How Worried About China AI Should We Be?

0:00Today on the AI Breakdown, we're discussing how worried we should actually be about China and AI. Before that on the brief, Mercedes has put ChatGPT in its cars, Meta is looking to commercialize Llama, and 42 % of CEOs fear AI could end humanity within the next 10 years. The AI Breakdown is a daily podcast and video about the most important news and discussions in AI. Like, subscribe, and share, and go to breakdown.network for more information. Welcome back to the AI Breakdown Brief, all the AI headline news you need in five minutes or less. We kick off today with a story that has been blowing up all over Twitter and the media because it is such a sensational headline.

0:42Now, of course, the question is, is it just a headline? Is it just sensationalism? Or is there something more here? This week, Yale professor Jeffrey Sonnenfeld held a virtual event called the Chief Executive Leadership Institute. As part of that, he surveyed 119 CEOs that include companies such as Walmart, Coca-Cola, Xerox, Zoom, and more on what their thoughts around AI were. The big banner headline was that 34 % of those CEOs said that AI could potentially destroy humanity in 10 years, and 8 % said it could happen in 5 years. Overall, that means that 42 % believe there is a chance of AI destroying humanity within just a decade.

1:19Now, 58 % said that that could never happen and they are not worried, but 42 % is an enormous number. Said Professor Sonnenfeld, it's pretty dark and alarming. Now, obviously, on this show, alongside the rise of all these new technologies, we've also been profiling the rise in the conversation around AI safety and AI risk. There is no doubt that it has taken an outsized place in the media. And in fact, I believe that's why you're seeing a counter response such as Mark Andreessen's Why AI Will Save the World piece, because there are many who feel like the rhetoric has just gotten out ahead of itself.

1:50Now, interestingly, Sonnenfeld has a framework for understanding five distinct camps within AI, and I think it's pretty useful. The first group he describes include curious creators who are naive believers who argue that everything you can do, you should do. Sonnenfeld compared them to Robert Oppenheimer before the bomb. A second category are euphoric true believers who only see good in AI. A third category are the commercial profiteers who are seeking to cash out and cash in on this new tech. Sonnenfeld says they don't know what they're doing, but they're racing into it. And then there are two categories of safety people.

2:22One, alarmist activists, and two, global governance advocates. The net point, and this I agree with more than anything else that has been said, is as Sonnenfeld puts it, these five groups are all talking past each other with righteous indignation. Now, when it comes to the Twitter sphere, I would characterize the response to this article in the survey as skepticism, to say the least. Eurasia Group president Ian Bremmer writes, percentage of CEOs surveyed at Yale CEO Summit that say AI could destroy humanity in 5 to 10 years, 42%. Percentage of CEOs surveyed at Yale CEO Summit that understand AI, 58 % max.

2:56Box's Aaron Levy writes, this is really getting ridiculous. And OpenAI's developer Relations Logan writes, 42 % of CEOs are out of touch. Ben Collins writes, this says more about the gullibility and or scam artistry of CEOs than the capabilities of the actual technology. Speaking of capabilities of the actual technology and CEOs interested in AI, it turns out that Meta is potentially shifting some of its strategy when it comes to AI. Their LLAMA model has been at the center of the open-source Cambrian explosion. It was released earlier this year, an open-source for non-commercial and research purposes.

3:30Now, according to the information, Meta is planning a commercial release to allow people to actually use this open-source technology for commercial purposes. This would of course be a very different strategy than the approaches taken by competitors like OpenAI and Google. Other companies outside those building foundational models are also looking for ways to incorporate AI into their systems. Mercedes-Benz made headlines yesterday when they announced that they were testing ChatGPT in cars to answer, quote, complex questions while people are driving. Basically, they're looking at ChatGPT as a voice assistant in cars.

