Judgment is the rarest organizational resource

You started as a physicist. How does someone trained in statistical mechanics end up becoming one of the world's most influential economists?

Hidalgo: The path was more natural than it might appear from the outside. The subfield of physics I specialized in is complex systems — the same discipline that gave us chaos theory, fractal patterns, and thermodynamics. By the time I was studying, complex systems were moving into networks. The 1990s were the decade of the internet, of nodes and links and webs of connection. Networks were as exciting back then as AI is today.

My PhD supervisor, Laszlo Barabási, had been applying network science to biology and social systems. What interested me most was economic development — specifically, why some societies seem to be collectively smarter than others. Why can some economies produce certain goods and others cannot? Network science gave me a way to model that.

What was the pivotal insight that shaped your research agenda?

Hidalgo: The realization that development is not a process of growth — it is a process of structural change. The standard economic model treats development as countries accumulating more of the same: more capital, more labor, more productivity. But that misses something fundamental. What changes when societies develop is the kind of things they can do, not just the quantity. And understanding that required a new lens — one that treated the economy as a network of interconnected capabilities, like an alphabet that some societies have mastered more fully than others.

That insight eventually became the economic complexity framework, and then the book, “The Infinite Alphabet.” The economy is, at some level, a giant game of Scrabble. Each country holds a set of letters — capabilities, skills, knowledge — and the products and services it can generate are the words it can form. Some countries have a rich, diverse alphabet. Others are stuck with a limited set of tiles. The goal of development policy should be to expand the alphabet, not just to play bigger words with the ones you already have.

Your book proposes that knowledge — not capital, not technology, not institutions — is the fundamental driver of economic growth. What do you mean by knowledge, and what are its governing principles?

Hidalgo: Knowledge is a word we use casually, but it covers genuinely distinct things. The educational psychologist Benjamin Bloom gave us a useful framework: factual knowledge — the clues at a crime scene; conceptual knowledge — the detective's theory of who did it and why; and procedural knowledge — knowing how to collect and analyze the evidence. These are not the same, and conflating them causes confusion about what AI can and cannot replace.

But the dimension of knowledge I find most important for understanding organizations and economies is the distinction between explicit and tacit knowledge. Explicit knowledge is codifiable — it can be written in a book, taught in a lecture, transmitted through language. Tacit knowledge is the kind you can only acquire by doing, by practicing alongside people who already have it. Michael Jordan could deliver a lecture on basketball. You would not be much better on the court afterward. You need to be on the court with the best, absorbing through channels that are not simply verbal.

The principles I propose in the book — and they are more like organizing categories than Newtonian laws — address three dimensions: the speed at which knowledge grows, the way it diffuses through physical space and across activities, and the mechanisms by which it generates economic value.

One of those counterintuitive features is that knowledge can decay. Can you explain that?

Hidalgo: This is perhaps the most important — and most neglected — principle. We tend to think of knowledge as something that accumulates passively, like books on a shelf. But productive knowledge, the kind that drives the economy, is more like a muscle. If you stop exercising it, it weakens.

The moment a society or an organization stops performing a complex activity — stops building nuclear power plants, stops running major infrastructure projects, stops manufacturing a certain class of goods — the knowledge required to do that thing starts to atrophy. Not dramatically, not all at once, but steadily and invisibly. Ten to fifteen years later, you discover that you cannot reconstruct the capability because the team that built the last one has scattered or retired, and no one who remains has the lived experience of doing it.

This has profound implications for organizations thinking about what to automate, what to offshore, and what to stop doing. Every one of those decisions is also a decision about which knowledge you are willing to lose.

AI is changing the knowledge landscape rapidly. How does your framework help leaders understand what AI can and cannot do?

Hidalgo: The key concept is what I call the jaggedness of AI knowledge. AI is not uniformly capable or uniformly limited — it is extraordinarily good at some things and genuinely absent from others. That jaggedness matters enormously when you try to integrate AI into real organizational processes, because the bottlenecks often lie exactly in the places where AI is weakest.

Think of it this way: introducing AI into an organization is like receiving a large supply of certain letters in our Scrabble game — many R's, many T's, many A's — but none of the letters that were already missing. Processes that required only those abundant letters will accelerate dramatically. But processes that depend on the scarce letters will still be bottlenecked — and now those bottlenecks will be even more visible because everything around them is moving faster.

And what about the complementarity between human expertise and AI capability?

Hidalgo: Complementarity is everything. Someone using an advanced coding tool who has deep experience in software architecture — who understands what questions to ask, how to structure a system, where the failure modes are — will produce something qualitatively different from someone with no computer science background. The tool gives you what you put into it. Human expertise shapes what you put in and how you evaluate what comes out.

