I’ve been on a bit of a hiatus from Superadditive.co. Vacation, and then all the work I didn’t get done during vacation, and then the work that piled up while I was doing that work...
TL;DR.
Knowledge exists in two forms: explicit (can be written down) and tacit (embedded in people and relationships). While explicit knowledge survives in books and databases, tacit knowledge lives only in our minds, hands and our connections to others. This includes the intuition, judgment, and craft wisdom that separates masters from students. Like family stories that exist in just one relative's memory, this knowledge can disappear when relationships break or experts retire without apprentices. The Tasmanians once knew how to make fire and craft bone tools, but isolation slowly erased these skills over generations. Today, as aging populations carry specialized knowledge, institutions dismantle, and AI replaces collaborative networks, we face a similar risk of forgetting.
Tracking Down Vanishing Knowledge
When I was a teenager, I was obsessed with genealogy. During family trips to India, I would track down distant relatives and ask them hundreds of questions about our ancestors. Names, stories, the origins of our family, what people did for a living, what they ate, and who I should talk to next. I was always on the lookout for little treasures: a photograph tucked away in piles of documents in an attic, a story that lived in only one person's mind, a historical document that provided a missing link to some distant branch of the family in a faraway place. I always carried a notebook with me (which I still have), where I would document all the little facts I learned, fearing that I too would forget these things one day.

As I’ve gotten older, and many of those same relatives have now passed on, what strikes me most is how fragile this knowledge was to begin with. These relatives were often the sole keepers of entire branches of our family history. When they died, this knowledge would vanish too.1
The Tasmanians
This fragility of knowledge turns out to be a much larger human phenomenon. My co-author and friend John-Paul Ferguson, when we were writing The Lives and Deaths of Jobs, shared what is now one of my favorite articles titled “Ten Thousand Years of Solitude” by Jared Diamond. The article is about collective forgetting among the Tasmanian people.
The Tasmanians had advanced technologies 10,000 years ago. They had bone tools for making needles and warm clothing, fishing techniques that provided 10 percent of their diet, and fire-making skills.
However, rising seas cut them off from the mainland, trapping just 5,000 people on the island.
Slowly, knowledge began to vanish.
By the time Europeans arrived, the Tasmanians had stopped eating fish. Their bone tools had disappeared around 3,500 years ago. They had lost the ability to make fire and instead carried firebrands, relighting them from neighbors when they went out.
Their knowledge had slipped away, one generation at a time, with no neighboring tribe to relearn from and no way to recover what was lost.
The Tasmanian case offers something that social scientists are always seeking: a controlled comparison. In this case, we know what technologies they started with and what they had lost by the end, giving us a much clearer picture of how knowledge vanishes over time.
Diamond's account is so compelling because it highlights several specific mechanisms behind knowledge decay: population thresholds below which skills can't be sustained, isolation that prevents relearning, and how useful practices can drift into taboos, ultimately leading to the disappearance of productive capabilities.
This prospect of societal forgetting seems impossible in the information age: everything, every click, is stored and shared worldwide in fractions of a second.
Yet, we face significant forces today that could make it happen: aging populations carrying specialized knowledge that dies with them, the deliberate dismantling of complex institutions that helped create the prosperity of the last 150 years, market forces and competitive pressures that make rare knowledge more valuable while also limiting its spread, and now, potentially, hierarchical corporate-owned AI systems replacing horizontal networks of specialists.
What is knowledge?
The concept of knowledge has been analyzed by philosophers for thousands of years. Plato's definition of “justified true beliefs” often serves as a helpful starting point2, and many philosophers have built upon and refined (and disputed!) this view over the centuries. (And if you’ve been around philosophers, you know that every word is scrutinized: What does justified mean? What counts as true? What is a belief?3)
From the perspective of an organizational researcher, the most significant modern advancement in our theory of knowledge comes from Michael Polanyi, a chemist-turned-philosopher who spent his later years grappling with a fundamental question: “why we can know more than we can tell?” I finally read his treatise, The Tacit Dimension, a few years ago and think about it nearly all the time.
Polanyi made a crucial distinction between two fundamentally different types of knowledge.
Explicit knowledge, which can be written down and transmitted through books, articles, or instruction. Think of the Pythagorean Theorem: Any 7th grader can memorize a² + b² = c² and apply it to solve problems.
Tacit knowledge is something entirely different in nature. It is knowledge that resides in our minds and hands, in both our practice and our intuition. It's what separates someone who can recite the Pythagorean Theorem from a carpenter who truly knows how to square a corner, or an architect who understands proportion and spatial harmony in ways that shape every design decision. And of course, a mathematician who has spent decades working with geometric proofs. In all these cases, the formula itself is just the beginning of what it means to truly know something.