4:00So if people want to understand things like how many miles to their destination, they can ask that in natural language. And while I've seen a lot of people being very skeptical of this or saying for some reason or why are they doing this, I gotta say, it actually strikes me as kind of an obvious use case. Think to how many times you've used the voice assistant in your car. The answer, if your car has one, is probably about zero times, because they're just not good. But what if it was of the quality of chat GPT? What if it was programmed directly into the display? What if you could actually use it to interface with your car in a more meaningful way?

4:32I don't think it's insane to think that might be a user experience that's common in the future. From June 16th on, that's today. People who have Benzes with the MBUX infotainment system, which represents around 900 ,000 vehicles in the U.S., can use the voice command, Hey Mercedes, I want to join the beta program, and start trying it out. Now going back to the theme of AI safety, outside of just the extinction risk conversation, there are other people who are thinking more practically about short-term concerns. Some of those arise from specific categories of AI, such as voice cloning. Eleven Labs is one of the companies at the very forefront of voice cloning, and they've just announced something that they're calling AI Speech Classifier.

5:04The basic idea, quite simply, is to allow people to upload any audio sample from which 11 Labs can identify if its technology was used to generate any part of that audio. In their announcement post, they write, Generative AI has simplified the creation of images, text, and audio clips to the extent that they are often indistinguishable from human-made content. At 11 Labs, we believe in the transformative potential of these technologies and their ability to unlock new frontiers of creativity and accessibility. At the same time, we also recognize that to fully harness the benefits of these technologies, we must prioritize the establishment of robust infrastructures that ensure their safe and responsible use.

5:40When it comes to the AI Speech Classifier tool, they write that audio generated by their system has very specific and detectable characteristics. So if people upload content that has been unedited, Speech Classifier can identify whether it was created with 11 labs with greater than 99 % accuracy. However, if it underwent transformations such as reverb transformations, the classifier has over 90 % accuracy. And then obviously the more the content has been post-processed, the harder it is for the model to understand it. Now this is a great step, no doubt, but the challenge is of course that no one company has a monopoly on AI speech or voice mimicking.

6:12Remember, I used Play.ht for my long read Sunday last week. Still, it's great to see companies taking these types of proactive steps, as voluntary private action is going to be a key part of any safe AI future. Last before we wrap for today, one really cool tool. Before Sunset is an AI task planner. Basically, it's a tool that asks you a set of simple natural language questions. What are you going to work on? How long will it take? And how much time is available for that to do? And creates an AI-generated plan for one's day. Now, I don't know if this will actually work in practice, but what I do know is that it's been number one on Product Hunt for the last couple days with almost 1 ,300 upvotes.

6:47So if you are someone who struggles with planning out your day, maybe Before Sunset is worth checking out. Anyways, guys, that is it for today's AI Breakdown Brief. If you're enjoying these, please like, subscribe, and share, and I will be back soon with the main AI breakdown. Today, we're discussing TikTok's massive AI GPU order and how worried we should actually be about China when it comes to AI. Here's a pretty remarkable statistic. This year, in 2023 so far, ByteDance, which is the parent company, of course, of TikTok, has ordered a billion dollars worth of GPUs from NVIDIA. Now what makes that remarkable, other than just it being a large number in general, is that's more than the total amount spent on NVIDIA GPUs by all of China and all Chinese companies just last year.

7:33Now this report comes from a Chinese language news outlet and sources both people close to ByteDance as well as people close to NVIDIA. One of those sources close to NVIDIA says that this year ByteDance has so far received a total of about 100 ,000 A100 and H800 accelerator cards. This is far from the first time that ByteDance and other Chinese companies have placed big GPU orders. Apparently in June 2020, after the release of GPT-3, ByteDance trained its own large language model with several billion parameters using the V100 card, but ultimately dropped the project as its capabilities at the time were just average.

