The Linkage Premium: Why AI Makes Knowledge Cheap and Linked Experience Scarce

Abstract

Most professionals treat knowledge as their competitive edge. They collect degrees, certificates, and frameworks. They underinvest in the experience they already hold. This belief made sense when knowledge was scarce. It makes less sense now. Generative AI is reducing the cost of explicit knowledge to near zero. Field evidence shows that AI helps novices most, which compresses the advantage that knowledge once provided. The same evidence shows something else. The people who benefit most from AI advice are those with the judgment to filter it. That judgment comes from experience. Yet experience has a structural weakness. It is tacit, local, and hard to explain. Unlinked, it cannot be tested, transferred, or recognized. This essay introduces The Linkage Model, which explains how experience gains value when it is connected to the right knowledge. It identifies four linkage moves: Surface, Name, Calibrate, and Transmit. AI can now assist each one. The central argument is plain. Knowledge is becoming a commodity. Experience is becoming the scarce input. The competitive edge belongs to those who link the two.

1. Introduction

In a large field study of customer support agents, researchers found that a generative AI assistant raised productivity by 14 percent on average. Novice and low-skilled agents improved by 34 percent. Experienced and highly skilled agents saw minimal gains (Brynjolfsson, Li and Raymond, 2025). The assistant had been trained partly on the conversations of the best agents. In effect, their experience was turned into knowledge and handed to everyone else.

This finding contains a contradiction. The experienced agents were the source of the value. The novices captured most of it. The experts' edge did not disappear. It was transferred. What had taken years to build became available on day one to someone who had never faced a difficult customer.

The standard reading of this result is that AI levels the playing field. That reading is incomplete. A second body of evidence shows the opposite pattern in different conditions. When the task requires judgment about which advice to follow, AI widens the gap between the experienced and the inexperienced (Otis et al., 2025). When the task falls outside what AI does well, people who rely on it perform worse than those who do not (Dell'Acqua et al., 2023).

These results matter because most professionals, institutions, and education systems still act on an older assumption. They assume knowledge is the asset. They invest in acquiring it, displaying it, and certifying it. Experience is treated as a byproduct of time served. It is rarely articulated, rarely tested, and rarely made transferable.

This essay argues that the assumption is now inverted. Knowledge is becoming abundant. Experience is becoming the scarce input. But experience only produces advantage when it is linked to the right knowledge. The essay develops four theoretical lenses, tests them against four cases, and proposes The Linkage Model as a reusable framework for individuals and institutions.

2. Theoretical Framework

2.1 Polanyi's Tacit Dimension and the Paradox That Protected Experience

Michael Polanyi observed that "we can know more than we can tell" (Polanyi, 1966). A skilled practitioner recognizes a pattern before they can explain it. A craftsman adjusts a tool without stating the rule. This is tacit knowledge. It lives in practice, not in documents.

David Autor later named the economic consequence Polanyi's Paradox (Autor, 2015). Tasks that people perform through tacit knowledge were hard to automate. Machines needed explicit rules. Humans could not supply rules for what they could not articulate. Tacit experience was therefore protected from automation.

Applied here, Polanyi's Paradox explains why experience held its value for decades. It could not be copied because it could not be written. Generative AI weakens this protection. Models trained on the outputs of experienced people can reproduce some of their patterns without anyone stating the rules. The customer support case shows this directly. Experience that is left unarticulated can now be harvested by others.

2.2 Nonaka's Knowledge Conversion and the Externalization Bottleneck

Ikujiro Nonaka described how organizations create knowledge through four conversions: socialization, externalization, combination, and internalization (Nonaka, 1994; Nonaka and Takeuchi, 1995). The critical step is externalization. This is the conversion of tacit experience into explicit concepts.

Externalization has always been the bottleneck. It requires time, reflection, and language. Most experienced professionals never complete it. They carry decades of pattern recognition that never becomes a model, a framework, or a teachable method.

Applied here, Nonaka's model identifies where AI changes the economics. AI can act as an interlocutor that draws out tacit patterns. It can ask the follow-up question. It can propose a name for a pattern. It can connect a single observation to an established body of research. The externalization step, once slow and rare, can become fast and routine.

