GuShiio.com Article source: intellectuals
Image source: Generated by AI
The birth of DeepSeek may mark the arrival of the “Edison moment” of artificial intelligence. Although Faraday discovered the phenomenon of electromagnetic induction as early as 1831, providing a theoretical basis for the emergence of generators and motors, it was not until half a century later that Edison invented durable and cheap electric lights and built the Pearl Street Power Station that stably output electricity. Only then did the human world enter the electrical age as a whole. DeepSeek’s unique technical solution may mark that artificial intelligence truly has the ability to enter thousands of households and change all walks of life.
The question I want to discuss here is, why was DeepSeek not born in the Internet giant and “Six AI Tigers” that have invested tens of billions of yuan in big language models, nor in universities and research institutes that undertake a large number of national-level artificial intelligence projects, but born in the deep pursuit of this small company that was not large and previously unknown to policy makers and the public-even though the former has far more resources than the latter?
Perhaps we can find several clear clues from the organizational system of modern science and technology.
The nature and positioning of major Internet companies and traditional scientific research institutions are very different, but their organizational characteristics are highly similar. They are both “bureaucratic” organizations written by Marx Weber. Such organizations have two clear characteristics:First, in terms of organizational models, there are clear authority, upper and lower levels, and professional division of labor; second, in terms of work processes, there are clear rules, procedures, and performance appraisal methods.
It is precisely by relying on these two characteristics that bureaucratic organizations can organize thousands of people to efficiently achieve specific goals. From ancient water conservancy projects to modern high-speed rail 5G, from manned spaceflight to the human genome project, bureaucratic organizations have proved their great power and value. The same is true in the field of artificial intelligence: although the big language model was not originally born in China, driven by bureaucratic organizations, China institutions have made amazing progress in the development of big language models. Long before the birth of DeepSeek, China’s big models could be found in major lists around the world, such as Tongyi Qianwen, Doubao, Kimi, Zhipu Qingyan, etc.
However, it should be noted that there are two clear prerequisites for bureaucratic organizations to exert their power:The goals are clear and the path to realization are clear-in other words, the project goals are “engineering”.Because only when the goals and paths are determined can the hierarchical organization split and refine the goals along the paths, and finally implement them at every level within the organization, so that the goals and paths can be implemented into the performance evaluation of each organization member.
It is also important to note that bureaucratic organizations clearly cannot be used to create source innovation from 0 to 1.Because source innovation essentially cannot define goals in advance, let alone plan paths in advance. Although the huge human genome project can be split up layer by layer, the prerequisite for all this is that Watson and Crick first understood the molecular nature of genes in 1953; domestic large models can emerge one after another, but without the foundation of convolutional neural networks, transformers and Llama, there is no way to talk about the fierce competition between domestic large models around scale and cost performance.
What’s more, bureaucratic organizations are not only unable to proactively breed innovation at the source, but actually destroy innovation at the source (intentionally or unintentionally).Because the essence of source innovation is that it cannot be accurately predicted, random, and even improper. Its emergence requires wild exploration, a fanatical pursuit of the origin of things, personality, and a flash of inspiration. The more strict and thorough the hierarchical system, strict division of labor and performance appraisal of a hierarchical organization are, the more there will be no opportunities and soil for source innovation-the latter will be identified by the organization as ineffective, wasteful, and destructive. After the success of ChatGPT, members of OpenAI wrote “Why Greatness Can’t Be Planned”. As the title suggests, OpenAI’s series of source innovations are the product of surprise, enthusiasm, bold ideas, and the courage to trial and error.
For example, bureaucratic organizations are like modern industry. As long as the goals are clear and the paths are clear, it can be fully advanced through a strict division of labor and assessment system and are invincible. But to breed real source innovation, traditional agriculture is needed. Once a handful of seeds are sown, all we can do is water, fertilize, and wait patiently.
Having said that, let us return to the discussion of China’s scientific research system.
To a certain extent, although people use terms such as “interest-oriented” and “free exploration” when talking about scientific research, modern scientific research activities around the world are carried out under the leadership of hierarchical organizations. This in itself is not surprising. On the one hand, the main supporter of modern scientific research activities is the government (that is, people’s taxes), and there should be strict organizational models and work processes to cope with the review of taxpayers and regulators; on the other hand, modern scientific research activities often need to organize a large number of scientific researchers to carry out long-term team research, and a clear division of labor and performance management are inevitable.
