The Human-Machine Knowledge Spiral

Abstract Nonaka emphasized that innovation is the result of a continuous back-and-forth between tacit and explicit knowledge. Artificial intelligence introduces a fundamentally new object into this process—tacit machine knowledge—but Nonaka’s ideas are more relevant than ever. The central role of the knowledge-creating company remains the same: to create the shared context in which different kinds of knowledge can feed off each other, become organizational knowledge, and set off further cycles of innovation.

Introduction. Recent advances in AI are forcing a cognitive version of the Copernican revolution. Just as it once required a profound conceptual leap to accept that the Earth is not the center of the universe, we may now need to accept that human intelligence is not at the center of intelligence (Klowden and Tao, 2026). In organizational terms, this means confronting a possibility that lies outside the assumptions of traditional theories of the firm: human knowledge may no longer occupy a central place in organizational intelligence (Ide and Talamàs, 2025, 2026; Chatterji, Rock, and Talamàs, 2025b; Beraja and Talamàs, 2026). This transformation implicates one of the most influential theories in our field, the knowledge-based view of the firm. We revisit this theory with this new orientation and argue that Ikujiro Nonaka’s insights are more relevant than ever.

Building on Polanyi’s (1966) famous observation that “we know more than we can tell,” Nonaka (1991) conceptualizes innovation as a process of organizational knowledge creation. Tacit knowledge—acquired through experience and practice—is articulated into explicit knowledge that can be shared across the organization. This explicit knowledge is then internalized through action and use, further expanding tacit knowledge. Through repeated cycles of articulation and internalization, human knowledge is amplified at progressively broader organizational levels—and eventually across organizational boundaries as well. The central role of the knowledge-creating company is to provide the shared context in which this spiral of knowledge can recur.

Artificial intelligence introduces a fundamentally new form of knowledge into organizations (Brynjolfsson, Li, and Raymond, 2025; Ide and Talamàs, 2025, 2026; Chatterji, Rock, and Talamàs, 2025b).

Unlike earlier technologies, AI learns from data and does not follow explicitly specified rules, so it acquires forms of knowledge that are also difficult to articulate or explain. In other words, machines now also “know more than they can tell” (Brynjolfsson and Mitchell, 2017; Brynjolfsson, Mitchell, and Rock, 2018; Autor, 2024). Thus, a new type of knowledge has been born: tacit machine knowledge (Ide and Talamàs, 2025; Chatterji, Rock, and Talamàs, 2025b). Like its human counterpart, tacit machine knowledge is hard to articulate. But unlike its human counterpart, it need not remain local: once embodied in a model, it can be made available at scale through software.1 In the age of AI, knowledge no longer moves only between tacit and explicit forms as Nonaka envisioned, but also into and out of machines. Tacit machine knowledge also amplifies the transfer of knowledge across teams and organizations that Nonaka described. But we argue that Nonaka’s insights are more relevant than ever. In particular, the central role of the knowledge-creating company continues to be to provide the shared context in which a virtuous spiral of knowledge can recur.

Related work. and contribution. This article follows the knowledge-based tradition that asks why firms exist when knowledge is difficult to create, transfer, and use (Epple et al., 1991; Kogut and Zander, 1992; Zander and Kogut, 1995; Kogut and Zander, 1996; Grant, 1996; Argote, 1999). Nonaka’s (1991) distinctive contribution is to show that knowledge creation requires making different kinds of knowledge available to one another in a shared context (Nonaka, 1994; Nonaka and Takeuchi, 1995; Nonaka and Konno, 1998; Nonaka and von Krogh, 2009).

A growing literature has begun to revisit this literature in light of AI and generative AI. Sanzogni et al. (2017) connects artificial intelligence to the knowledge-management debate by asking whether AI changes how we should understand the tacit dimension of knowledge. More recent work treats ChatGPT and related systems as tools that can support Nonaka’s original spiral of knowledge (Sumbal and Amber, 2025). Other work goes further and asks how generative AI can be treated as a participant in knowledge creation (Matsumoto et al., 2025; Böhm and Durst, 2026; Kirchner and Scarso, 2026; Uchihira, 2026). A related line of work asks whether AI-mediated interaction can itself become a new form of shared context (He et al., 2026). This literature is important because it recognizes that AI can participate in the processes through which organizations create knowledge.

