You probably don't think about it when you ask an AI to whip up a quick dinner recipe or publish an email.
But every time you do, you're using them hundreds and even thousands of times.
We call them tokens.
And here's the staggering part.
By March, China had recorded over 140 trillion token costs per day.
That's more than a thousand times the number from just two years ago and up over 40 percent since the end of last year.
What's driving this explosion?
How is China, now a global AI powerhouse, weaving artificial intelligence into the fabric of factories, hospitals and even our homes?
And as the country's economy goes smart, what does that shift mean for the rest of the world?
Hello and welcome to the panel discussion of World Today.
I'm Dou Hongyu in Beijing.
To delve deeper into China's AI token surge and more.
We're joined by Assistant Professor of Economics Li Lun at Peking University, Professor Andy Mock with Beijing Foreign Studies University and Professor Liu Baocheng, Director of the Center for International Business Ethics and the University of International Business and Economics.
Thank you, professors, for joining us in the show.
First of all, let's start with the very, very basic.
Let's start with you, Andy.
Could you please explain in the simplest way possible what exactly is a token?
Sure, absolutely.
Happy to do that.
And I think that's actually really, really important to understand the what makes large language models so powerful, but also maybe so alarming as well.
So the place to start, as you mentioned, is with a token.
And a token is really just a basic unit of language.
So it's a building block, maybe like a Lego block we can think of.
But the issue here is that it's not just one word.
So for example, unbelievable. would actually be three tokens, un, believed, and a bull.
And even punctuation like periods and question marks have tokens.
So what this means is that when we're talking with an AI, when it's producing content, it actually isn't seeing meaning or processing meaning.
It's really just running a series of tokens.
Now, a token actually is just a long list of numbers.
And here's where it gets really pretty incredible.
So humans can think about a word with maybe three, four, five, maybe 10 different attributes.
So we can think a person has height, they have weight, they have a tone of voice, they have a color of their eyes.
But how, uh these llms work is?
They have thousands, sometimes maybe millions, of these what are called dimensions, so they can see things about language that humans have no possibility of even understanding.
And this is what allows uh, large language models to not only guess the next word in a sentence, but actually find deeper meaning, to construct paragraphs, et cetera.
And we sometimes will call this an emergent process, meaning that this is not something that the humans that developed the AI planned for it to have, but just because it's processing all these numbers, it discovers things by itself.
That gives it capabilities that no one thought possible.
And then what is the deeper technological infrastructure behind all the scenes?
For example, when I ask AI to maybe publish an essay for me, What kind of computing power or maybe chips or server support is actually required behind it?
Well, that also is a great question.
And most people think about, you know, it's chips.
There's a large language model and then there's an application like DeepSeq or ChatGPT.
But in fact it goes far, far deeper than that, that without electricity we cannot have an AI and LLM function.
So this is really where it starts.
And what this means is that when we talk about tokens, it's not just the ability to produce a lot of tokens, but it's the ability to produce the cheapest possible tokens.
And here we measure it by energy.
So we say joules per token.
So joule is just a unit. of energy measurement.
And whoever can produce the most amount of tokens using the least power or electricity is the one that will ultimately win.
So it involves power generation.
It involves the data center.
It involves the connections between the servers.
It includes the chips, which you mentioned.
It includes the, the language models that run on top of the chips or the servers, and then it's the application.
So it's really very long value chains.
Now in China, we've been witnessing a token surge.
China's average daily token calls have exceeded 140 trillion by this month, which represents a surge of over a thousand times compared with two years ago and with an increase of over 40 compared with just a quarter before.
And China's weekly AI token calls have surpassed those of the United States.
And here's a question for all of you professors.
What does this tell us about the current state of supply and demand in China's AI services market?
Professor Li, could you please start?
Yeah, so I think there's multiple reasons under the surge of China's average daily token cost.
So I think to summarize, there's an intensive margin versus an extensive margin.
So basically, what I mean by that is, First, we have more models to begin with.
Right now we have multiple, even dozens, of viable models that can compete with ChatGPT and Cloud, for example Dobao DeepSeq, QN and so on and so forth.
Because we have more models to begin with, naturally we'll have more calls for for these tokens, and also in terms of intensive margin.