This distinction between explicit knowledge and tacit knowledge is everywhere. In my field, there's a marked difference between a first-year student who has taken a few econometrics classes and a theory seminar and someone with decades of experience who has an intuitive feel for data, models, and their connection to the structure of a scientific theory and its predictions.
Where Tacit Knowledge Lives
Where these types of knowledge are located also matters.
Explicit and codifiable knowledge can be stored for long, uninterrupted periods, in libraries, databases, and, now, the cloud4.
Tacit knowledge is far more delicate.
It primarily exists in two places: within people (not just their minds, but also in their hands, feet, and muscles) and in relationships among people.
When scholars say that knowledge lives “in people,” what they mean is that it is embedded in their practice, muscle memory, and judgment.
The master sushi chef, Jiro Ono, possesses decades of tacit knowledge about rice temperature, fish quality, and timing that no cookbook can capture. Moreover, he probably can’t convey all the things he “knows.”
Similarly, the sarod masters in Amjad Ali Khan's lineage, both past and future, carry knowledge about instrument construction and playing techniques that have been refined through generations of practice. This produces a sound that takes decades to perfect in one person and centuries to reach its current state.
But perhaps more importantly, tacit knowledge lives in relationships, in the social fabric that connects people.
The idea that knowledge resides “in networks” is central to many organizational theories (and, obviously, in Sociology and Anthropology too).
March and Simon, in their 1958 masterpiece “Organizations,” were among the first to argue that organizations adapt through routines or programs that emerge from repeated interactions between people with bounded rationality.
Building on this foundation, Nelson and Winter's “An Evolutionary Theory of Economic Change” (1982) provided the most systematic treatment of organizational routines. Their key argument was that knowledge does not reside solely in individual minds, but in the practiced interactions between people engaged in interdependent work. (For a thorough review of the literature on knowledge and knowledge transfer, the work of Linda Argote is a must-read.)
For example, a surgical team's tacit knowledge emerges from their repeated collaboration, their unspoken communication, and their ability to anticipate each other's needs.
This relational dimension explains why knowledge loss in organizations and society can be so sudden, and so devastating.
When relationships dissolve, when teams disband, when master craftspeople retire without apprentices, when communities scatter, the tacit knowledge embedded in those connections can vanish with them. The knowledge held in human networks requires constant activation and use to survive.
Hierarchical vs. Network Knowledge Systems
We can consider knowledge systems at two ends of the spectrum: hierarchical and network-based. Recognizing this difference helps explain why some knowledge is more prone to forgetting.
Hierarchical systems follow a top-down, one-to-many model.
For instance, textbooks, online courses, and AI systems are quite hierarchical in nature. These systems excel at transmitting explicit information efficiently and at scale. A single textbook can teach millions of students; it’s a 1:m system.
An AI system can instantly provide answers to billions of queries. The efficiency and scale are remarkable, but these systems can only transmit what can be explicitly documented or programmed (it’s an interesting question as to what a non-documented but robust statistical relationship between concepts that emerges in these models counts as explicit or tacit?).
Network systems operate through peer-to-peer (1:1), many-to-many (m:m) relationships. For instance, research laboratories, craft workshops, or apprenticeship programs are network-type systems. In these systems, knowledge flows through interaction, observation, co-creation, and practice. A master craftsperson does not just tell an apprentice how to shape wood; they work side by side, sharing countless little adjustments, corrections, and insights that accumulate into knowledge. These networks are slow and costly to maintain, requiring sustained relationships.5
The critical difference lies in what each system can transmit:
Hierarchies excel at distributing codified knowledge but cannot convey the intuition, judgment, and embedded know-how that constitute tacit knowledge. You can learn music theory and chord progressions from a YouTube video, but you cannot learn the timing, feel, and musical instincts required to actually jam with other musicians this way.
Networks, by contrast, are specifically designed for this kind of tacit knowledge transfer through repeated interaction and shared work.
This distinction in how knowledge is stored highlights a vulnerability: once a knowledge network collapses, its tacit core can vanish forever.
When the last master craftsman retires without an apprentice (or is laid off when a private equity firm takes over), or when research teams disperse due to funding cuts, or when communities of practice dissolve because no one wants to organize them, the tacit knowledge embedded in those relationships disappears.
No amount of documentation or AI can recover what was never codified in the first place.
The Economic Value of Tacit Knowledge
Tacit knowledge is interesting from both philosophical and sociological perspectives, but it is also the foundation of durable competitive advantage in many industries.
Consider ASML's extreme ultraviolet (EUV) lithography machines, which enable the production of the world's most advanced computer chips (e.g., those used in AI). ASML’s dominance in this space provides a textbook example of how tacit knowledge can create insurmountable competitive moats even in high-tech manufacturing.