The novelty of this paper is to revisit Nonaka’s ideas in the context of the rise of a fundamentally new form of knowledge—tacit machine knowledge—which does not sit comfortably in the traditional tacit–explicit divide that Polanyi (1966) emphasizes and Nonaka (1991) builds on. We argue that Nonaka’s insights and prescriptions apply mutatis mutandis once the relevant spiral includes tacit machine knowledge. The key property of this new type of tacit knowledge is portability: it can be deployed at scale while remaining opaque to humans. AI therefore dramatically amplifies the transfer across teams and organizational boundaries that Nonaka emphasizes, as well as the importance of shared context for the knowledge-creating company.

Method. 2 The Human–Machine Knowledge Spiral Nonaka (1991) emphasizes that making personal knowledge available to others is the central activity of the knowledge-creating company. As the following example illustrates, in the age of AI, a new but analogous activity becomes central for innovation: to make human and machine knowledge available to one another.

In 2015, Matías Muchnick, Karim Pichara, and Pablo Zamora founded NotCo with an ambitious goal: to reproduce familiar animal-based foods using plants. They built an artificial-intelligence system called Giuseppe, after the Renaissance painter Giuseppe Arcimboldo, who composed human faces out of fruits and vegetables.2 Giuseppe searched across ingredient data, chemical analyses, scientific publications, previous formulations, and sensory results on a scale no food scientist could match. Given a target such as milk, it proposed plant combinations that might reproduce its flavor, aroma, texture, appearance, nutrition, and behavior in cooking. Because Giuseppe could explore a vastly larger space than any human team, it was tempting to think that NotCo’s creative problem had been solved.

Experience proved otherwise. In one early attempt to create NotMilk, Giuseppe proposed a formulation containing dill. The milk came out green. Giuseppe had detected properties in dill that appeared useful in the data, but it missed something obvious to anyone standing in NotCo’s kitchen: whatever else milk may be, people do not expect it to be green. The failure made it obvious to everyone involved that—despite having access to large amounts of data—Giuseppe had a serious lack of context.3 The green milk vividly illustrates the different nature of machine knowledge. Giuseppe excels at detecting potentially useful relationships among ingredients and molecular compounds. But it cannot taste the product, smell it, or understand everything people expect from a glass of milk. In contrast, chefs can recognize if a prototype is too grassy, too thin, too sweet, or simply wrong. They can appreciate color, aroma, mouthfeel, aftertaste, stability, and performance under heat. Much of this knowledge has been acquired through experience and is difficult to reduce to rules.

NotCo’s response to the green milk fiasco was to orchestrate a spiral of knowledge between Giuseppe and its human teams. Giuseppe proposed formulas. Research chefs made them. Chefs, food scientists, and sensory panels tasted, smelled, measured, and modified the results. Their judgments about color, flavor, texture, mouthfeel, and appearance were recorded and returned to Giuseppe, which then generated new proposals in light of the feedback.4 By 2021, more than twenty chefs and food scientists were working with Giuseppe—testing more than one hundred recipes a month.5 Many proposals failed. Others opened paths that the human teams were unlikely to have explored.

In searching for a plant formulation with the aromatic qualities of cow’s milk, Giuseppe helped direct the team toward an improbable but winning combination: pineapple and cabbage.6 NotMilk—with its pineapple and cabbage—reached Whole Foods stores nationwide in the United States in 2020.7 A 2021 financing round valued NotCo at $1.5 billion.8 Nonaka emphasized that products embody organizational knowledge. They are the concrete forms in which organizational knowledge is crystallized, and they are also triggers that set off the next cycle of knowledge creation (Nonaka, 1991, 1994). Matsushita’s bread machine carried the baker’s “twist dough” insight beyond the team that had created it. The machine made that insight available outside the original process in a durable form. In this sense, the spiral of knowledge in one team can help set further spirals in motion in other teams.9 Similarly, NotMilk embodies the formula created through NotCo’s human-machine knowledge spiral. Its ingredient list makes the pineapple-and-cabbage combination visible. Other consumers and companies can taste the result, inspect it, and learn from it.

But NotCo can now make a second object travel: Giuseppe itself.

Discussion. 3 From Chaos to Concept—Again Nonaka (1991) emphasizes the value of redundancy: the conscious overlap of information, activity, and responsibility. Redundancy creates the common cognitive ground on which people situated in different contexts can sense what others are struggling to articulate. It places different participants around the same problem and allows their different perspectives to interact.