So, you know, users are becoming more and more dependent on these models.
So, for example, a user may previously only ask one questions per day to these large language models, but now they could be, you know, asking questions all day long.
And not to mention, for example, the ai agents.
That's a, you know, a new phenomenon.
That's basically constantly calling for the AI large language models to participate in helping the users to solve problems.
So I think these phenomenons together kind of helps to explain why there's a sudden surge in tokens.
But that's just my take.
Okay, now AI has become a mentor for a lot of people around us.
Then let's go to Andy, what do you think?
How do you see this high number of token costs in China?
Well, I think that first of all, you know, we think about cost, it's quantity times price.
And I would want to focus a little bit more on the quantity here because, First of all, there's a couple of things worth pointing out that to me, you know are very important to understand what's going on in China.
So the first is attitudes towards AI.
So surveys in the United States show that many more people are anxious or uncomfortable with AI, whereas in China it's the exact opposite people are generally very optimistic and very eager to use AI.
So I think that is one important driver of demand.
I think.
Second then is the speed at which the market moves in China.
So we see, you know, things appearing seemingly overnight because the market is so competitive uh, the infrastructure, generally speaking is is so efficient that new products, new services can be rolled out uh far more quickly.
And then we look at also the major uh tech companies like uh tencent alibaba, etc.
Have been very, very proactive in integrating AI into their offerings and offering standalone models.
I think this is the second very, very important factor.
The third and last one I would say, which goes back to Professor Lee's point, is something called Jevons Paradox, which is an economic phenomenon where, when you make something more efficient, instead of people using less of it, they actually use more.
Because what happens is it drives the cost down, essentially.
So before maybe to use AI to help you, I don't know, make a video, was too expensive.
But now, because it's become so much more efficient, people now use it for making videos, making songs, not only for research.
So, because of Jevons paradox, I think what we're seeing is again, the quantity of usage really has exploded.
In fact, some would describe it as an exponential increase.
And it's likely to continue, precisely because of Jevon's paradox, because it becomes more and more efficient.
It becomes rational to use it for more and more things.
Great.
Professor Liu, what do you think of the token surge in China right now?
I think both of the professors have mentioned a good deal about those important drivers.
The other is really the Chinese lifestyle, where consumers really see the reliability and the convenience the AI can provide.
And therefore they would file more of the queries into it.
When they look at a flower which they do not really recognize, they take a picture and ask AI what is the name of the flower and where is its origin, et cetera.
And also that in China, we have roughly 12 million college graduates.
So therefore, on the campus, we accommodate more than 5 million college students.
So they are also the most active users on a regular basis, although we do not really allow them to use AI to do their essay.
They constantly refer to it.
And the other is that the competition between those providers are getting so intensified.
So they try to push and promote the popular usage of their AI model.
And of course, there is also continuous improvement to make it more friendly and more deep thinking.
So that's quite helpful.
It's really the interaction of the provider versus the consumer that really boosts such sort of a surge.
So AI is part of our lifestyle now.
And then, Professor Li, why is there a surge of token costs in recent couple of years?
Is it because of better models or larger data clusters or other factors?
Yeah, so I have, for example, for the short-term observation that Chinese models are topping, are surpassing US models in terms of the token usage.
So I actually checked the website with the ranking that actually I think the model that's ranked number one is a model by Xiaomi.
It's called Mimo V2 Pro.
So it's basically officially introduced in March of 18th.
So a little more than a week ago.
Its previous name was Hunter Alpha.
So basically we're seeing that, you know, Mimo V2 Pro has been using over 3 trillion tokens per week, which is a significant number.
So I think I did a little bit of more research and you know, on the website called Open Router, basically it ranks.
You know the usage of different large language models.
And the reason MIMO actually have, you know, the number one position.
I think it's largely because of its, you know usage for being optimized for agents, because you know Open Router, and also MIMO says on the official webpage that it's designed to be the brain of agent systems.
Because, first of all, it has a so-called 1 million token context window so it can go back to the conversation with a maximum length of 1 million tokens.
And also in terms of the pricing model.
It's also fairly competitive, fairly cheap, compared with, for example, cloud or OpenAI.
