Artificial intelligence is becoming a new force in knowledge production
2026-07-10
In recent years, the development of artificial intelligence has raised a serious question worth discussing: is AI just a tool to help humans complete retrieval, computation, writing, and analysis, or has it already participated in knowledge production in some sense? A common view is that AI lacks human consciousness, emotions, and life experiences, making it impossible to truly understand the world and produce knowledge. This view emphasizes the important differences between humans and machines, but it also tends to excessively narrow the judgment of knowledge to the subjective experience of humans.
The key to the problem is not whether AI thinks like humans, but how we understand knowledge production itself. Modern knowledge production is never just a direct expression of individual experience, but a cognitive outcome formed in tools, methods, evidence, models, and community norms. A more appropriate statement is that artificial intelligence is becoming a generative force in the knowledge production chain. It may not necessarily understand the world in a human way, but it has begun to push humanity to re-examine what can be seen, what is worth studying, and what can become knowledge.
Knowledge judgment cannot be limited to subjective experience
To discuss whether AI can participate in knowledge production, it is necessary to clarify a premise: whether the effectiveness of knowledge must be based on subjective experiences similar to those of humans. Of course, human knowledge cannot be separated from experience, body, language, and social life. Especially in the fields of ethics, aesthetics, historical understanding, etc., human situations and experiences have irreplaceable importance. But this does not mean that any system without human experience cannot participate in knowledge formation at all.
From an epistemological perspective, knowledge is not equivalent to temporary feelings, judgments, or opinions. A judgment to become knowledge usually requires the support of reasons, evidence, and methods, and can be tested, revised, and discussed within a certain community. The knowledge forms of different disciplines are not the same: natural sciences place more emphasis on laws, predictions, and experimental verification, social sciences place more emphasis on mechanisms, contexts, and causal explanations, and humanities place more emphasis on meaning, text, and historical interpretation. However, regardless of the type of knowledge, individual subjective experience cannot be the sole criterion. The reason why knowledge becomes knowledge is that it can go beyond personal feelings and enter into a public process that is arguable, communicative, and verifiable.
In fact, human understanding of the world has never relied solely on direct experience. The microscope has no life experience, but it opens up the microscopic world; Telescopes, without subjective consciousness, have changed humanity's understanding of the universe; Statistical models do not experience social pain, but can reveal the structural relationships behind population, disease, poverty, and mobility. Modern knowledge production already relies on intermediaries such as instruments, models, algorithms, databases, and computational simulations. They do not "experience" the world in a human way, but profoundly change what humans can know, how they know it, and how knowledge is proven.
Artificial intelligence is closely related to these cognitive mediators and has new features. It is not just about extending senses, accelerating computation or storing information, but also about discovering patterns in large-scale, high-dimensional, non intuitive data spaces and transforming these patterns into predictions, classifications, hypotheses, and explanatory clues. That is to say, AI is transitioning from "processing existing knowledge" to "generating cognitive objects".
Therefore, to determine whether AI is involved in knowledge production, it is not appropriate to simply ask whether it has human feelings and intentions, but rather to ask whether it reveals relationships that were previously difficult to see? Can verifiable questions be raised? Can it be transformed into stable knowledge through community discussions? Can we improve human understanding of nature, society, and ourselves? If these questions can be answered positively in some fields, we cannot simply deny the knowledge production significance of AI. This is not lowering knowledge standards, but expanding standards. Knowledge should not be based solely on machine output, but it should also not be based solely on human subjective experience. The reliability of modern knowledge comes more from evidence, methods, explanations, verification, and community review. The emergence of AI requires us to re understand the boundaries of knowledge production under new technological conditions.
AI has entered the stage of knowledge generation
Artificial intelligence has become a generative force in knowledge production, which can be illustrated more intuitively in the fields of Go and materials science.
Go has long been regarded as a field highly dependent on experience, intuition, and aesthetic judgment. Players accumulate an understanding of "good chess" through game charts, layouts, patterns, hand muscles, and long-term matches. The Go intelligent system, represented by AlphaGo, was initially known to the public for defeating the world's top Go players, but what is truly worth discussing is not just winning the game, but changing the way the Go community understands "good chess". This type of system did learn human chess manuals in the early stages, but the real breakthrough came from reinforcement learning and self play. It repeatedly explores the chessboard space determined by rules, gradually forming judgments that many human chess players have not fully understood.
The "37th hand" in the game between Lee Sedol and AlphaGo is often regarded as a landmark moment. This was a game that many professional chess players were not optimistic about or even surprised by at the time, but later proved to have strategic value. Afterwards, professional Go training generally shifted towards AI replay, win rate judgment, and change chart research. Some traditional formulas were re evaluated, and new "AI formulas" and "AI layouts" continued to emerge. This is not just a simple calculation acceleration, but an adjustment of the Go knowledge system: what is urgent, what is thickness, what is global efficiency, and what kind of local losses can be exchanged for greater overall benefits are all reopened by AI. This indicates that AI can generate practical and strategic knowledge in a specific rule space. It may not necessarily experience victory, competition, and aesthetics like human chess players, but it can discover new effective patterns through self play and prompt humans to revise existing judgments. In other words, AI is not just reproducing the chess principles that humans have already mastered, but expanding human understanding of chess principles within established rules.
Another more scientifically representative field is the involvement of AI in the discovery of new materials. Materials are the foundation of modern technology and industrial development. From high-performance batteries, semiconductor devices, aerospace materials to catalysts and biomedical materials, a deep understanding of material structure, properties, and preparation conditions is essential. In the past, searching for a new material often required researchers to rely on theoretical deduction, empirical judgment, and extensive experiments, repeatedly trial and error in a vast space of element combinations, structural types, and process parameters. In recent years, machine learning and generative models have been used for material performance prediction, candidate structure screening and generation, and experimental path optimization, enabling researchers to discover more targeted material combinations worth verifying in a vast space of possibilities.