The real problem is thatAgainst the backdrop of the imperative of this strict organizational model, do we leave enough flexibility for real source innovation?
There are also many successful cases in this regard. Google’s unique 20% time policy allows employees to devote 20% of their working time to free exploration, bringing them important innovations such as Gmail and AdSense. Bell Labs is certainly a large traditional scientific research institution, but a culture that allows free exploration has also given birth to great inventions such as the transistor.
But people familiar with the domestic scientific research system may immediately realize that this space is very cramped and scattered, if not completely disappeared.
In terms of organizational model, our scientific researchers have an extremely complex division of labor and hierarchy. As the academicians at the top of the pyramid, they are basically completely separated from the front line of research, but they have huge rights to allocate research resources. However, scientific researchers who have just entered the industry are trapped in the complex chain of senior “hats” such as graduate students, postdoctoral students, young teachers, professors, four young talents, outstanding youth/Yangtze River. Rather than focusing on important scientific and technical issues, researchers are often more concerned about how to obtain higher-level titles in the chain.
In terms of work flow, complex and dynamic scientific research activities are also cut into fragmented fragments vertically and horizontally. The plan and actual reimbursement of each funding; the application, opening, annual summary and closing report of each research topic; scientific research findings are quantified as impact factors and citations of the research paper; educational innovation is quantified as number of class hours and teaching rewards. Compared with how to solve an important scientific and technical problem, researchers ‘time and energy are consumed a lot of time and energy to meet a variety of complex quantitative procedural performance appraisal indicators.
To be fair, such problems are not unique to China. Taking the United States as an example, Katalin Kariko, the inventor of the mRNA vaccine, has not received financial support or permanent teaching positions for a long time, and Keytruda, the most successful anti-cancer drug in history, has been neglected in the corner of large companies for a long time. To be equally fair, we must also admit that even with restrictions of one kind or another, original innovation continues to be born on the land of China. I don’t need to list them one by one. Readers who pay attention to science and technology news can naturally list them as well.
But what I want to ask more is, can our scientific research system be loosened again so that the same resource investment can be more efficiently used in the breeding of original innovation?
For example, within the scientific research community, is complex division of labor and hierarchy really unavoidable?Every year when the application and review season of scientific research projects comes, we see a lot of resources and energy being invested in unnecessary circles, greetings, and intimacy. Will we have eliminated the hat and the layers of promotion ladders, and we no longer have the ability to evaluate the work of scientific researchers? Of course not! Comments from peers are the most direct to the essence. After the birth of DeepSeek, the CEO of OpenAI publicly admitted that “we are on the wrong side of history”; while Microsoft and Amazon’s cloud computing services quickly deployed and opened DeepSeek’s model portal-feedback from peers is a direct recognition of DeepSeek’s capabilities. Could it be that if the fund review agency does not give it a title, good scientific and technological achievements will be buried and ignored?
For another example, when the goals and paths are clear,”organized scientific research” can certainly reflect the power of efficiency and scale. The Human Genome Project is a good example: When the molecular nature of genes has been revealed and gene sequencing technology has matured, it is a reasonable choice to organize scientists and scientific research organizations from all over the world to tackle key problems. But faced with a large number of important issues whose goals are still undefinable and whose paths are still unclear-such as how to overcome cancer and aging, how to understand the working principle of the human brain, how to use AI technology to predict complex life activities, etc. –Should we let go of our obsession with efficiency and scale and let scientists with great enthusiasm for these issues follow their own inspiration and make explorations that have a huge possibility of failure?
Summary and several specific suggestions:
·Bureaucratic organizations are only suitable for engineering projects with clear management goals and clear realization paths, and are not suitable for source scientific and technological research full of uncertainty
·Source innovation projects like DeepSeek are difficult to be born in hierarchical organizations, whether they are major Internet companies or traditional scientific research institutions.
·Source science and technology research is inherently difficult to produce through targeted support, but it can be incubated through extensive support; source innovation requires an agricultural perspective rather than an industrial perspective.
·It is inevitable that the modern scientific research system will gradually move towards a bureaucratic system, but the key lies in how much room for expansion is left for source innovation.
·The domestic scientific research management system has gone too far and deep under the bureaucratic system and needs to be strongly reversed. In areas where source innovation is needed, it may be necessary to eliminate labels on a large scale, eliminate most quantitative assessment indicators, and reduce the coverage of organized scientific research.