And finally, there's other evidence that suggests that the usage of AI agents are actually driving this, because MIMO actually keeps track of how much of those tokens are from the user's prompt, how much of that tokens are from completion and reasoning.
So the vast majority of MIMO's daily token usage is from the prompt, so basically from the input.
So we're seeing that the users are actually putting a lot of prompt into MIMO and using up a lot of the so-called input tokens, and that's basically the majority of the token usage.
So that's the short-term answer of why we're seeing, for example, Chinese models, especially for the last week or the last couple of weeks, having this rising trend.
But on a longer term, I do think our guests have mentioned a couple of reasons.
For example, we have a lot of different models.
We have lower prices, we have strong infrastructure and also in terms of the AI computing centers.
China has built over 40 intelligent computing clusters and each of them have over 10000 GPUs.
So basically, we have the capacity, we have the electricity, we have the market demand for these large language models.
And right now we have something new, which is the AI agent, which is basically giving the large language models even more scenarios for its token to be used.
Thank you for that explanation.
And Chinese AI models seem to be not only popular domestically, but also in other countries.
For example, it is estimated that over 90 of academics in the United States use Chinese models in their research.
And in fact, the top four most called AI models globally are all Chinese.
So Andy please, why have Chinese models become so widely adopted, even in what we see as a leading AI powerhouse like the United States?
That really is quite remarkable, isn't it?
And I think that it shows the philosophical difference in how China and the U.S. think about AI.
So for the United States, the development of the modern industry is largely by private companies looking to set up essentially monopolies as their goal and extract economic rent from people using their model.
So basically very, very much a profit driven approach.
However, China has taken the approach of AI as a global public good.
What this means is that this is why we see, I think, not only a lot of models, but almost all of them are open source, which means that you have access to the weights, you can change, adapt them.
But also they are essentially free, or at least if we look at the total cost sometimes only 10 percent of what a comparable US alternative might be.
So I think this is also what is driving a lot of the usage and the token consumption, although Russell Lee makes an excellent point.
We see this enormous surge with this age and craze that we're in the middle of right now.
But I do think in a long term, secular basis, these Chinese models have an enormous cost advantage because again, the Chinese approach is to treat this as a global public good.
And what's even more interesting is that the dirty little secret is even in Silicon Valley.
Many American startups are using Chinese models.
However, they're very careful to not disclose that or to keep it as quiet as possible.
But in fact this again shows, I think, the powerful competitive advantage that Chinese AI models have globally.
But now there seems to be a quite contradictory phenomena.
Professor Liu, I want to ask you this because many Americans have expressed concerns about data security when using Chinese AI services.
But on the other hand, there's a very strong adoption of Chinese AI models.
How do we explain this?
Are those worries fading?
Actually, this is really the concern for many countries, both at the government level, business level and also individual level.
And this is a fair consideration.
But people are rational decision makers, so therefore they calculate the cost and the risk, both in terms of the technical risk and also in terms of the regulatory risk.
Therefore, the Chinese model is so much attractive because, as Jason Huang of Invita has mentioned, it becomes a commodity and China is so competitive in the provision of all commodities, from tangible ones now into intangible ones.
As also Andy mentioned, most of the users, both at the corporate and also individual level.
They are private operations so therefore the do not have to attach themselves to the policy directions as long as it is not really a regulatory requirement by the White House or many other legislators in the United States.
So therefore, they would definitely pick on that.
And the other is that the rise of the open source ecosystems have also lowered the barriers for such sort of adoption.
And researchers, developers, on a worldwide basis, they can experiment and build on these models.
They can also compare the cost and benefit of many other models that are available.
So in the AI era, users are really increasingly pragmatic.
So they choose tools based on the effectiveness and the efficiency and convenience.
And so, rather than the origin or simply to follow the political orientation, because that's not really very reliable and it may change over time.
And now the world's giant AI players are very excited about this token surge.
As Professor Liu mentioned, NVIDIA CEO Jensen Huang called tokens the new commodity.
And he also talked up the new concept of tokenomics, if I pronounced it correctly.
Professor Li, what does this token economics or tokenomics really mean for businesses and the broader economy?