8:04Now when it comes to the AI race, people have been examining China versus the US for some time. In 2019, Sapien's author Yuval Noah Harari wrote a piece for Foreign Policy called Who Will Win the Race for AI? Even back then, pre-ChatGPT, the stakes were clear. Harari writes, The world could soon witness a new kind of colonialism, data colonialism, in which raw information is mined in numerous countries, processed mainly in the imperial hub, and then used to exercise control throughout the world. Those who control the data could eventually reshape not only the world's economic and political future, but also the future of life itself.

8:35Now, more recently, China has been the big boogeyman lurking around the corner of calls for any sort of AI pause. Marc Andreessen's seminal AI will save the world piece that we read on Long Read Sunday last week identified the quote actual risk of not pursuing AI with maximum force and speed as China. He writes, there is one final and real AI risk that is probably scariest of all. AI isn't just being developed in the relatively free societies of the West. It's also being developed by the Communist Party of the People's Republic of China. He goes on, China has a vastly different vision for AI than we do.

9:08They view it as a mechanism for authoritarian population control, full stop. They are not even being secretive about this. They are very clear about it, and they are already pursuing their agenda. The single greatest risk of AI is that China wins global AI dominance, and we, the United States and the West, do not. Another example of this discourse comes from an interview with Senate Intelligence Chair Mark Warner earlier this week. In an interview with Politico, he said, Many of us believe that we are in an enormous technology competition, particularly with China, and that national security means winning the battle around AI.

9:37Warner, who's a Democrat, said that China, quote, has a variety of efforts in AI, and they already actually moved even farther than Europe in having specific legislation. Warner said that he's worried the Chinese government will use AI on a, quote, offensive basis or on a misinformation and deceptive basis against the balance of the world. And that, quote, China is very much ahead of the game in terms of self-regulating AI within their own nation state. Warner repeated what has become a congressional and Senate mantra, that there were mistakes made when it came to regulating social media, with the belief among politicians being that they should have exerted more control over social media before it got out of their hands.

10:09But the big gist of this interview, hold aside everything else, was that China is far out ahead. In many ways, it should be concerning to the U.S. Senator Ted Cruz also had rough words for the U.S. when it came to AI regulation, saying that Congress, quote, doesn't know what the hell it's doing. Echoing things he said before about other tech areas like crypto and Bitcoin, he said, Congress doesn't know what the hell it's doing in this area. This is an institution where I think the median age in the Senate is about 142. This is not a tech-savvy group. Now, Cruz didn't necessarily think that we should do just what Europe had done, saying, quote, I think Europe historically has been far less concerned with creating an environment where innovation can flourish.

10:46But for him as well, China remained the concern. He said, quote, I believe that we need a comprehensive strategy for dealing with China, much like we had under Ronald Reagan for dealing with the Soviet Union and for ultimately winning the Cold War. He says that if Beijing ends up leading the world on AI, I quote, that would be profoundly dangerous to the United States from a national defense perspective, but also certainly from an economic perspective. OK, so on the one hand, you've got all this political discourse where China is the reason, if nothing else, that the U.S. should be focused on AI policy.

11:15But then at the same time this week, we also got a couple examples of, if not detente, then certainly a different interaction between China and the U.S. on AI as well. A Wired report reads, good news. China and the U.S. are talking about AI dangers. Scientists from the world's two superpowers are concerned about the risks of AI, which may offer a bridge to developing global regulations. Now what all this comes from is a recent summit at the Beijing Academy of Artificial Intelligence, or BAAI. OpenAI CEO Sam Altman was there and said, China has some of the best AI talent in the world. Solving alignment for advanced AI systems requires some of the best minds from around the world, so I really hope that Chinese AI researchers will make great contributions here.

11:53BAAI Chair Zhang Hongjiang told WIRE that a number of Chinese scientists, including the director of the BAAI, had signed that pause letter a few months ago, but also pointed out that BAAI had a record of being focused on more immediate AI risks as well. Now, Sam Altman was far from the only AI guru to be at this event. Jeffrey Hinton and Max Tegmark, both who have been sounding the alarm on AI risk for some time, were there, as was Jan LeCun. Now, Jan is a fellow Turing Award winner with Jeffrey Hinton and current head of AI at Meta, but who has a very, very different perspective on AI risk than some of his fellow award winners.

