AI in the Techno-Economic War with China with Vinod Khosla, Todd Young, and Jacob Helberg

I've called for treating state-of-the-art models, which are called SOTA models or frontier models, and restricting them like we would like a rocket technology or a Lockheed technology. I don't think we can keep permanently ahead, but we can slow China down. The argument I hear for not closed sourcing these state-of-the-art models is there's Chinese implants in our companies, there's cyber attacks that take this on. All that is true, but we can slow them down.

Hey everyone, Erik here from Turpentine. In this episode of The Hill and Valley, recorded live in Washington, D.C. at the 2024 Forum, Vinod Khosla and Senator Todd Young join Jacob Helberg to discuss the national security implications of AI. Vinod and Senator Young expand upon the question of, would you open-source the Manhattan Project? The first voice you'll hear is Jacob Helberg, followed by an initial response from Vinod Khosla, and then Senator Young. Let's dive in.

Welcome. So, we're going to get started with a discussion on the U.S.-China techno-economic war. Vinod, you've been a vocal proponent that China and the U.S. are in a techno-economic war. Where does AI fit into that war? Is AI the modern-day equivalent of the atomic bomb? And if so, Should the US be thinking about mobilizing for a new Manhattan Project?

AI is unlike most other technologies. In fact, the last big change in technology was when we got the steam engine, the electric motor, those when we replaced human muscle power. This time it's replicating human brain power. So think of AI as almost all expertise will be near-free, whether you're talking structural engineers or oncologists or primary care physicians. Are rocket engineers. So why it's so important in this techno-economic war with China? There's clearly the hard power war machine AI. There's cyber warfare. There's what I call persuasive AI on things like TikTok. But the most important piece is often ignored. Whoever wins this war. Wins the economic race. And I think because of that will influence most of Southeast Asia, Africa, Latin America with their social philosophy. So winning the race for AI wins economic power, which then lets you influence social policy or ideology. And President Xi has been very clear on what ideology he wants. And he said internally in China, there's no compromising on that. So I think AI has really strong economic consequences, and hence social consequences.

I want to turn to a question for you both. A lot has been said about how AI is drastically changing warfare and outcomes on the battlefield. What does an AI-enabled conflict between 2 superpowers look like? Senator Young.

Yeah, thank you so much, Jacob, for the questions. It's great to be here with all of you and with Vinod. And, and I would certainly embrace what Vinod said about the, the tectonic implications of continued development and deployment of AI technologies. I think as a policymaker, It's incumbent upon us to prepare for whatever the future holds. Perhaps it is more modest in terms of the impact across certain areas of society. But if we're not preparing for the most dramatic implications, we've missed something. And no area is this more important than in warfare. You know, I think one of the things, this is perhaps the most discussed implication of AI technologies and what we anticipate it enabling in the fairly near future will be the automation of not just certain physical processes, but literally of decision-making, right? Life and death decisions, which can be automated. And so I actually believe there will be, in fairly short order, a lot more comfort developed around that idea, knowing that our adversaries are likely to empower machines to make very intelligent decisions very quickly through advances in AI technology. And if we have to have an intermediary using our gray matter to make those decisions, it will slow us down. It could lead to suboptimality. There are other decisions that will be easier, one could see, to automate areas of cybersecurity, logistics. We'll see automation, we'll see adoption of AI technologies probably more quickly in some of these areas. There'll be unglamorous areas of warfare that will be transformed by AI. Think preventive maintenance, maintenance for vehicles and other platforms. And I think even those decisions that remain under the purview of human beings, and there'll be a whole lot of those, will be improved handsomely by the use of AI technologies to come up with certain, you know, a handful of options in any given scenario. And then you apply human judgment informed by human values. To those different options, leading to faster decision-making, the ability to learn from your decisions very quickly and change course if you need to, again, informed by AI technology. And then sort of the last prediction I have is that the data we use to train different AI technologies for warfare will oftentimes be wrong. We'll be surprised, right? It'll be an entirely different scenario than, than was expected. And so some systems will very quickly be found to fall short, and we'll have to learn how to replace those decision-making tools with other systems, be very adaptable.

But you've been very eloquent in a Medium post about this as well.

Yeah, obviously there's traditional uses, smart robots, smart aircraft, smart drones, but there's even more creative uses of AI. About 2 or 3 weeks ago in The Economist, there was a great article of how over the last month Ukraine has been using AI. It's all hacked together. Picking up unusual cues in the environment to target Russian soldiers because they've been short of weapons. So there will be lots of creative uses. But the one, and this pertains to the TikTok issue, President Xi has said the battle for mind control, I think this is a quote, will be in a smokeless battlefield. And he's referring to the internet. I think that's pretty close to a direct translation. And I think that's where we will see what I call persuasive AI, both influencing our elections. I'd be surprised if there aren't millions of bots trying to be persuasive in our election cycle this time. But beyond that, influencing young kids. He even said in this particular paper I saw, or translation I saw, that this battle for mind control is long and will take patience, yet there shouldn't be any compromise on ideology.

