The 2025 tipping point: AI's IQ will surpass that of humans, and economic rules are about to be rewritten.
The average IQ of AI has exceeded 110, officially surpassing that of ordinary humans. In 2025, AI began to participate in the "full - chain operation" of the economic system. From information collection, decision - making to actual execution, for the first time, a non - human entity independently runs the complete economic chain. AI is rewriting the underlying rules of business! Keynes' century - old prediction is finally coming true, and the AI economy is emerging.
In the wave of digitization of human economic activities, the Internet and the mobile Internet have completed the first two steps. The emerging AI economy may bring even greater changes.
Human economic activities are being digitized:
In 1946, humans invented the computer, which marked that after thousands of years of evolution, human calculation finally reached the electronic form from manual to mechanical.
The emergence of the computer has increased computing power to a level far beyond that of the human brain.
In 1874, the British William Shanks spent 15 years calculating pi to 707 decimal places (however, in 1945, it was discovered that Shanks' calculation of pi was incorrect after the 528th decimal place). In 2019, Google Cloud Platform helped humans calculate pi to 31.4 trillion decimal places.
Humans are in the natural environment and have two fundamental tasks:
One is to utilize and transform the natural environment to support human survival;
The other is to improve personal life after achieving material prosperity, so that each person's nature can be fully developed, that is, the all - around development and self - actualization of humans, "be the best version of yourself."
Under the first task, in the interaction with nature, humans have developed some classified methods, such as astronomy, divination, mathematics, engineering, physics, biology, and nature. Among these methods, the most scalable ones later dominated the process of human - nature interaction. Such fields are mathematics, physics based on mathematics, and the later - developed computer science.
The emergence of the computer means that humans have begun to enter the digital age.
From this moment on, all human economic activities have begun to be digitized in sequence. After digitization, algorithms can play a role, and economic activities can be driven by algorithms to achieve intelligence. From this perspective, the digitization of the entire human economic activity seems to be an inevitability.
Chart 1: The digitization process of economic activities
Regarding the digitization process of human activities, Nicholas Negroponte's Being Digital is a milestone - worthy work.
This 1996 work acutely pointed out the above - mentioned trend of the digitization of the entire human economic activity and condensed it into a piece of advice: "Move bits, not atoms." (Being Digital has influenced many people globally. Among the believers of "Move bits, not atoms," many later became influential leaders in the digital economy, such as Wang Xing, the founder of the Chinese company Meituan.)
The efficiency of the bit world is millions of times that of the physical world. Due to the efficiency gap, in the era dominated by computers, the entire physical world will eventually be replicated in the bit world, and economic activities will be carried out in the bit world. This process started at the end of the 20th century and may take a considerable part of the 21st century to complete.
Currently, we are in the first stage of the above - mentioned process where digitization is not yet fully completed and the second stage where algorithm - driven development is booming.
The first stage/digitization is the Internet era and the mobile Internet era. Computers have helped humans digitize daily activities in fixed scenarios, and mobile phones have helped humans digitize daily activities in mobile scenarios. The essential characteristic of this stage is the digitization of the physical world, but thinking and decision - making still need to be done by the human brain, and the biggest role of the digital world is matching, which has greatly improved the matching efficiency.
In the second stage, thinking and decision - making can be done by algorithms, and algorithms can deliver work results. Its starting point is that algorithms have thinking abilities close to humans. In the medium - to - long term, algorithms will have better thinking abilities than humans. We are currently at the critical point in the second stage where algorithms begin to have the ability to deliver work in a generalized manner. The contribution of the second stage to human economic activities will far exceed that of the first stage.
The digital - world economic activities in the first stage
As mentioned above, compared with previous economic activities, the most significant feature of digital - world economic activities in the Internet era and the mobile Internet era is that the matching efficiency has been greatly improved. Through the two types of hardware, desktop PCs and mobile phones, which emerged successively, the economic activities brought about by the mainstream daily needs of humans have all been digitized.
In the new economic forms emerging in the Internet era and the mobile Internet era, the three largest tracks are search, social networking, and e - commerce, corresponding to human information needs, social needs, and commodity needs respectively, and also corresponding to the matching of information and people, people and people, and commodities and people respectively.
Why can the Internet and the mobile Internet greatly improve the matching efficiency in the above three scenarios? We use the following table to illustrate this process.
Chart 2: The matching methods of information, commodities, and social partners in the pre - Internet stage, the Internet stage, and the mobile Internet stage
We can see that the three types of needs, namely information, commodities, and social partners, in the pre - Internet stage, the Internet stage, and the mobile Internet stage, are matched through proximity acquisition, global search, and personalized recommendation respectively.
The choice set for proximity acquisition is very limited, which is also the norm for human choice - making since the birth of humans. For thousands of years before the birth of the Internet, humans made choices in this way.
Compared with proximity acquisition, the choice range and richness of global search have been increased by an order of magnitude. People can make choices in a choice set close to "exhausting all possibilities," and users are more likely to get a choice with relatively high scores in both the "liked" and "suitable" dimensions. Such a choice may be outside the choice set in the proximity acquisition stage.
Compared with global search, personalized recommendation better solves the problem of "inefficient choice due to insufficient knowledge of an individual in a certain field." That is, although users can make choices in a range close to the entire set, since judging each type of choice object requires specialized knowledge, an ordinary user cannot have such a high - level knowledge reserve in every field, so he still cannot always make choices with relatively high scores in both the "liked" and "suitable" dimensions.
Personalized recommendation essentially recommends "the best - verified choice of a certain type of users with commonalities in a certain field" to all such users with commonalities, thereby improving the quality of their choices.