At NotCo, Giuseppe, the chefs, the food scientists, and the sensory panels repeatedly confronted the same prototype. The machine examined relationships among ingredients and molecular properties.

The chefs experienced its color, aroma, taste, texture, mouthfeel, and culinary behavior. Food scientists measured stability, nutrition, and production performance. Their work overlapped at the prototype on the kitchen table.

The prototype was a shared context—what Nonaka and Konno (1998) call ba—in which knowledge could acquire meaning through interaction. From the standpoint of information processing, having multiple expert chefs carefully evaluating Giuseppe’s formulas may look like duplication. From the standpoint of knowledge creation, this redundancy creates the common cognitive ground whose value Nonaka emphasized.12 Discrepancies are especially valuable. Nonaka argued that ambiguity, conflict, and breakdown can force members of an organization to reexamine what they take for granted. The green milk did exactly that. The fiasco exposed the distance between molecular resemblance and the lived meaning of milk, and made part of the chefs’ tacit knowledge visible: milk has a color, and that color belongs to what people understand milk to be.13 Giuseppe’s successful proposals also echo Nonaka’s movement from metaphor to model. “Pineapple and cabbage as milk” brought distant domains together before anyone possessed a complete explanation of their connection. The chefs and scientists had to examine precisely how these ingredients were and were not like the product they hoped to reproduce. A wild proposal turned into useful shared knowledge through a process that sounds very similar to the one Nonaka described.14 Knowledge creation also requires direction. Nonaka insisted that the knowledge-creating company is as much about ideals as it is about ideas. Giuseppe could search what was possible. NotCo’s ideal—familiar food made from plants without surrendering the qualities people valued—gave that search a purpose. The ideal does not specify dill, pineapple, or cabbage. Instead, it gives the team a way to decide which wild proposals deserve another turn of the spiral.15 Nonaka’s account of organizational roles reads with similar force today. Frontline employees possess detailed knowledge of “what is.” Senior leaders give voice to “what ought to be.” Middle managers operate between the two, translating broad ideals into concepts that can encounter the reality of products, technologies, and markets. They are, in Nonaka’s (1991) words, the knowledge engineers of the company.

AI adds another source of knowledge to this structure. Domain experts understand the reality of the product and the customer, AI specialists know AI’s capabilities, and managers have a sense of where these can be useful. The machine contributes patterns and proposals that neither group may have anticipated. The work of connecting these perspectives to one another and to the organization’s purpose is analogous to the work Nonaka emphasized was central to the knowledge-creating company.

Nonaka’s insights also suggest that knowledge creation cannot reside inside an isolated AI unit.

Nonaka gives no department or group of experts exclusive responsibility for creating knowledge.

Conclusion. While everything about Giuseppe would have been unimaginable when Nonaka wrote about Matsushita’s bread machine 35 years ago, we believe the organizational process shaping it would have been very familiar to him. An insight appears in one place, encounters a different form of knowledge, acquires meaning in a shared context, and returns to begin another cycle of the knowledge spiral.

The AI revolution itself offers another illustration of how relevant Nonaka’s insights are today.

Before ChatGPT could converse with the world, human trainers staged conversations in which they played both the user and the assistant. They were given model-written suggestions to help compose their responses. They then ranked alternative replies generated by the model, and those judgments shaped subsequent training (Ouyang et al., 2022).16 The same movements seen at NotCo appear here at a different scale. Vast bodies of explicit text are encoded in the model. Human demonstrations and rankings embed judgments about what constitutes a useful response. Model outputs make learned capabilities available for inspection and evaluation. By observing unexpected successes and failures, the people building and using the system acquire new knowledge about what the model can do and where it remains limited.

This is the human–machine knowledge spiral operating inside an AI laboratory. The main product of this spiral is tacit machine knowledge that can be made available far beyond the laboratory.

Through software products and APIs, tacit machine knowledge formed inside one laboratory is now becoming central to people and organizations around the world (Chatterji et al., 2025a; Johnston et al., 2026). The spiral operating inside a small number of AI laboratories is therefore feeding the knowledge spirals of countless other companies—at scale.

Here again, Nonaka explains why access to tacit machine knowledge alone—no matter how impressive— does not create organizational knowledge. AI models arrive to an organization without its context or ideals. The imported knowledge becomes organizational knowledge only after it enters the humanmachine knowledge spiral.