That makes sense.

We'll continue our interview in a moment after a word from our sponsors. If you don't already subscribe to Turpentine's industry-leading newsletters, like our new daily AI newsletter, emergent behavior, or media empires, you should. But that's not what I'm here to tell you about. The platform we use to power these newsletters is called Beehive, and it's excellent. First of all, it was started by the same early team who helped build Morning Brew into a $75 million newsletter business. And they built Beehive to offer that same powerful functionality to anyone sending emails, from essayists to business owners. The platform is beautiful, their text editor is intuitive, and they help you scale your audience with custom growth features. Beehive has powerful tools to help you monetize your content. You can easily launch paid subscriptions or pursue an advertising model. The Beehive platform will even connect you to premium brands to sponsor your newsletter. Not only do we use them, but thousands of the top newsletters in the world also use them, like Milk Road, Blockworks, the Lindy Newsletter, and so many more. Beehive's founder hooked up Upstream listeners with a sweet deal. Get 20% off for 3 months with code Upstream. Visit beehive.com, that's B-E-E-H-I-I-V.com, to get started. Hey all, I'm hearing more and more that founders want to get profitable and do more with less, especially with engineering. Listen, I love your 30-year-old ex-Fang senior software engineer as much as the next guy, but honestly, I can't afford them anymore. Founders everywhere are trying to turn to global talent, but boy, is it a hassle to do at scale, from sourcing to interviewing to on-the-ground operations and management. That's why I teamed up with Sean Lanahan, who's been building engineering teams in Vietnam at a very high level for over 5 years, to help you access global engineering without the headache. Squad, Sean's new company, takes care of sourcing, legal compliance, and local HR for global talent so you don't have to. With teams across Asia and South America, we can cover you no matter which time zone you operate in. Their engineers follow your process and use your tools. They work with React, Next.js, or your favorite frontend frameworks. And on the backend, they're experts at Node, Python, Java, and anything under the sun. Full disclosure, it's going to cost more than the random person you found on Upwork that's doing 2 hours of work per week, but billing you for 40. But you'll get premium quality at a fraction of the typical cost. Our engineers are vetted top 1% talent and actually working hard for you every day. Increase your velocity without amping up burn. Head to choosesquad.com and mention Turpentine to skip the waitlist.

If you look at the AI supply chain, it's quickly becoming global. TSMC is in Taiwan, DeepMind is in London, ASML is in the Netherlands, and now the UAE is positioning itself as a global compute hub to help our companies train advanced models. There is a policy debate going on between those who are comfortable with a fundamentally global supply chain And those who favor the idea of a more sovereign approach to AI. The basic rationale is that if AI is like the atomic bomb, it wouldn't make sense to exacerbate our dependence on other countries for our most vital capabilities. So my question for you both is, are you concerned at all that unless we drastically upgrade our energy infrastructure and ramp up our investments in large computing clusters here at home, that we could find ourselves in a situation a few years down the line where we're overly reliant on foreign partners like the UAE to train our most advanced AI models and capabilities. I'm curious as to where you both fall in this debate, and maybe we could start with Vinod and work our way—

I'm less concerned about the UAE. I think there'll be enough compute power in the US and in Western Europe, and I think it's generally a good thing that various countries have their own efforts. And it reduces critical dependence. So I think that's in general a good thing. I'm especially not that worried about the Mideast effort. And if there's another center of compute there, that should be fine. I am more worried about what we do in this country and especially the state-of-the-art models and how effective they can be. I do think most of the talent is in this country. And if you're going to do a good job of competing, I think immigration of PhDs and computer science majors and others is so critical in winning this battle. And we've slowed down that import and it has served us very well the last 20 or 30 years. So I do think that's where the battleground will be, not so much efforts, uh, like, uh, like the UAE might mount. We have an effort in Japan based on, based on kanji script, which is tokenized differently. We have an effort in India based on the 13 different scripts in India. So those are good, healthy things. Uh, we tend not to do anything ever in China, uh, ourselves, uh, so So I think more distribution is a good thing, especially, and I think of the supply chain as more people who believe in liberal philosophy as opposed to our freedom of speech and some of the values we have versus people who believe in a different ideology. Supply chain within those areas, I think, will work fine. I'm particularly concerned about Taiwan and TSMC's our dependence on TSMC. I think that's our biggest vulnerability.

That makes sense.

Senator Young, have people been talking about this in the halls of the Senate at all?