The key ASML technology consists of machines (often costing up to $370 million) that manufacture transistors through “printing.” The precision is incredible: the machines can print transistors that are the length of five DNA strands (or approximately ten thousand times narrower than a human hair, according to the article I link to). Given how delicate and nuanced this production technology is, it is not surprising that it took nearly two decades for this technology to hit the market (with billions in investments). What was learned over these nearly two decades? Lots of tacit knowledge that came from solving specific problems, both big and little ones.
Furthermore, these problems are not solved solely by ASML’s scientists or engineers, but are solved collaboratively with thousands (yes, thousands) of suppliers, whose IP constitutes nearly 85% of all the IP (this linked article is superb!) that goes into the final product. Moreover, its buyers (Samsung, Intel, TSMC) also work with ASML to solve specific problems that they face. The knowledge is embedded in very bespoke networks inside ASML, among its suppliers, and with its buyers. In an interesting CNBC article, the CEO talks about the embeddedness of ASML’s knowledge, as per Granovetter:
“We’re unique to some of our customers, and some of our supplies are unique to us,” Wennink said. “And those almost symbiotic relationships, some people say, are worse than being married because you cannot divorce.”
That is embeddedness.
This knowledge is so subtle and tacit that ASML also embeds customer support engineers with their machines (the full WSJ article is worth a read because it highlights the surprising solutions that come from tacit knowledge—e.g., using two Home Depot buckets to fix a multi-hundred-million-dollar machine).6 These engineers are trained for over a year on all the subtleties of these incredibly complex machines. So if problems arise, they can be solved on the spot.
As one might imagine, this knowledge is remarkably valuable, and people have attempted to steal it or replicate it in other ways by investing billions in the development of similar technology. Thus far, to no avail.
There is, however, a flipside to this economically valuable trove of tacit knowledge that ASML's dominance rests on. If key holders of unique, tacit knowledge leave, the network begins to fray, or reckless cost-cutting removes important but invisible nodes, ASML may lose critical knowledge; more importantly, the world may regress in terms of its technological capabilities because the knowledge is embodied in people and relationships.
Examples of such technological forgetting are plentiful. For example, the loss of knowledge about Roman concrete as the Roman empire disintegrated is a notable example of a significant and superior technology that was lost until modern science was able to reverse-engineer it (described here in a more straightforward manner). Similarly, Roman knowledge on how to build domes was also lost until the Renaissance, when it was revived through the ingenuity of Brunelleschi.
Likewise, the gradual decline in the skill of Ukiyo-e woodblock printing in Japan was driven by multiple factors, including the emergence of competing technologies, public disinterest, and government policies that discouraged its practice (see the section on its decline on the Wikipedia page). Yet, the art form survived—though not in its original form—due to unexpected reasons, particularly interest from Western artists such as Whistler, van Gogh, and Cassatt. In recent times, David Bull, a Canadian printmaker who has been based in Japan for several decades, has also worked to preserve the tacit knowledge of this endangered art form (see, for instance, this three hour video where he makes a woodblock print from start to finish).
If you watch David Bull’s hands in this video you can understand Polanyi’s refrain: “we can know more than we can tell.”
The Economic Paradox of Tacit Knowledge
When I read about the ASML example, it suggests to me that we should not be too enamored by the competitive implications of tacit knowledge held by a single, dominant firm, nor too nostalgic about the unique wisdom of the experienced craftsperson.
This is because a fundamental economic tension is at play.
Tacit knowledge can be incredibly valuable both socially and economically. As a result, sharing that knowledge creates competition and reduces the holder's ability to capture value. That is, the subtle skills, judgment calls, and hard-won insights that make tacit knowledge valuable become less unique when it is shared with others. This economic dynamic creates powerful incentives for knowledge hoarding. And, in turn, this hoarding behavior is precisely what ensures the knowledge’s eventual disappearance.7
I’m not sure if this paradox has been supported with data, but if it is, it would be quite tragic.
However, today the forces that threaten tacit knowledge extend far beyond the endogenous economic incentives described above.
Our Tasmanian Dilemma: An Era of Mass Forgetting?
What we're witnessing today may be a perfect storm of endogenous (see above) and exogenous forces at play, which can cause societies to forget important, productive knowledge.
