Summary of Key Points
In the policy on intelligent agents issued by Beijing, what is most noteworthy is not the subsidies such as "computing power vouchers" or "Token vouchers," but rather the participation of the Municipal Bureau of Statistics in the responsibility for "encouraging the development of the Token economy." The logic behind this is that the intelligent economy has grown to the point where it requires its own "measurement tools"—a statistical system. Tokens are currently the closest candidate to serving as a "unit of measurement" for the intelligent economy. However, we must avoid merely focusing on increasing the volume of usage and shift towards charging based on value. Ultimately, this will enable effective governance of the intelligent economy by establishing a credible measurement infrastructure.
1. Why Does the Bureau of Statistics Play a Role in AI Policies? The Intelligent Economy Needs Its Own "Dashboard"
Every era has its own way of measuring the economy: in the industrial age, we looked at electricity production and steel output; in the mobile internet era, we used metrics like DAU (daily active users) and GMV (gross merchandise value). The intelligent economy also needs such a measure. Otherwise, like during the Great Depression in the United States, governments would have no idea how bad the economic situation was, and policies would be based on guesswork.
Previously, the measurement of the intelligent economy relied on private indicators, such as NVIDIA's financial reports (on the supply side) or Microsoft's capital expenditures (on model development). However, these only reflected the production side and did not account for whether the demand side was actually benefiting from the technology. By involving the Bureau of Statistics, Beijing aims to create an official "dashboard" for the intelligent economy. This dashboard will help quantify data on the volume, quality, and efficiency of intelligent operations, allowing the intelligent economy to transform from a vague concept into a tangible goal that can be managed, just like GDP.
2. Why Can Tokens Serve as the "Measurement Unit" for the Intelligent Economy?
Tokens are the basic units used to invoke intelligent models—whether you use Baidu Wenxin or Alibaba Tongyi, each model invocation is charged based on the number of Tokens used. Like "kilowatt-hours" in electricity measurement, Tokens have three key characteristics:
1. Universality: They can be applied across all models, just as kilowatt-hours are applicable to all power plants.
2. Measurability: Every invocation is recorded, preventing any confusion in accounting.
3. Market Value: Global model manufacturers offer real-time pricing; for example, GPT-4 costs $0.03 per thousand Tokens.
Without a unified measurement unit, it would be impossible to price, subsidize, or tax intelligent services. Before the establishment of a standard for Tokens, it was unclear how much an intelligent service was worth. By early 2024, the daily Token usage in China had reached 100 billion, and by March, it had surpassed 140 trillion, representing a 1,000-fold increase in just two years. The National Data Administration's description of Tokens as the "value anchor" of the intelligent era is entirely justified.
3. Concerns About Manipulating Tokens to Obtain Subsidies
The lessons from fraud in new energy vehicle subsidies are still fresh in our minds. If subsidies were based solely on the number of Tokens used, companies might engage in unnecessary operations (e.g., having intelligent agents repeat meaningless statements to generate more Tokens). Beijing's policy addresses this by prioritizing quality over quantity in evaluation criteria, focusing on the accuracy of responses and the effectiveness of the services provided by intelligent agents in helping businesses make money. It also encourages charging based on value, rather than simply on the amount of Tokens used. For instance, an overseas customer service intelligent agent like Fin is charged $0.99 for each successfully resolved issue; Salesforce's acquisition of Fin (for $3.6 billion) reflects this shift towards outcome-based pricing.
4. Where Does the Profit from the Intelligent Economy Come From?
The pricing model for the intelligent economy is evolving:
- Initially, fees were based on the number of service accounts purchased (whether used or not).
- Now, fees are calculated based on the number of Tokens used.
- In the future, fees will be based on the results achieved—e.g., a percentage of the revenue generated by the intelligent services.
By including RaaS (Software as a Service) in its policy documents, Beijing makes it clear that the value of the intelligent economy lies not in the models themselves but in their ability to solve real problems for users. For example, if an intelligent agent helps a factory save 10% in costs, that result is far more valuable than simply using millions of Tokens.
5. The True Ambition of the Policy: To Measure the Intelligent Economy First to Govern It Better
The real purpose of this policy is not to distribute subsidies but to establish a measurement infrastructure for the intelligent economy. This is similar to how GDP was developed: growth was not measured until there were tools to quantify it, which then made it a tangible goal to pursue.
Whoever establishes the proper measurement framework for the intelligent economy will gain control over its development—identifying where intelligence is being effectively utilized, where there are inefficiencies, and where subsidies are needed. This is more important than distributing a few vouchers, as it determines the future direction of the intelligent economy.
Whether Tokens are the perfect measurement unit or not is secondary. What matters is that Beijing has recognized that the intelligent economy has grown to a scale where we can no longer afford to develop it in a haphazard manner; we must first measure it accurately to ensure its sustainable growth.
In Conclusion
The intelligent economy is like a new car, and the Bureau of Statistics plays the role of the dashboard designer. Without a dashboard, you wouldn't know the vehicle's speed or fuel consumption, making driving chaotic. Beijing's policy aims to install such a dashboard, enabling us to understand, manage, and develop the intelligent economy in a more controlled and effective manner.