Yeah, so from my perspective, I think the so-called tokenomics is treating token as a commodity money or as a currency.
But I mean at the deeper level.
I think it represents a new ways of production, meaning that when we're measuring the output from AI, when we're measuring the output from the smart economy, from the future production, we are measuring them in the number of tokens you know, just as we're measuring, you know, electricity in watts.
Or we're measuring the cloud services in the storage of gigabytes, or we're measuring the internet service in the bandwidth.
So what Jensen Huang means, I think, is that you know, in terms of the future, you know operational, you know approach tokens are going to become a new kind of almost like a currency.
So, in terms of for firm, for its budgeting, for its productivity, for talking about its revenue generation, firms need to take into careful consideration of how much token they need to consume for the next quarter, how much tokens are going to cost, and that's going to change basically, the mindset of these firms, right?
So, instead of asking whether we should use AI or not, like the previous generation, I think the answer or the right question have increasingly become which AI model or what workflow is going to give us the highest return per token?
So for every token that I spend, how can I generate the most revenue?
So, in other words, we're not talking about ai, you know, like a chat bot or like a standalone software tool.
It's becoming more like a workflow or the entire way to organize the, you know, the production process.
And, at the more macro level, i do think this also means that there's going to be more intensive competition.
Uh is going to center on, you know, the cost per token and also in terms of the revenue per token.
So the cost per token, As you know, Professor Annie Mock mentioned, you know,
It's the electricity, the energy you consume basically to generate one token.
And also the revenue per token is how much dollar value or how much RMB value can that token generate for the firm?
So both of these efficiencies are going to be crucial for our futures workflow because they matter significantly for the productivity
There's a McKinsey estimate where it says generative AI could add 01 to 06 percentage points to the annual labor productivity growth all the way through 2040.
I do think, you know, with more innovations, that's probably a understatement.
But anyway, I do think that you know if we're talking about broader use of AI, you know in a society, you know framework.
I do think you know it's going to generate a real productivity gain and it's all going to be measured in the unit of tokens.
And now we're probably seeing the tokenomics taking shape in China.
We can see the push for the integration of AI into economy in China's official documents over the years.
And first, the government work reports introduced the AI Plus initiative.
For example, AI Plus manufacturing, education, and finance, et cetera, et cetera.
And last year, the focus continued.
And this year, we're hearing a new term, the smart economy.
So Andy, how are AI plus and the smart economy related?
Is smart economy the next stage?
Absolutely is.
And I think you know again, we have to step back and look at the distinctive Chinese approach to governance, where uh China can articulate a long-term vision.
Or in my recently published book The Innovation Machine, how China creates and adopts technology through governance.
That what China does is first articulate a vision based on a worldview, a philosophy, translates that step-by-step into ultimate impact.
And what we see here is the philosophy or the idea is high quality growth through new productive forces, right.
And this is exactly what China is doing to offset the demographic headwinds of an aging population, a global environment that is very challenging when we think about the free flow of technologies.
So what this AI plus?
The goal is to increase productivity more to offset these headwinds, these structural headwinds that China is facing.
And how it's going to do.
That is through deploying AI, not as an industry vertical, but we can think of it horizontally, as a substrate through the entire economy.
So touching every industry, every facet of human activity.
And I think that it is the Chinese approach where they can actually take an idea, turn it into policies instruments, infrastructure and ultimately deliver impact.
All right.
Thank you all for your in-depth analysis.
Let's have a short break.
Coming back, we'll continue our discussion.
Hello, my name is Alessandro Golombievski Teixeira.
I'm a professor of public policy and management at Tsinghua University in Beijing.
I am a great listener of The Wall Today.
In my opinion, The Wall Today is one of the best China radio programs.
In The Wall Today, we can get the best news and analysis in what is happening now in the world.
So please, come to join us.
Welcome back.
You're listening to the panel discussion of World Today with me, Dou Hongyu.
We've been unraveling China's AI surge and exploring what it means to China's economy.
Let's continue our discussion.
We've already heard of another term and we're very familiar with it.
It's already a very natural part of our life when you shop online, when you order food online, or maybe when you hail a taxi on an app.
It's already digital economy.
It's part of our life already.