AI is an additional instrument for exploration, like a telescope that lets you survey codified knowledge at scale. But the quality of the exploration depends on the skill of the explorer. In areas where you have deep expertise, you will catch the errors — the plausible-sounding but subtly wrong claims, the citations to papers that do not quite say what AI attributes to them. In areas where you lack expertise, you are navigating with a powerful but unreliable instrument and no map. That is a dangerous combination.

Your book contrasts two very different approaches to building innovation ecosystems. What happened in Ecuador, and what did China do differently?

Hidalgo: Yachay is one of the most instructive cautionary tales in the history of innovation policy. The Ecuadorian government invested approximately one billion dollars — about one percent of the country's GDP — to build a city of knowledge from scratch on a greenfield site several hours north of Quito. They wanted to attract Google, Huawei, biotech companies. They wanted to excel at everything simultaneously within a decade.

That is not how knowledge works. Knowledge loves density. It propagates most effectively in environments where the relevant people, institutions, and activities are already concentrated — in downtown Tokyo, in San Francisco, in Zhongguancun in Beijing. It does not spontaneously emerge from infrastructure investment in remote locations, no matter how well-intentioned or well-funded.

Zhongguancun, by contrast, worked precisely because it built on existing density. Beijing already had Peking University, Tsinghua University, the Chinese Academy of Sciences, and decades of accumulated scientific expertise. What the government added was targeted: a single street segment they called the Avenue of Entrepreneurs, guiding funds that blended public and private capital into venture financing, and a structure that created capitalization dynamics rather than simple government spending. The result was self-reinforcing — Lenovo came from there, Baidu came from there, and the model was eventually replicated across China as the country made its deliberate shift from mass manufacturing to mass innovation.

If you were advising a Fortune 50 Chief Innovation Officer today, what would you tell them?

Hidalgo: Three things, in order of importance.

First, you are a victim of your own success, and you need to make peace with that. Every large organization is organized around its current dominant products. The processes, the incentive structures, the culture, the metrics — all of it is optimized to defend and extend what is already working. That is rational and necessary. But the future will not arrive through your current dominant product. It will arrive through something adjacent, something that looks inferior at first, something that your best people will be too busy to pay attention to.

The Xerox PARC story is not unusual. The technology for the personal computer was developed inside Xerox, by brilliant people, while everyone who mattered was focused on selling more toner for the copier division. IBM's Project Chess was a better model: take a small team, physically relocate them, free them from the constraints of the main bureaucracy, and give them the explicit permission to operate differently. That is what produced the PC.

Second, watch the bottom of your market obsessively. The disruptions that kill incumbents almost never arrive at the top. They arrive at the bottom — cheaper, simpler, often inferior on the dimensions you are optimizing for, but improving fast. Toyota entered the American market with inexpensive cars that serious buyers did not take seriously. Today, BYD is building electric sports cars at forty thousand euros that are genuinely competitive with vehicles costing ten times more.

Third, someone has to cut the cake. Deliberation is necessary but at some point someone must decide. The ability to make that decision, and to make it well, is what we call judgment. And judgment is the rarest and most valuable organizational resource in an environment of uncertainty.

You have studied innovation successes and failures extensively. What is your honest assessment of why so many AI implementations fail?

Hidalgo: Part of it is simply evolutionary reality, and I think people are too hard on themselves and on AI about this. We are living through a period of extremely high entry rates into AI development and deployment. Most of those entries will fail. That is how evolution works — it is inherently wasteful. You need many trials to get the few successes that survive, replicate, and eventually dominate. The same thing happened during the first wave of internet companies in 1997 and 1998. We remember Amazon and Google. We have forgotten the thousands of companies that attempted the same thing and disappeared.

That said, there are structural reasons why AI implementations fail that are not just evolutionary noise. The most important is the misunderstanding of context. Organizations approach AI as if it can simply replace a task that was already clearly defined and well-understood. But in most real organizations, the most valuable processes are not clearly defined and well-understood — they are embedded in tacit knowledge that nobody has ever made explicit. The context is the tacit knowledge.

The organizations that succeed are the ones that do the archaeology first — that invest the time to surface what people actually do, in practice, including all the informal knowledge, the workarounds, the patterns that experienced workers cannot fully articulate but that they perform every day. Only when that knowledge is visible can you make intelligent decisions about what to automate, what to augment, and what to protect.

And scarcity — does it help or hurt?

Hidalgo: Scarcity combined with skill is a powerful advantage as it enforces ownership. Budget constraints force a quality of attention and ownership that abundant resources often undermine. When you have a large budget, responsibility diffuses — the manager delegates, the next person hires, the next person subcontracts, and eventually nobody truly owns the problem. The people closest to the work have the least resources and authority; the people with the most authority are the furthest from the work.

With constrained resources, a small team of genuinely skilled people who own the problem completely will consistently outperform a larger, better-funded effort that has distributed ownership across many functions and meetings. This is not always true — scarcity without skill is just poverty. But in organizations that have the skill, budget constraints often produce the focused, personally-invested team dynamic to solve hard problems.