12:25This Wired author writes, wherever you stand on the doomsday debate, there's something nice about the U.S. and China sharing views on AI. The usual rhetoric revolves around the nation's battle to dominate development of the technology, and it can seem as if AI has become hopelessly wrapped up in politics. Given that AI will be crucial to economic growth and strategic advantage, international competition is unsurprising, but no one benefits from developing the technology unsafely, and AI's rising power will require some level of cooperation between the U.S., China, and other global powers. Perhaps more interesting than just some shared ideas at a conference in Beijing, however, is that apparently Bill Gates recently met with Chinese President Xi Jinping, and AI was one of the topics they discussed.

13:04According to Reuters, Chinese President Xi Jinping discussed the global rise of artificial intelligence with Bill Gates on Friday and said he welcomed U.S. firms, including Microsoft, bringing their AI tech to China. Now, Reuters claims two sources close to the talks, but interestingly, these comments were not disclosed in reports of the meeting that were published by official Chinese state media or in a post by Gates about his trip. It's not exactly clear to me about what that says about the Chinese position on these questions. The Wall Street Journal also published a piece today called U.S. Grapples with Potential Threats from Chinese AI.

13:34Defining risky artificial intelligence poses challenge as Washington moves to curb investment in Chinese tech. The specific context for this is the forthcoming executive order that will potentially restrict U.S. investment into geopolitical rivals like China. As the WSJ puts it, Washington's efforts to preserve American technological superiority over China, including by banning the export of some advanced semiconductors last year, has been a defining issue in its relationship with Beijing. Now, it appears, based on this article, that the tricky line that the Biden administration is struggling with is how to identify which technology and capital controls relate to national security threats, while at the same time maintaining the goal of not undermining broader commercial trade between China and the U.S.

14:15As the article puts it, quote,

14:26Now, one thing that the article does note is that even before they get into further restrictions on American investment in Chinese companies, the ban on semiconductors is already impacting AI companies in China from accessing computing power that could be necessary to develop more sophisticated models. As they put it, the export ban has pushed some Chinese AI companies to try to develop advanced AI without cutting-edge chips. But at least as of now, it doesn't seem to be enough for the Biden administration when it comes to the potential for Chinese AI development. Now, this is a topic that's going to do nothing but increase in the public discourse.

14:58The most recent Foreign Affairs has a piece about it called The Illusion of China's AI Prowess. Regulating AI will not set America back in the technology race. And any Google search, in fact just a good old-fashioned one not even using BARD, will show you just how many people are writing and thinking about this. I think that the discourse about China when it comes to AI is important. The China-US geostrategic battle is one of the most formative forces in the world right now, and AI will be one of the battlegrounds on which that conflict plays out. But at the same time, it's hard not to see it as at least a little distracting and in fact obscuring of some of the other challenges.

15:34We forget, maybe, that for a big part of the last 30 years, especially before the China-U.S. great power competition became front and center, the biggest thing that social scientists and political scientists were focused on was the rise not of state actor competition to the U.S., but of non-state actors. Now, the Al-Qaeda's and ISIS's of the world may have receded in the public imagination, but it doesn't mean that those threats are gone. Should the U.S. be exclusively dictating its foreign policy around AI in the context of China? Or might it also make sense to think once again about rogue individuals and the increased capacity that these new technologies give them as well?

16:07Anyways, there is a lot here to think about, and so it will be a topic that I'm sure we return to over and over again. For now, that is it for today's AI Breakdown. If you are enjoying this, if you are finding them instructive, please like, subscribe, and share. Click the notification button so you don't miss an episode. Subscribe to the podcast or the newsletter version. You can find all the information you need at breakdown.network. Until next time, guys. Peace.

16:31Thank you.