Bloomberg: How will artificial intelligence AI disrupt the organizational structure of enterprises?

The economic system has long been built on the idea that professional knowledge is scarce and expensive. Artificial intelligence is about to make this professional knowledge abundant and almost free. This article is sourced from an article by AI Pioneer, reprinted by TechFlow. (Background: Senior engineer's feelings: Junior developers now rely on AI for 'independent thinking,' Musk responds) (Background: OpenAI unlocks Deep Research: Paid users can query 10 times a month, Microsoft releases multimodal AI agent Magma) For most of human history, hiring a dozen experts with PhDs often required a huge budget and months of preparation. Now, just by entering a few keywords in a chatbot, one can instantly access the wisdom of these 'brains.' When the cost of wisdom becomes lower and the speed faster, the fundamental assumption supporting our social system — 'human insight is scarce and expensive' — will no longer exist. What changes will occur in company organizational structure when we can call on the insights of a dozen experts at any time? How will our innovative approaches evolve? How should each of us approach learning and decision-making? The question facing individuals and businesses is: when wisdom is readily available and almost cost-free, how will you act? The historical program of 'discounting' wisdom In history, we have witnessed many times a significant drop in the cost of knowledge and a rapid expansion of dissemination channels. The emergence of the printing press in the mid-15th century greatly dropped the cost of disseminating written information. Before that, texts were often hand-copied by professionals such as monks, which was both costly and time-consuming. With this bottleneck broken, Europe witnessed profound social changes: the Protestant Reformation caused massive upheaval at the religious level; literacy rates rose rapidly (laying the foundation for the popularization of elementary education); and scientific research flourished through printed publications. Commercially oriented countries such as the Netherlands and the UK benefited greatly, with the Netherlands entering a 'Golden Age' and the UK continuing to play a significant role on the global stage for several centuries. Over time, as mass literacy and public education became widespread, overall societal wisdom improved, laying the foundation for industrialization. Specialized positions in factories led to a more complex division of labor, driving economic growth. By the late 18th century, countries with higher male literacy rates were the first to industrialize; by the late 19th century, the most technologically advanced economies were often those with the highest literacy rates. People acquired new skills, creating more professional positions and forming a virtuous cycle that continues to this day. The advent of the internet has taken this trend to new heights. In my childhood, if I wanted to research a new topic, I had to bring notes to the library to search for references, which could take up half a day just for this step. Back then, acquiring knowledge was both expensive and difficult. Today, artificial intelligence has taken over the relay baton of this millennium-long 'discounting of wisdom,' opening a new chapter for our economy and way of thinking. My 'Eureka moment' with ChatGPT When I first used ChatGPT in December 2022, I felt that it was a milestone product. Initially, I just used it for some 'digital tricks,' like having AI 'rewrite the Declaration of Independence in Eminem's style' (the adapted lyrics it wrote were something like 'Yo, we gotta shout it out, the people here won't be knocked down,' and so on). Looking back, it was like having a Michelin-star chef make grilled cheese sandwiches for you, a bit of an overkill. It wasn't until one afternoon in January 2023 when my 12-year-old daughter and I spent a few hours designing a brand new tabletop game with the help of ChatGPT that I truly realized the power of these tools. At that time, I told the AI which board games we liked and disliked, and asked it to analyze the commonalities among them. It found that we liked game mechanics that involved 'laying paths,' 'managing resources,' 'collecting cards,' 'strategic planning,' and had a 'suspenseful outcome,' while disliking certain modes common in Risk or Monopoly. I asked it to come up with some less obvious but important game ideas based on these elements and hoped for a historical background. ChatGPT then came up with a game called 'Elemental Discoveries,' where players take on the role of 18th-19th century chemists, conducting experiments, collecting and trading resources to earn points, and being able to interfere and sabotage each other. I then had it further refine the resources, gameplay, game mechanics, and roles suitable for players to play. It proposed positions like 'Alchemist,' 'Saboteur,' 'Merchant,' 'Scientist,' and matched them with historical chemist figures such as Lavoisier, Joseph-Louis Gay-Lussac, Marie Curie, and Carl Wilhelm Scheele. With the help of the then 'basic' ChatGPT, we created a somewhat rough but playable tabletop game in just two to three hours. In the end, I had to stop, partly because of lack of time, and partly because I was exhausted. That experience made me truly realize how an AI 'collaborator' could compress what would typically take weeks of R&D into just a few hours. Just imagine the enormous potential it could bring if used for product development, market analysis, or even corporate strategy? In this process, I saw that ChatGPT was not just repeating or assembling facts; its performance demonstrated analogical and conceptual thinking capabilities, linking ideas with real-world references and truly outputting creative solutions based on demand. From 'Random Parrot' to 'Depth Thinker' At one trillion, this scale is already astonishing. Large language models that support ChatGPT often have tens of billions, hundreds of billions, or even trillions of parameters, their complexity is staggering. We still do not fully understand why these models work and how they operate. When these models repeatedly made breakthroughs over the past seven years, some theoretical scholars insisted that they could not produce truly new things — in 2021, some researchers even coined the derogatory term 'stochastic parrots.' This is because large language models basically predict text based on statistical patterns in training data, as if parrots randomly repeat phrases. However, for those who continue to experience and admire these tools, it is hard to believe that they are merely repeating. This viewpoint seems even less tenable in the past six months. The initial large language models were more like 'speaking intuitively,' lacking 'reflective' ability and 'self-awareness.' In the words of Nobel laureate Daniel Kahneman, humans mostly rely on System 1 (intuitive, quick response) thinking most of the time, but when deep thinking is required, we switch to System 2 (slower, more cautious, and less error-prone). Earlier versions of ChatGPT and its competitors mostly only exhibited performance similar to System 1, lacking the reasoning process of System 2. This situation began to change in September 2024, when OpenAI released a reasoning model called o1, which can decompose and verify intermediate conclusions for complex logical problems spanning multiple steps (and can backtrack and correct if necessary), thereby better arriving at the final result. Compared to traditional large language models that rely solely on memory or surface pattern matching, the new reasoning model gradually acquired...