But how is the smart economy different?
Professor Liu, what's fundamentally new here?
Well, the digital economy is primarily about connecting information by digitizing the process to enable communication and building platforms.
But the smart economy goes further by enabling systems to understand analyze, to deep search and act on those information.
So it's more dynamic and more user-friendly.
In other words, the shift is really from a data-driven to intelligence-driven economy.
So the key difference is that digital systems store and transmit knowledge information, where the smart systems can generate insights and help to make decisions and provide guidance.
So this marks a transition from connectivity now to co-creativity and also to more of the knowledge sharing with people.
This is really the major difference.
Yeah, from data driven to intelligence driven.
And I believe China is not the only country who wants its economy to be driven by intelligence.
Professor Li, what are China's unique advantages in building this smart economy?
Yeah, so I think China does have a lot of advantages in making this transition.
First of all, we're already talking about a very large internet usage base.
So over 1.1 billion people are already using the internet.
And of those people, over 600 million people are using generative AI.
So we're talking about a penetration or adoption rate of 40% around the entire population.
So a lot of people are welcoming to the idea of AI models or AI agents, because I think the most important difference between a digital economy and a smart economy, as Professor Liu also elaborated, is in terms of using AI to help us not only to collect the information, but also to help us make decisions to solve our day-to-day problems.
If our user base are generally welcome to the idea, there's going to be little psychological resistance to the adoption of these technologies.
In terms of policy support, we're seeing very clear signs in, for example, the government work report which says ai plus is combining the digital technologies to china's traditional market sectors, including manufacturing.
So basically, we're trying to combine, you know, the most cutting edge technology in ai, also the most traditional technology of manufacturing, and creating new forms of smart economy, making sure that AI is helping to transform these traditional technology sorry, traditional sectors into the next winner in this AI era.
And I do think this is going to have a huge spillover effect because our logistics, our supply chain management can be hugely optimized by AI, because AI can help with routing, warehouse management and also inventory management.
Also, we can talk about AI being integrated into commodities and devices.
For example, when we're talking about our EVs, we can have these AI-enabled devices or AI-enabled cars that could better help the users for navigation, for driving.
And also AI-enabled phones and computers and eyeglasses.
Even toys have emerged as new phenomenon in Yiwu, for example.
So I would say that smart economy is especially timely for China right now, because we're currently looking for the drivers for the new productivity growth.
So I think this is a natural path towards a high-quality growth, because the technology is usable, the infrastructure is built and the economy is actually ready for the wide adoption of the AI technology.
Earlier, Andy said that China is seeing AI as a public good.
So here's a question I think is very important for our global listeners.
And I'd like to invite all of you professors to answer.
What does China's smart economy mean for the world's economic landscape?
Andy, let's start with you.
Sure.
For the world economic landscape, I think it means a couple of things.
I think, first of all, it means a narrowing of the digital divide.
So one thing we've seen, as the world has digitized primarily through the Internet and now the mobile Internet, is a wide gap between the haves and the have nots.
So primarily to develop countries and the global south.
So I think what China is doing again by treating AI as a global public good, making it accessible from a cost perspective.
But also the other powerful feature of open source models, as I mentioned earlier, is that the weights are actually available and you can run the model, adapt the model to your own particular circumstances.
So this also has security implications as well.
So if you say, for example, I'm not comfortable having my data hosted in another country, you could certainly then run it in your own country, in your own city, in your own company.
So I think this combination of cost and adaptability is a very, very compelling feature.
I think the other is that there are i would call them maybe economies of scale, in that what i mean is that the more people you have using a particular model, a particular standard, the more likely it is you'll have better applications, faster problem solving, etc.
So sometimes we call this a network effect, right?
So if you only have Two people using a telephone, not that useful.
If you have 50 million people all using a phone network that can talk to each other incredibly powerful.
So I think this is one dimension to this as well.
And again, as we have more and more people experimenting, solving specific problems using AI, I think we can all benefit.
So I think this is another way.
And then frankly, I would say, another aspect is less obvious.
But in addition to providing these tangible things like a technology foundation, the application layer, what China also is doing, I think, is providing a stable, predictable and safe political environment, or companies that decide to invest in China and with China can feel a greater sense of security than if they were working with other parts of the world.