2.3 Kahneman and Klein's Conditions for Valid Expertise

Not all experience deserves trust. Daniel Kahneman and Gary Klein resolved a long debate about intuition with a shared conclusion (Kahneman and Klein, 2009). Intuitive expertise is valid only under two conditions. The environment must be regular enough to be learnable. The practitioner must receive timely feedback on their judgments.

Where these conditions hold, experience produces reliable intuition. Firefighters and nurses are examples. Where they fail, experience produces confident error. Long-range political forecasting is an example.

Applied here, this lens prevents the argument from becoming a celebration of seniority. Experience is not automatically valuable. It becomes valuable when it is calibrated against knowledge that tests its validity. This is the precise meaning of "the right knowledge" in the essay's thesis. The right knowledge is the knowledge that confirms, corrects, or bounds what experience claims to know.

2.4 Absorptive Capacity and the Complement Problem

Wesley Cohen and Daniel Levinthal argued that an organization's ability to use external knowledge depends on its prior related knowledge (Cohen and Levinthal, 1990). They called this absorptive capacity. Knowledge from outside is useless without something inside that can recognize, assimilate, and apply it.

Applied here, absorptive capacity explains why AI advice produces uneven results. AI supplies external knowledge on demand. The value of that knowledge depends on what the recipient already holds. Experience is the absorptive capacity for AI-generated knowledge. Without it, advice cannot be filtered. With it, advice becomes leverage.

2.5 What the Four Lenses Predict Together

The four lenses produce a combined prediction. AI will reduce the value of explicit knowledge held alone (Autor). It will make the externalization of experience cheaper (Nonaka). It will reward experience that has been validated and punish experience that has not (Kahneman and Klein). It will widen gaps wherever judgment is needed to use the knowledge it supplies (Cohen and Levinthal). The cases test this prediction.

3. Case Studies

Evidence labels are applied throughout. Observed marks findings documented in peer-reviewed or primary sources. Claimed marks statements made by the organization itself. Inferred marks this essay's interpretation.

Case 1: Customer Support Agents. Experience Converted Into Shared Knowledge

What existed before? A Fortune 500 software company employed thousands of customer support agents. Performance varied widely. The best agents had learned, through repetition, how to resolve difficult cases and calm frustrated customers. That knowledge sat in their heads and in their conversation logs (Observed: Brynjolfsson, Li and Raymond, 2025).

What was built or changed? The firm introduced a generative AI assistant that suggested responses in real time. The study covered 5,179 agents. The system learned from patterns in successful conversations, including those of top performers (Observed).

What happened afterward? Productivity rose 14 percent on average. Novice and low-skilled agents gained 34 percent. Experienced and highly skilled agents saw minimal impact. The authors report that the tool disseminated the practices of more able workers to less experienced ones (Observed).

What does this reveal? Experience was the raw material. The AI system performed the externalization that the experienced agents had never done themselves (Inferred). The value of their tacit knowledge was real, but it flowed to others. The case shows the first half of the thesis. Explicit knowledge is becoming cheap, and unlinked experience can be extracted by whoever builds the linking system. The experienced worker who does not link their own experience does not keep its premium.

Case 2: BCG Consultants. Knowledge Without a Map of Its Limits

What existed before? Boston Consulting Group consultants are selected for analytical ability and trained in structured problem solving. They represent a population rich in explicit knowledge (Observed: Dell'Acqua et al., 2023).

What was built or changed? Researchers from Harvard Business School and other institutions ran a preregistered experiment with 758 consultants. Some received GPT-4. Tasks were designed to fall either inside or outside the frontier of what AI could do well (Observed).

What happened afterward? Inside the frontier, consultants with AI completed 12.2 percent more tasks, worked 25.1 percent faster, and produced work rated more than 40 percent higher in quality. Lower performers gained 43 percent. Higher performers gained 17 percent. Outside the frontier, consultants using AI were 19 percentage points less likely to reach a correct answer than those without it (Observed).

What does this reveal? Knowledge access is not the constraint. Every participant with AI had the same knowledge source. The difference was knowing where that source could be trusted (Inferred). The authors describe the frontier as "jagged." Tasks that look similar can fall on opposite sides of it. Recognizing which side a task falls on is a judgment built from experience with the domain and with the tool. Knowledge without that map of its limits produced confident error. This is the Kahneman and Klein condition in practice.