The significance of such progress is not only to improve experimental efficiency, but also to change the way "candidate objects" are generated in scientific discoveries. In the past, many materials that may have special properties were difficult to enter the research field due to their large combination space and high experimental costs; Today, AI can infer a batch of potentially valuable candidate materials based on existing data and models, and then have researchers validate, revise, and interpret them through experiments. In other words, AI has not replaced the theoretical judgment and experimental verification of materials scientists, but has transformed some of the previously difficult to exhaust possibilities into research objects that can be screened, compared, and verified, promoting the formation of a new cycle of "model prediction experimental verification theoretical explanation" for scientific discoveries.
More noteworthy is that in the past year or two, research has attempted to have multiple artificial intelligence agents collaborate and simulate different roles within a research team, undertaking tasks such as literature search, hypothesis formulation, scheme design, tool invocation, result analysis, and report generation. Although this type of exploration is still in the process of development and requires strict validation, it has shown a trend: AI entering knowledge production is not just providing an answer, but beginning to embed a continuous chain of problem posing, path design, and result generation.
From the intelligent system of Go to the discovery of new materials, and to the exploration of multi-agent scientific research collaboration, it can be seen that the role of AI has exceeded the scope of auxiliary labor. It can discover patterns in complex spaces that humans have not yet mastered, and transform these patterns into new judgment criteria, research paths, and knowledge resources. It is in this sense that AI has entered the generation stage of the knowledge production chain.
AI expands the known world
The generative power of AI is not only reflected in natural sciences or fields with clear rules, but it is also expanding the observable scope of social sciences and humanities, albeit in a different way. In social sciences, many important issues are not lacking in materials, but are limited by the conditions and methods for dealing with large-scale complex materials. Public opinion, policy texts, social media, news reports, judicial documents, corporate announcements, interview materials, etc. all contain rich social information. In the past, it was difficult for a researcher to systematically read and compare such vast materials; Today, AI can help researchers identify changes in discourse, emotional structures, group identities, issue diffusion, and shifts in institutional language. For example, when studying public issues, AI can help discover how certain words move from peripheral expressions into mainstream discussions, how certain emotions are interconnected with specific events, and how different groups use different narratives to understand the same problem. These findings may not necessarily constitute a direct conclusion, but they can provide important mechanistic clues for social science research. They suggest to researchers: which phenomena are worth explaining, which variables may be relevant, which group differences need further investigation, and which institutional backgrounds cannot be ignored.
In the humanities, the value of AI is not to replace literary critics, historians, or philosophers, but to change the way materials are organized and problems are discovered. Literary research can use AI to compare the themes, imagery, narrative structures, and stylistic changes in a large number of works; Historical research can use AI to extract people, places, events, and relationships from archives, newspapers, letters, chronicles, and judicial documents; The study of intellectual history can use AI to track the usage changes of a concept in different periods and contexts. These tasks do not equate to completing explanations, but they can make previously scattered, hidden, and difficult to exhaust associations visible again.
This means that AI not only improves efficiency, but also expands the boundaries of the 'known world'. It enables more natural structures, social patterns, and cultural associations to enter into a knowledge process that is analyzable, comparable, and debatable. Relationships that were difficult to see in the past may now be prompted through algorithms; Materials that were difficult to organize in the past may now be rearranged into problem networks; The trend that relied on individual experience to capture in the past may now manifest in large-scale materials.
Of course, social mechanisms and cultural significance cannot be automatically judged by algorithms. AI can discover patterns in social texts, but cannot independently judge the institutional logic behind these patterns; AI can propose associations between literary and historical materials, but cannot automatically complete contextualized explanations; AI can generate summaries and counterexamples of philosophical arguments, but it cannot replace human normative judgments. The production of knowledge in social sciences and humanities still relies on the community of theory, history, value, and interpretation. Therefore, AI is more like a discoverer of mechanism clues and meaning associations in these fields, rather than a judge of final conclusions. It expands the ability of humans to ask questions and also requires humans to take on a higher level of explanatory responsibility.
Towards a knowledge future of human-machine co construction
The profound changes brought about by artificial intelligence are not in its ability to complete existing tasks faster, but in its ability to change what can be seen, what is worth studying, and what can become knowledge. It does not have a subjective human experience, but this does not prevent it from generating testable, applicable, and debatable cognitive outcomes under certain conditions. It is difficult to explain the ongoing knowledge practice by denying the knowledge production function of AI based solely on whether it has human experience.
Admitting that AI is becoming a generative force in the knowledge production chain does not mean handing over the truth to machines. On the contrary, the more AI can generate patterns, structures, and hypotheses, the more humans need to strengthen verification, interpretation, and standardization. The key to the future is not to make simple trade-offs between humans and machines, but to establish new connections between algorithm discovery, empirical verification, theoretical explanation, and public discussion.
The knowledge production in the AI era is neither a retreat of humans nor a monologue of machines, but a way for humans to reorganize, understand the world, explain society, and reflect on themselves under new intelligent conditions. AI is becoming a new force in knowledge production. Only by facing up to this can we take a more proactive approach in building an open, reliable, and responsible human-machine collaborative knowledge system.(Outlook New Era)
Author: Chen Hao (Professor at the School of Sociology and the Institute of Frontier Interdisciplinary Research, Nankai University)
Recommended Reading Change it