So I think these are some of the more important ways that I think China is really driving, I would say, systemic change and improvements in the global landscape.
So the stability and predictability are also unique advantages of China's AI development.
And then, Professor Li, what do you think?
What does China's smart economy mean for the world's economy?
Yeah, so I think, besides what Andy has said previously, I think There's going to be some ripple effects, both from the most central sectors related to AI to the more peripheral or the more traditional sectors.
And that's the first meaning of the ripple effect.
Also, it's going to have a ripple effect through China's trade network.
So it's going to benefit China first, but it's going also to benefit China's largest trading partners, including Africa.
ASEAN countries and so on and so forth.
So what do I mean by that?
For example, when we develop the smart economy, the most central sectors, including the tech companies that make the AI models, including the companies that's making the robotics, for example, for these human robots, and everything that's you know being central at the AI sector, that's going to directly benefit not just the central sectors but also the sectors, like I mentioned manufacturing, like you know healthcare, like service sector.
So these are going to, with the help of AI, help us to generate, for example, new products or, you know, new scenarios for usage of AI, and also at lower costs.
So, for example, you can right now make AI toys at very affordable prices.
So if you're a parent, you know, at a South, you know at an Asian country, you can import one of these AI toys without having to, you know, make them domestically.
And you can also enjoy the benefit, the surplus from purchasing that good.
Okay.
And finally, Professor Liu, what do you think?
AI, since it's based on the internet exchange, so therefore it can really have a very positive spillover around the world, as long as you are more creative and and give you one example.
Is that uh, actually one of my co-author from the united states, who are the professor, and came to my house and he really admires the robot running around our floor to clean it and listen to order, et cetera.
And then my student also gave him a two-case toy, which can be AI
But then I think that also shows that how China can really continually to improve on.
Its advantage is that they only follow the Chinese direction.
So those kids are really being pushed to speak to the right Chinese to be able to direct the toy robots, although not very successfully, I think now it is really something that's spreading very quickly.
I think the strongest advantage for China is in dealing with any type of new gadgets or technologies.
They are able to produce it on a large and massive scale.
So particularly for innovation, so scale does really show the best promotions.
So only when people have the access and larger population having the access and with big scale they can really popularize very soon.
And not to mention that it is always the Chinese.
Another advantage is cost and benefit that's there to share with the rest of the world, and particularly with the global South.
And they are very happy to engage in such sort of activities co-creation in terms of localization, but the main creator will be more of the Chinese providers, both for the software and also for the hardware.
And now, as the smart economy is taking shape and China's AI development is progressing really quickly, it is noticeable that China's plans emphasize building a solid foundation mainly through three elements, namely computing infrastructure, algorithm innovation and data resources.
So Andy, can you help us understand how these three work together fundamentally?
Sure.
Well, so you know again, if we go back to this idea, that you know we have all this, these exciting capabilities that are available online, and you know we haven't talked about embodied ai yet, or smart robots but basically, without electricity you cannot run any of this, and i think this is something the us is now encountering that its electrical grid is not able to support uh, the demand for ai and, of course, it is having a pricing impact in terms of making electricity more expensive.
Now, What China has done, I think not only in providing this foundational layer of infrastructure for China, but I think probably globally as well, is that China has for years, if not decades, invested in electrical generation and transmission capacity.
So, I'll give you just a quick statistic.
Last year, China produced 10 trillion kilowatts of electricity The U.S. produced 4 trillion.
And the number three country, I believe, was India, produced 2 trillion.
So, again, without sufficient electricity, you really can't run any of these models.
And China not only is producing a lot of electricity, it's producing it in a green, sustainable way.
And it is also producing at a very globally competitive cost.
So I think that's the foundation.
And again, we've talked about the data center processing layer.
I think China again, because of its governance system, is able to coordinate the infrastructure, the water, the connection to the electrical grid, the building permits, et cetera, in a very smooth and efficient way to allow the rapid build-out of data centers.
And in the past, what we've seen, as Professor Leal mentioned, is this bottleneck with high-end GPUs that China really could've used more of them.