Case 3: Kenyan Entrepreneurs. Experience as the Filter

What existed before? Small business owners in Kenya ran farms, restaurants, shops, and service businesses. Their performance varied. Some had built profitable operations. Others struggled (Observed: Otis et al., 2025).

What was built or changed? Researchers gave 640 entrepreneurs access either to a GPT-4 business mentor via WhatsApp or to standard business training guides. The experiment tracked revenues and profits over several months (Observed).

What happened afterward? The average effect was close to zero. That average hid a split. High-performing entrepreneurs improved performance by more than 15 percent. Low performers did roughly 10 percent worse. The researchers found the difference was not in how often participants used the tool. Low performers tended to apply generic advice such as price cuts and more advertising. High performers selected advice that fit their specific circumstances. Co-author David Holtz described the difference as the ability to separate more useful advice from less useful advice, and the knowledge of how to act on it (Observed: Columbia Business School, 2024).

What does this reveal? This case completes the logic of Case 1. In customer support, AI carried experience to novices and narrowed the gap. In entrepreneurship, the task required choosing among options, and AI widened the gap (Inferred). The difference is where judgment sits. When the AI embeds judgment, novices benefit. When the human must supply judgment, experience decides the outcome. Experience functioned as absorptive capacity. It determined whether knowledge helped or harmed.

Case 4: Aramco Metabrain. An Institution Linking Ninety Years of Experience

What existed before? Saudi Aramco has operated for more than ninety years. Over that period it accumulated drilling records, geological surveys, operational data, and cost histories across one of the world's most complex hydrocarbon systems. Much of the judgment built on that record sat with engineers and in archives. The institutional experience was vast. Its accessibility depended on who held it and where it was stored (Inferred from Claimed scale data).

What was built or changed? In March 2024, at the LEAP technology conference in Riyadh, Aramco announced Aramco Metabrain AI. The company described it as a generative AI model with 250 billion parameters, trained on 7 trillion data points drawn from its operating history. Upstream, it was designed to analyze drilling plans, geological data, and historical drilling times and costs, and to recommend well options. Downstream, it was designed to support forecasts for refined products (Claimed: Aramco, reported by AGBI and Offshore Technology, 2024). CEO Amin Nasser said the company's data "gives us an edge in understanding and solving the big challenges we face as a company" (Claimed).

What happened afterward? Aramco has described using AI to reprocess historical seismic data. Ahmad O. Al-Khowaiter, Executive Vice President of Technology and Innovation, stated that the company is "now able to generate new results from old data," with processing that once took months now taking hours (Claimed: Aramco Europe, 2025). Independent evaluations of Metabrain's operational outcomes have not been published. This is an empirical gap and is named as such.

What does this reveal? The case shows linkage at institutional scale. Aramco's competitive position does not rest on geological knowledge alone. Much of that knowledge is published and widely taught. The distinctive asset is ninety years of accumulated operating experience in specific reservoirs (Inferred). Metabrain represents an attempt to perform Nonaka's externalization on an entire institution's history. It converts accumulated experience into a queryable form and links it to present decisions. The phrase "new results from old data" is the Linkage Model in one sentence. The experience already existed. Its value increased when it was linked to new analytical capability. The case also reflects a broader pattern in Saudi transformation. Advancement builds on institutional memory rather than replacing it.

4. Synthesis Framework: The Linkage Model

The four cases point to one mechanism. Experience and knowledge are complements, not substitutes. Each is weak alone. Their combination produces advantage. AI changes the cost of combining them.

The Linkage Model has three parts: a map of positions, a set of four moves, and one governing condition.

4.1 The Linkage Map

Two dimensions define a professional's or an institution's position.

The first is experience depth. This is the volume of validated practice in a domain.

The second is knowledge linkage. This is the degree to which that practice has been connected to explicit concepts, evidence, and transferable form.

The combination produces four positions.

Silent Mastery. Deep experience that has never been articulated. The practitioner knows more than they can tell. Their value is real but local. It cannot be taught, audited, or recognized beyond direct observers. In the AI era, Silent Mastery is exposed. Others can extract its patterns, as in Case 1, while the practitioner captures little of the premium.

Borrowed Fluency. Rich access to knowledge with little practice behind it. This position is expanding fastest. AI allows anyone to speak fluently about any domain. The BCG case shows its weakness. Fluency performs well inside the frontier and fails outside it. It cannot tell the difference.