Here are just a few:
Dismantling of Core Institutions
Perhaps the most urgent issue is the large-scale dismantling of core public institutions. For example, agencies like the NIH are consolidating from 27 institutes to 8 with a 40% budget cut, ending over 800 research projects, and breaking apart 70-year-old interconnected research networks. These institutions are webs of tacit knowledge, and losing one node can be devastating; losing hundreds or even thousands of crucial nodes in the knowledge network may be nearly impossible to recover from. It is not just the NIH; dozens of institutions are being dismantled, and tacit knowledge — embedded in people’s minds, practices, and their relationships — may slowly be forgotten. This is perhaps the least of it, as the ripple effects of this institutional loss will be felt more widely.
The Aging Workforce
Another pressing issue is the problem of aging workforces in developing economies (my colleagues Yoko Shibuya, Ines Black, and Maria Zhu, and I) have an integrative review titled “The Aging Firm” (happy to share it if you are interested) and an early-stage empirical paper examining the firm-specific factors that lead to skewed (older) age distributions inside firms. A key tension a firm faces is getting work done now (and competing!) and passing on that knowledge to the next generations.
Japan, for instance, has seen a marked inversion of its population pyramid which threatens the loss of important production knowledge in the economy.
The United States hasn’t inverted yet, but its population pyramid is now more rectangular. We have already seen a decline in manufacturing throughout the US, and it’s not clear that policy or financial incentives can bring it back quickly, as the TSMC experience in Arizona demonstrates.
As the world ages, and the torch of tacit knowledge isn’t passed down, we may witness a monumental scale of forgetting.
Artificial Intelligence and the Hierarchical Knowledge Systems
Finally, the long-term effects of AI on knowledge worker productivity raise serious concerns that go beyond just short-term efficiency gains documented in some research studies.
Large language models excel at generating novel combinations from codified knowledge. However, these technologies raise important issues for organizations and the economy as a whole. These technologies risk replacing the natural growth of young talent and the networks of practice that have traditionally linked both explicit and tacit knowledge sharing. For example, you can ask ChatGPT to explain a mathematical theorem, but it cannot teach you how to think like a mathematician, build mathematical intuition, or identify which problems are worth pursuing. The widespread use of these systems may accelerate the erosion of human networks that store knowledge.
That is, when we can instantly access synthesized information, we may lose the slower but essential processes through which expertise is cultivated through study, observation, and human interaction.
Enrique Ide's compelling new working paper, “Automation, AI, and the Intergenerational Transmission of Knowledge,” explores this tension and should be essential reading for anyone grappling with how knowledge flows between generations in an age of artificial intelligence.
What Can Be Done to Prevent Forgetting?
While these are quite significant forces and most of these are things that individuals cannot solve on their own, there is probably a lot we can do to preserve valuable tacit knowledge in our own backyards.
At the organizational level, we can all start being more thoughtful about who holds what knowledge within our organizations and what networks store the relational knowledge that enables things to run smoothly.
For transmitting crucial, productive, tacit knowledge, we should think about how to create apprenticeship programs that go beyond traditional mentoring (or classroom education). Apprenticeship programs should be designed so experienced practitioners work alongside newcomers on real problems rather than sage-on-stage lecturing.
Similarly, companies can redesign incentives to reward knowledge sharing rather than hoarding (check out research by Alex Oettl on helpfulness here and here, as well as the brilliant work on awards and social contributions by Jana Gallus).
We should consider promotion criteria that value the development of others.
There is also much we can do to preserve the hard-won knowledge we have acquired. We too should focus on mentoring the next generation in our fields, and contribute on a 1:1 basis in our professional communities.
Most importantly, we should probably recognize that preserving knowledge is not just about our legacies, but about maintaining the collective tacit knowledge that enables societal progress.
When I reflect on my fascination with genealogy, one thing is clear: each of us may be the sole keeper of some valuable knowledge.
We have a responsibility to ensure it doesn't die with us.
Herbert Simon once observed that the fog of the past is as thick and impenetrable as the fog of the future.
It’s useful to note that the Greeks were not the only ones with developed ideas of knowledge. Indian, Chinese, and Muslim philosophers also engaged in profound discussions and debates on this topic.
I was a philosophy major.
Though the undecipherability of the Indus Valley script provides an example of the sometimes unrecoverability of codified knowledge.
In fact, our PhD program at Duke is explicitly designed around an apprenticeship model, where most learning occurs through writing papers. Classes are few and taken either early to establish core skills/frameworks, or later as needs arise in solving specific research problems.
This also reminds me of an anecdote I read, I think in the Philosopher of Palo Alto, about how Xerox embedded ethnographers with their repair people to learn how they fixed problems. For instance, while a copy machine’s manual provided a very complex set of directions to figure out the source of a jam, the experienced repair person could go through the garbage can next to the copier and figure this out based on how the last discarded copies looked.
One can see this knowledge hoarding in a variety of contexts, from farmers hoarding advanced agricultural technology, in corporate contexts where experts (technical and others) hoard knowledge,