But at the same time, I think China's made enormous advances in building around these constraints.
For example, what the Huawei 384 Ascend right is basically hundreds of these chips that can outperform the comparable Nvidia solution, but it uses more chips.
So each chip is not as capable, but the unit itself is actually has greater processing power than the Nvidia comparison or alternative.
So I think again we see this happening right before our eyes the tremendous innovation in the models and I think also the applications.
So China, I think, is not building this, not just for domestic use but, I think, also at a scale that has global implications as well.
And then let's talk about data.
Professor Liu, China is building high quality data clusters for various fields, including transportation, education and healthcare.
But what does this high quality data mean in this context?
How is it different from the data we currently collect?
Yes, I think that's a very important question because the AI is engaged in deep learning.
So they have to rely on not only the volume, but a rather clean structured accurate, relevant data to specific domains.
For example, you know, we are talking about the AI-aided diagnose over health issues.
So the medical data for such diagnosis must be very accurate and very well labeled.
This is unlike the general Internet data.
People can really talk about stories, writing the poems and also spread some gossips, which may be noisy and inconsistent.
So high quality data will have to satisfy these criterias.
So the shift towards the high quality data by Chinese policymakers reflect a move from quantity now to usability and scalability, which is really essential for building a reliable AI system that is there to serve positive means of all users.
And Professor Li.
As there are more users of AI, as there are more interactions between human beings and artificial intelligence, and as there are more data, what are the biggest challenges China faces in AI governance, which can be quite pressing?
Yeah, so I think the most significant challenge is to balance the technological growth and safety or privacy concerns.
So we have seen for example, recently regarding the AI agent Open Claw there was some debate about Open Claw actually misuses some of the user's private files and actually in some cases, cause damage financially to these users.
Basically, I mean including the traditional.
You know challenges like false information like, for example, personal infringement in terms of deep fake or using AI to generate fraud or harmful contents.
You know data security problems or discriminative or biased content.
So those are all you know combined, you know, As part of the challenge between how to balance between innovation and safety concerns.
So I think there needs to be some filtering, some regulation in terms of making sure that these contents are managed in a proper and responsible way.
Also, whoever posted these contents, especially on social media.
There need to be identity verification.
There needs to be clear labeling for AI-generated contents, where we need to have the users to have a deeper understanding of what these models are really capable of, especially when it comes to more powerful models like agents or open claw.
You need to have the agent you know, have the users understand what authorities or what permissions you're essentially giving these, these agents.
And I think in terms of accountability there's also some challenge, because if we have an AI model or AI agent makes a harmful decision, for example, a harmful suggestion to some of his users, then who is responsible for that answer?
Are we taking the model provider accountable or not? is it the platform is it the user is it whoever deploys the model so i think there's uh you know needs to be more rules that's uh basically going to give more clear clear guidance about the accountability in these cases and finally i think in terms of authentic authenticity and trust because there are going to be, as I mentioned, so many contents that being generated and altered.
So in the future, you know, there could be this vicious cycle of people using fake information or fake videos to generate, you know, even more, or faker, or even even more outrageous materials.
So, you know, because I can also use some of the AI generated material for its future training.
So if we're talking about a deep contamination of our knowledge pool, then the AI models could learn from the mistaken or biased materials and generates things that's even more biased or in a discriminative.
So how to regulate those things and prevent, you know, those mistakes from happening is also a challenge that we need to face.
Indeed, a lot of problems to consider and, professor liu, how can these governance challenges be addressed better, including the issues of accountability and trust?
To address these challenges require a combination of regulation education, technical solutions and also collaboration.
First, The government needs a clear and adaptive policy where they can really support innovation but also to prevent misuse, with hard thoughts and clear type of guidance and rules.
Education is there to educate consumers with the right type of ethical standard so that they can really prevent political and moral bias.
So, like you know, how we can really train our students in how to use AI is that you cannot really engage in plagiarism with simply essay written by AI software.
That also is very much challenging to teachers as how to check out some of the plagiaristic innovation by our students.
And then the company's business community can also work together to collaborate.
One is that they are able to prevent the monopoly to have expected rent on consumers.
And the other is that they will give the true information as people need it.