Drift. Neither deep experience nor linked knowledge. AI can lift performance here on routine tasks where the system embeds the judgment. It cannot supply direction where judgment is required, as the Kenya case shows.

Linked Judgment. Deep experience connected to the right knowledge. This is the position that captured the largest AI gains in Case 3. It is the position Aramco is attempting to build institutionally in Case 4. Linked Judgment is the new competitive edge.

4.2 The Four Linkage Moves

Movement from Silent Mastery to Linked Judgment happens through four moves. Each was historically slow. AI can now assist each one.

Move 1: Surface. Bring tacit patterns into language. The practitioner describes decisions, exceptions, and instincts. AI acts as an interviewer. It asks why a choice was made, what would have changed it, and what a novice would miss. This is Nonaka's externalization made conversational.

Move 2: Name. Connect surfaced patterns to existing concepts. A practitioner's private rule often matches a known principle in economics, psychology, or operations. AI can propose those matches and supply the literature. Naming turns a personal habit into a recognizable idea. It also reveals where the practitioner's pattern departs from established knowledge, which is where original insight often sits.

Move 3: Calibrate. Test experience against evidence. This is the Kahneman and Klein step. The practitioner asks whether their domain offered regular patterns and honest feedback. AI can surface counter-evidence, base rates, and cases where the pattern failed. Calibration separates valid expertise from confident habit. It is the step most often skipped, and the one that makes linkage "right."

Move 4: Transmit. Convert linked experience into a form others can use. This may be a framework, a method, a decision checklist, a training program, or a published essay. Transmission is where experience becomes visible and attributable. It is also where the practitioner, rather than a third party, captures the value.

4.3 The Governing Condition: Validity Before Value

The Linkage Model has one condition. Experience only earns linkage if it was formed in a valid environment. Where the domain was regular and feedback was honest, linkage amplifies real skill. Where the domain was irregular or feedback was absent, linkage should correct experience rather than amplify it.

This condition protects the model from a predictable misuse. Seniority is not the same as expertise. Years in a role do not guarantee learning. The Calibrate move exists to enforce this condition.

4.4 How the Model Connects to the Prince Researcher Corpus

The Linkage Model sits upstream of the Achievement-Reputation Gap series. That series explains why accomplished people remain unrecognized. Its Evidence Gap describes achievement that is poorly documented. The Linkage Model explains one reason documentation fails. Experience that has never been surfaced or named cannot be documented. Silent Mastery is the cognitive precondition of the Evidence Gap.

The model also maps onto the four-dimension instrument. Transmission produces Recognition. Calibration produces Credibility. Naming creates Coherence between a practitioner's work and a recognized domain. Sustained Linked Judgment accumulates into Standing.

4.5 Applications Beyond This Essay

The model applies at three levels.

Individuals. A senior professional can audit their position on the Linkage Map. The practical question is simple. What do I know from practice that I have never written down, tested, or taught?

Institutions. Organizations facing retirement waves or leadership transitions hold large reserves of Silent Mastery. The Four Moves describe a knowledge-capture program that preserves experience before it leaves. This connects to Strategic Memory, one of Prince Researcher's core domains.

Nations. Economies that invested heavily in formal education now hold large populations with deep sector experience. The Linkage Model suggests that the next productivity gain may come less from acquiring new knowledge and more from linking existing experience to it.

4.6 Boundaries and Objections

Three objections test the model. Each sharpens it.

Objection 1: AI will let people skip experience entirely. If AI can supply the patterns of experts, novices may never need to build their own. The evidence does not support this for judgment-heavy work. The Kenya case shows that access to expert-quality advice did not replace the judgment to choose among it. There is also a formation risk. If novices rely on AI during the years when experience normally forms, they may reach seniority with Borrowed Fluency rather than Linked Judgment. The model therefore implies a design principle for organizations. AI should support the formation of experience, not substitute for it. Novices still need exposure to real decisions and honest feedback.

Objection 2: Experience becomes obsolete faster in a changing world. This is partly correct. Experience formed in one environment may mislead in another. The Governing Condition addresses this directly. When the environment changes, experience must be recalibrated. The Calibrate move is not a one-time step. It is a recurring discipline. Experience that is never recalibrated moves from Linked Judgment back toward confident habit.