Increasingly, you know, my colleagues are discussing these things, which is why we've had a lot of staff interest in this gathering. And, you know, Senator Schumer, myself, Heinrich Rounds, roughly a year ago convened these, these AI forums to educate ourselves. A number of staff members participated. As well. And this has been among the conversations we're having. So in addition to investing in workforce, which is clearly going to be very important, you know, unlocking data, what are, what are some of the perhaps what's, what's some of the business that we've been working on for a number of years that really has not come to fruition that will become very important in this, in this new era, so to speak. One is the need to permit. So permitting reform for our fabs so that we can make the chips that will, you know, lead to processing the data and, and stay ahead of the competition. That's going to be really important permitting with respect to, you know, data centers and the compute clusters. So all of these things are going to be increasingly important. But, but really, I think a good model here may be a variant of the CHIPS and Science Act. So for semiconductors, we have been willing to work with most countries, most countries in the world, but increasingly work less with those who we have a high level of distrust. Knowing that we have extended supply chains for semiconductors. Your average chip crosses 70 international borders between it by the time it ends up with a consumer. So this could be analogous. We're going to have a collaboration across multiple international borders to develop and deploy these systems. And if we can come up with common standards, then our adversary, China, there's some others out there, they're going to have to sell into this market embedded with our standards for things like privacy and bias and, and so forth. So I think that is a more modest vision than coming up with a sovereign AI system. I happen to think it's a better vision as well because, you know, the UAE, as you mentioned, but there's a lot of countries in the Global South who want to partake in this AI-enabled world, and they will find a way to partake with China, if not with us. So let's co-opt them. Let's partner with them and mutually benefit in the process.

Another contentious point of the— on the AI policy debate has been on the open source versus closed model approach. And Vinod, you've been a vocal critic of allowing to open source our most advanced AI models and called for national security You tweeted, would you open source the Manhattan Project? Are you in favor of a know your customer rule? How do you think about whether we should cut off China's access to our most advanced capabilities? And Senator Young, I'm curious to know, how is the Senate thinking about that approach? Because obviously there are trade-offs.

You're gonna start with Vinod, right?

I'll start with Vinod.

No, I'm sorry. I've called for treating state-of-the-art models, which are called SOTA models or frontier models, and restricting them like we would like a rocket technology or a Lockheed technology. I don't think we can keep permanently ahead, but we can slow China down. The argument I hear for not closed sourcing these state-of-the-art models is there's Chinese implants in our companies, there's Cyberattacks that take this on. All that is true, but we can slow them down. I'll give you one example. I can't name the company. They were publishing a lot on their developments. This is not in AI; it's in energy technology. And the Chinese developed a parallel effort that replicated that technology. They then decided to not publish. Cyberattacks went through the roof on this company, but the Chinese effort slowed down. It's fallen behind the US effort. So that's a classic example of what can happen. Not that they won't eventually catch up, but it's pretty important that we slow them down in any way we can. We won't keep them permanently behind. I don't buy the cat's out of the bag argument. I think we can still slow them down.

Senator Young, can you give us a brief peek into the science?

Well, that, that where you ended up, that, that appeals to me because it strikes me as a modest and therefore a realistic approach, try and, and slow down your adversaries, or those who might, you know, use these amazing tools to harm others. And so that would be very consistent with my thoughts about partnering. With other countries and regional partners. Export controls, for example, is an area where I commend the administration for making a lot of progress. Hasn't been perfect, but they've worked with the Japanese, they've worked with, you know, Europeans. We still have some more work to do, but I'd like an expansion of those efforts and those kind of efforts moving forward. I think, you know, open source, closed source, That's highly unresolved by members of Congress. I think we will see a lot of debate about that issue in coming years. My suspicion here again is that we don't end up gravitating towards an all open source or all closed source solution. We're going to have some of both moving forward, recognizing the benefits of the openness but also the vulnerabilities. So the key will be based on people like Vinod and others offering counsel over the last year or so to stay ahead of our competitors, which requires constant iterative, sometimes more than iterative innovation, the training of our workforce and making sure that we have kind of a legal and regulatory atmosphere that lends itself to that. It also requires one other thing. It requires greater visibility into how much progress your adversaries are making. And right now, most of the assessments of that strike me as somewhat facile. We don't have enough information. So we need to institutionalize that effort almost like an Office of Net Assessment within our government to determine where we are versus where China is in developing AI technologies. I have some legislation with Senator Bennett that would that would bring this about.

So I just want to add one comment to that. You know, people talk about international treaties. That's really naive. In nuclear, use of nuclear was verifiable. In biowarfare, use of biowarfare was mostly verifiable. If we look at the SARS virus, if it's used or deployed. In AI, It's not a verifiable technology. And so international treaties are worthless because there's no verification of the technology. And so we should keep this in mind for those who propose we have international treaties. It's very naive. Do we trust President Xi?

I don't.

Thank you so much, ladies and gentlemen, Vinod Khosla and Senator Young.

Great stuff. Thank you. Thank you.

Hey everyone, Erik here. At Turpentine, we're building the first media outlet for tech people by tech people. We're the network behind the show you're listening to right now. We have a slate of hit shows across a range of topics and industries, from our AI and investing cluster of podcasts to shows that drive the conversation in tech with the most interesting thinkers, founders, investors, and influencers like Econ 102 with Noah Smith. We're launching new shows every week and we're looking for industry-leading sponsors. If you think that might be you and your company, email me at erik@turpentine.co. That's erik@turpentine.co, and let's partner together.