necessarily well thought out in terms of okay is that really what would happen um so we're sort of in this weird world where we don't really know how a AGI will manifest and how we'll know this but we're making assumptions about how governments you know who recognize that we're there would react and how they would use that a certain capability which we don't even you know we haven't really defined very carefully so that that's one of my sort of quibbles with or you know major concerns with that because in the meantime you know there's all sorts of risks that we're entailing like we're not going to have collaboration with the US and China and then I've argued as you know um in many cases in many places that you know if China believes that the US is nearing this and they're and they're basing it the hardware bases in Taiwan you know China I think it's again naive to think that China is not going to take action on this you know sooner rather than later force them to act on this
R. Wright: no this is uh you were the first person I found as you may recall who had thought about this seriously uh a fact that if you you know if you take the factories in Taiwan and cut off the flow of advanced AI chips to China but keep the flow going to the west that just alters China's incentive structure in terms of of whether to invade Taiwan i'm not saying it would be the only reason to invade but it's clear how it plays out on the cost benefit side and and also
P. Triolo: also if you listen again let's listen to go back to the Dario Amodei's um essay right so in there he says that he thinks his best guess or his guess is that the current strategy of of denial and export controls and all that is the best strategy um to ensure that that the west or you know dem democracies get AI first but then his he he's pretty clear there that the goal of that is so that they can they can basically use that capability to blackmail China into you know dropping its dropping the communist party you know becoming a democracy so again if chi if you're China and Chinese government and you see that you know here's a here's the CEO of a leading US company saying this you know what are you going to make right i mean it's a pretty big deal and I don't think people really appreciate the the risks that you rightly point out and I pointed out that you know I mean if you're if you're sitting back watching this how do you your calculus on something like Taiwan and its critical role and central role in in the global supply chain you know you're going to it's going to have an impact right you're not going to just just blow it off and and Ben Ben Buchanan in the white the former White House czar in that interview with Ezra Klein seem to seem to believe that oh well China understands the what the export controls are about and they're not going to really they don't you know we should still be able to talk to them about safety and other things and I'm like again I think that there's there's a lot of um sort of misjudgment of China and the Chinese system and how it works and you know what what what's going on here right so I'm a little concerned that that we're um you know again we're sort of sleepwalking at the conflict um based on these these ideas that haven't been very very uh closely examined or well thought out um and the collateral damage you know could be considerable and just one other one other quick item I just ripped a piece on this but you know there's the rare earth too right like in the next couple weeks we might see major companies in the US have to shut down production because the Chinese cut off rare earth because of the export controls right so further collateral damage um of the type that I've been talking about for some time on this So anyway so there's there's a huge sort of cost side to this the benefits are at some future point when we may get to AGI and then we may have some lead over China and then we may you know do something with it right so it but that's the problem is that there's these very different um sort of costs and benefits that are out there and um and then there opportunity costs right because for example we we won't likely collaborate with China or China will not see fit to collaborate with us on AI safety um and so then we're going to be not only in a race but we're going to you know we're not going to probably know ultimately what what what China is doing in the space either just as at some point probably US companies like OpenAI and Antropic will not you know will not be releasing their latest they'll kind of go quiet because they're getting close to something so anyway so it sets up a very dangerous dynamic in my view the whole thing Yeah sets up a very dangerous dynamic that we're not quite in yet but we're sort of in the middle we're like in the first inning of a nine inning game and by the time we get to the seventh inning of this game you know people are going to be running for the exit
R. Wright: and just to drive home something people may not realize in terms of danger uh so all of the most advanced AI chips that Open AI uses and so on they're made in only one place Taiwan there isn't a second factory where Nvidia makes these chips okay and Taiwan is you may have noticed is very close to China china considers it part of Taiwan and in fact the US doesn't even disagree we don't recognize Taiwan as a sovereign nation we have a one China policy which means that yeah in some sense they're part of the same polity we just want them to work it out uh peacefully and you know it had been the case that the the fact that these factories are sending these great chips to China was a deterrent to invade because the factories might get blown up and obviously you care a hell of a lot less that they might get blown up if you're not getting the the best chips
No comments:
Post a Comment