Objection 3: Extraction benefits the organization, so why should the individual care? In Case 1, the firm and its novice agents gained. The experienced agents who supplied the patterns saw minimal benefit. From the organization's view, this is an efficient outcome. From the individual's view, it is a transfer of value without attribution. This connects the Linkage Model to the corpus concept of Attribution Capture. When a system links a person's experience on their behalf, the system often captures the credit. The individual who performs the Four Moves personally keeps authorship of what they know.

These boundaries define the model's scope. It does not claim that experience always beats knowledge. It claims that validated experience, linked by its owner to the right knowledge, is the most defensible source of advantage when knowledge itself is abundant.

5. Conclusion

The belief that knowledge is the competitive edge belonged to a period when knowledge was scarce. That period is ending. AI now supplies explicit knowledge to anyone who asks.

What AI cannot supply is the judgment to know which knowledge applies. The field evidence is consistent on this point. Where judgment is embedded in the system, novices catch up. Where judgment must come from the person, experience decides the outcome.

Experience has always carried this value. It rarely captured it. It stayed tacit, local, and unrecognized. The cost of linking it to knowledge was too high for most people to pay. AI has lowered that cost. Surfacing, naming, calibrating, and transmitting experience can now happen in hours rather than years.

This creates a choice. Experienced professionals and institutions can link their experience themselves and keep its premium. Or they can leave it unlinked and watch others extract it. Knowledge tells you what is possible. Experience tells you what is true. The advantage now belongs to those who can connect the two.

6. References and Further Reading

Academic

• Autor, D. H. (2015). Why are there still so many jobs? The history and future of workplace automation. Journal of Economic Perspectives, 29(3), 3–30.
• Brynjolfsson, E., Li, D., and Raymond, L. (2025). Generative AI at work. The Quarterly Journal of Economics, 140(2), 889–942. Earlier version: NBER Working Paper 31161 (2023).
• Cohen, W. M., and Levinthal, D. A. (1990). Absorptive capacity: A new perspective on learning and innovation. Administrative Science Quarterly, 35(1), 128–152.
• Dell'Acqua, F., McFowland III, E., Mollick, E., Lifshitz-Assaf, H., Kellogg, K., Rajendran, S., Krayer, L., Candelon, F., and Lakhani, K. R. (2023). Navigating the jagged technological frontier: Field experimental evidence of the effects of AI on knowledge worker productivity and quality. Harvard Business School Working Paper 24-013. SSRN 4573321.
• Kahneman, D., and Klein, G. (2009). Conditions for intuitive expertise: A failure to disagree. American Psychologist, 64(6), 515–526.
• Nonaka, I. (1994). A dynamic theory of organizational knowledge creation. Organization Science, 5(1), 14–37.
• Nonaka, I., and Takeuchi, H. (1995). The Knowledge-Creating Company. Oxford University Press.
• Otis, N., Clarke, R., Delecourt, S., Holtz, D., and Koning, R. (2025). The uneven impact of generative AI on entrepreneurial performance: Evidence from a field experiment in Kenya. Management Science. https://doi.org/10.1287/mnsc.2024.06909
• Polanyi, M. (1966). The Tacit Dimension. Doubleday. Reissued University of Chicago Press, 2009.
• Schön, D. A. (1983). The Reflective Practitioner: How Professionals Think in Action. Basic Books.

Institutional

• Aramco Europe. (2025). How AI helps Aramco turn old data into new opportunity. https://europe.aramco.com/en/news-media/news/2025/aramco-deploys-ai
• Aramco. Building the AI future. https://www.aramco.com/en/what-we-do/energy-innovation/digitalization/building-the-ai-future
• Columbia Business School. (2024). AI is widening the gap between high and low performers. https://business.columbia.edu/insights/ai-widening-gap-between-high-and-low-performers

Journalism

• AGBI. (2024, March). Aramco launches world's first industrial AI model. https://www.agbi.com/ai/2024/03/aramco-launches-world-first-industrial-ai-model/
• Offshore Technology. (2024). Saudi Aramco unveils industry's first generative AI model. https://www.offshore-technology.com/news/saudi-aramco-unveils-industry-first-generative-ai-model/

© Strategic Essay — Prince Researcher | For Academic and Editorial Use