And third, that they need to refrain from, you know, pushing those viewers and consumers hard with the junk type of information.
So right now this is really a problem that is largely complained by so many people is that they do not want that content, but as long as the just by accident hit on a certain type of a topic and then they get flooded by the same type of information or even advertisement.
So, and lastly, I think you know how can China work together with the rest of the world, as you know how to manage the right type of data flow so that we together can work out certain type of norms to prevent risk.
So this really something that we can really need to look into and address it effectively.
And finally, as we embrace the AI revolution and, in the meantime, being aware of all these challenges professors, could you please give every one of our listeners a some suggestions?
I mean, as AI becomes deeply integrated into economic activities, our life will be changed.
But what mindset or skills should we individual users develop to make a better life and to thrive in the era of the smart economy?
Let's start with Andy.
Sure.
And I think this is an incredibly important question, which I have been spending a lot of time thinking about and engaging with different sectors of society in about this, because we can, I think, pretty clearly see a future where we have AI that is more capable than the most capable human.
And what does that mean for our jobs?
What does it mean even for what it means to be human?
And here I'm teaching a class on this, in fact, and using Man's Search for Meaning, which is a book by Viktor Frankl, who survived the Nazi concentration camps.
And he wrote that even in such terrible conditions one can still find meaning and even a sense of contentment.
In such terrible circumstances.
And the idea is that When everything is taken away from you, you still have your dignity.
And that is one of the most important things.
And it is up to us individually to choose what is meaningful.
And that perhaps, is the most important thing to keep in mind as we move seemingly very quickly to a world where humans may be displaced.
It's quite a philosophical question to think about.
And let's move on to professor lee.
Do you have any other suggestions?
Yeah, so i totally agree with what andy said and also i'd like to add that uh, you know, ai literacy is probably going to be one of the most important skills in this new century, because we need to have to be able to use AI not as our substitutes, but as our complements.
We need to use AI as our helper to help us accomplish the skills or accomplish the achievements that we, by ourselves, couldn't achieve before.
So I think the key is to focus on those skills that's complementary to AI.
For example, how do we write better prompts to AI?
How do we check the outputs from AI and see if they're correct?
How do we understand, for example, in which context we should use which AI and how to make these higher level judgments or decisions and leave AI to do the lower level decisions?
So the AI literacy knowing when and how to use this AI, I think is going to be the most important a skill that anyone can maneuver in the future.
And also, I think, what makes us human, because before the Industrial Revolution, humans can use their muscle or manpower to earn a salary.
And after the Industrial Revolution, machines replaced that function of humanity and we use our brains to kind of, for example, intellectually or cognitive skills, to earn our salary.
But AI is quickly going to replace that as our scarce resource.
So what are we going to do to strengthen our edge as a human is not just producing or thinking about things or ideas, but more of developing a taste, developing a responsible and ethical way to communicate and work with AI, and also to think across borders, to think across different fields and domains.
Those are the skills that AI can help us to amplify but cannot really replace.
Great.
When we learn about how to work and chat with AI, it is also helping us to know who we are.
And last but not least, Professor Liu, what suggestions would you like to offer us?
I think, first to have a very open mind to adopt such sort of innovation, because that can really enhance productivity and creativity.
Second, be careful with the correct usage, because critical thinking is utterly important and people are not only prepared with skills, but they should also be prepared with their taste, with their own sense of judgment, before they can really fully rely on the AI support.
And so definitely, this is really a transformative power and those who will miss it will have to be social laggards.
So therefore, open mind, critical thinking and also the preparedness in the entire ideology and the knowledge work, that's something important to fully take advantage of such a transformative power.
Okay, thank you all professors for your wonderful and thought-provoking suggestions.
You leave us a lot to think about after the show.
Thank you.
Assistant Professor of Economics Li Lun and Peking University, Professor Andy Mock with Beijing Foreign Studies University and Professor Liu Baocheng, Director of the Center for International Business Ethics at the University of International Business and Economics.
And that is all for this edition of World Today.
To listen to this episode again or to catch up on our previous episodes.
You can download our podcast by searching World Today or you can follow our ex-accountant, CDTN Radio.
Thank you so much for listening.
Bye for now.
