Summary of Key Points
Professor Sun Tianshu from Cheung Kong Graduate School of Business believes that the value of AI does not lie in token consumption (the “fuel” on the supply side), but rather in its ability to be integrated into industrial scenarios to complete measurable tasks and form intelligent agent systems. The greatest opportunity for AI in China is not to replicate the consumer internet model (To C) but to adopt an AI-to-B approach (using AI to transform various industries). This transformation requires “Henry Ford-style” architects who understand both industry and AI—those who can restructure production methods from scratch using fundamental principles, rather than simply adding AI capabilities to existing processes, making intelligent agent systems a core asset for businesses.
1. Tokens are merely “fuel”; true value lies in the application scenarios
Tokens are like gasoline for a car: the amount of fuel you use does not necessarily reflect the value you create; it could be wasted or lead to a significant deal. Professor Sun emphasizes that tokens serve as a measurement unit on the supply side (connecting chips, computing power, and models), but they do not equate to value itself. The same tokens can generate vastly different amounts of value when used in contexts such as drug research (helping scientists quickly screen molecules) versus writing trivial essays.
Business leaders are concerned not with how many tokens are used, but with what outcomes are achieved with those tokens—e.g., whether production efficiency has increased, customer complaints have decreased, or the time to develop new drugs has been shortened. Tokens only gain commercial significance when they are transformed into measurable results in real-world scenarios (for example, AI helping a factory save 10% on raw materials).
2. AI-to-B has more potential than AI-to-C; China’s industrial foundation is an advantage
Why is the AI-to-C approach less promising? Because users’ attention and spending power have already been exhausted by mobile internet usage (7 hours per day, with no room for a significant increase). AI-generated content (novels, short dramas) does not attract more viewers just because it’s created by AI. Most AI-to-C products merely improve supply efficiency without creating new consumer experiences.
However, the situation is different for AI-to-B: China has one of the most comprehensive industrial landscapes in the world, with real-world challenges in various sectors (automotive, pharmaceuticals, manufacturing, etc.). AI-to-B can directly enhance productivity by helping doctors analyze CT scans, optimizing factory processes, and accelerating drug development. These improvements represent tangible value, and China’s extensive industrial data and experience are natural advantages.
3. The core asset of a business is not the model, but the intelligent agent system
Many think that AI simply means purchasing a large model for use, but Professor Sun argues that the future core asset will be intelligent agent systems—closed-loop systems that integrate models, data, processes, and feedback. For instance, an intelligent production line in a “dark factory” can sense production data, automatically adjust processes, and continuously improve efficiency, rather than relying on 500 human robots.
For example, an intelligent system in the livestock industry could link sensor data (body temperature, food intake) with breeding processes and business outcomes (milk production) to automatically optimize feeding plans. Similarly, a retail system could integrate inventory, sales, and replenishment to optimize the supply chain in real-time. These systems can evolve continuously, becoming the “industrial brain” of the business.
4. Industrial restructuring requires “Henry Ford-style” AI architects
Professor Sun frequently mentions Henry Ford, who did not just add electricity to old factories but used it to reinvent the assembly line, turning cars from luxury items into mass-market products. The AI era needs such individuals: those who understand both industries and AI, capable of reorganizing production methods. These could come from within industries (e.g., factory owners collaborating with AI teams) or be cross-disciplinary collaborations. Those who only know technology may not understand the complex dynamics of industries like pharmaceuticals, while those who only understand industry processes may not know how to use AI effectively. Their goal is not to add an app to existing services but to use algorithms to fundamentally transform industries (e.g., like滴滴 reorganizing the transportation sector).
5. Focus on deep restructuring with “AI+”, not just superficial “+AI”
How can you tell if a project is genuine AI or not? Look at whether it changes the production process and business model. For example:
- +AI: Adding an AI chat function to old software might increase margins by 1%, but it doesn’t change the industry structure.
- AI+: Using real-time location data and algorithms to restructure transportation scheduling, as with滴滴, can significantly disrupt the industry (e.g., making taxi companies less competitive).
Professor Sun advises businesses to choose the right scenarios (e.g., areas with shortages of specialized services or high-value decision-making processes, such as drug research) and establish dedicated AI teams with resources for experimentation. If existing businesses are still performing well, it might be better to invest in new AI-based initiatives to avoid the “innovator’s dilemma.”
What is most needed in the second half of the AI era?
It’s not cheaper tokens or better models, but “AI architects” who can integrate industrial scenarios, data, and intelligent agents. Each industry will need its own architect to redesign processes, organizations, and business models with AI. In Professor Sun’s words: “The second half of the AI era requires a thousand Henry Fords.”
(End of translation)