So, the entire Internet/mobile Internet digitizes the economic activities brought about by the mainstream daily needs of humans and then solves the matching problem. Just this alone has brought a huge boost to economic efficiency and consumer utility.
From the perspective of the digitization of the entire human economic activity, the Internet and mobile Internet stages are just the beginning.
First, in terms of the scope of digitization, the economic activities related to individual consumer behavior have a relatively high degree of digitization, while the economic activities related to enterprises still need to improve their digitization degree.
Second, the Internet and the mobile Internet mainly provide great value in the aspect of "matching."
The interaction between humans and nature can be described by the chain of "collecting information - making decisions - taking actions." Among them, the Internet and the mobile Internet have optimized the information - collection link and partially optimized the decision - making link (in the case of global search, decisions are still made by the human brain; in the case of personalized recommendation, the human brain can make decisions by referring to the options recommended by the algorithm).
Logically, after the complete digitization of economic activities, the entire chain of "collecting information - making decisions - taking actions" can be optimized.
Here, we can see that in the entire digital wave, the Internet and the mobile Internet are just a small step for humans.
The emerging AI economy
After the emergence of AI in 2017, the digitization process of humans entered a new stage. Different from the Internet and the mobile Internet, which mainly provide matching functions, AI can actually complete some online tasks. For example, image - recognition technology can accurately identify human faces, and knowledge - graph technology can analyze where a faulty machine has problems.
However, these work capabilities are all linked to specific models. OpenAI's GPT series of models have made AI capabilities generalized, that is, the same AI model has the ability to deliver work in a generalized manner. For example, GPT - 3 is the first model that simultaneously has the capabilities of dialogue, search, drawing, and coding.
Here, it is necessary for us to discuss the "collecting information - making decisions - taking actions" chain of human interaction with the natural world. When constructing this analysis framework, we referred to the "Perception–Decision–Control (PDC)" theory widely used in cybernetics, artificial intelligence, robotics, and autonomous driving.
The reason for this is that when analyzing human interaction with the natural world, we found that disciplines such as cybernetics, artificial intelligence, robotics, and autonomous driving have comprehensively considered the entire activity chain, that is, the "Perception–Decision–Control" chain, when considering the interaction between machines and the natural world. And human interaction with the natural world essentially also consists of these three steps. Considering the expression habits, we express it as the "collecting information - making decisions - taking actions" chain.
The ability of AI to (generically) deliver work means that in the "collecting information - making decisions - taking actions" chain of human interaction with the natural world, computers can play a role in all three links.
Computers can complete information collection, part of "decision - making," and part of "action." The specific situation can be shown in the following table:
Chart 3: The participation of computers in the "collecting information - making decisions - taking actions" chain at different stages
Specifically, in terms of decision - making, algorithms have a more detailed and accurate understanding of the needs of economic entities (individuals/organizations/enterprises) than in the mobile Internet stage, and can make more precise and effective decision - making suggestions. As a result, the scope of authorization of algorithms by the human brain in decision - making will expand, and algorithms will play a greater role in decision - making.
In terms of action, in the first stage, computers can complete work in the pure digital world, such as programming, writing a copy, building a website, generating an advertising video, filling out an insurance policy. This part of the work was mainly done by programmers, copywriters, designers, and repetitive mental workers before.
In the second stage, after the development of embodied intelligence matures, computers can participate in completing work in the physical world, such as cleaning housework, working on the factory assembly line, logistics handling, and taking care of the elderly, which are currently done by human labor.
In 2025, in the process of the digitization of human society, it is an important time point. In this year, the ability of AI to (generically) deliver work begins to exceed that of humans.
Since GPT - 3, when AI began to have the general and generalized ability to complete work, if evaluated based on the human IQ benchmark, the IQ of AI has always been lower than that of humans. TrackingAI.org uses the human IQ test, the Mensa test, to evaluate the reasoning ability of AI, which can be used as a reference.
The mainstream models before 2025, such as GPT - 3.5, GPT - 4o, Grok - 3, Llama 3, Mistral, and GLM - 4 of Zhipu AI, all have an IQ lower than 100, which is the average level of humans. So when we use these models and AI applications developed based on them, we will feel that these products are "a bit stupid" and cannot well meet our needs. However, the models released since the end of 2024, especially since 2025, such as OpenAI o3, Gemini 2.0, Gemini 2.5 Pro, Claude 4, DeepSeek R1, etc., have an IQ level that has exceeded the human average of 100. From the actual performance, many models have reached the range above 110.
The IQ of these models is equivalent to the top - ranked level among humans, even the top 10% level, or the IQ level of students from prestigious universities. (For AI engaged in economic activities, a better evaluation benchmark is to specifically evaluate its ability to engage in economic activities. We can initially define this evaluation benchmark as the "economic Turing test." The specific criteria for the "economic Turing test" will be elaborated in subsequent articles.)
For example, OpenAI o3 is evaluated to have reached the "genius - level," and ByteDance's Doubao model also achieved a score in the 2025 Chinese college entrance examination that would qualify it for admission to Tsinghua University or Peking University. This is why, from the user experience perspective, many AI agents since the end of 2024 have become "useful," and there have been many AI agents with outstanding effects.
Chart 4: The scores of various large - scale AI models in the Mensa IQ test. Source: https://trackingai.org/home, accessed in May 2025
In view of the above, at this moment, in May 2025, we are at an important juncture in human history.
The computer, invented by humans with the desire for "automated calculation," has fully developed the ability of "collecting information - making decisions - taking actions" in the interaction between humans and the natural world about eighty years after its birth, and its ability is at the critical point of exceeding that of humans.
The basic chain of economic activities,