Albert-László Barabási

In the [YOU] on AI Field Guide

[YOU] on AI is pervaded by the tension between capability equalization and the persistence of structural inequality—the observation that AI makes more people able to build without automatically making more people able to succeed. Barabási supplies the mathematical substrate for this tension. The book’s account of Bob Dylan is illuminating in network-science terms: Dylan was not merely a better musician than a bedroom performer. He was a hub—a structurally different kind of node whose removal would reshape the topology of an entire network. AI tools do not change this. A musician in rural Mississippi who uses AI to produce a major-label-quality album has gained capability. She has not gained playlist curators, concert promoters, or music journalists. The distribution network remains governed by preferential attachment.

The Death Cross—the trillion-dollar repricing of software companies that the cycle documents in early 2026—is, in Barabási’s terms, a phase transition: the crossing of a critical threshold at which the cost of code production drops below the level that sustained the old network topology, forcing a rapid reorganization around a new structural principle. The pre-transition hubs were defined by the code they had produced. The post-transition hubs are defined by their ecosystems—the accumulated networks of users, data, integrations, and institutional relationships that surround the code and remain valuable precisely because the code barrier that once protected weaker competitors has dissolved.

Barabási’s 2025 concept of human-AI coevolution adds the temporal dimension the static analysis of networks cannot capture. The cycle’s builders do not merely use AI tools; they coevolve with them. Their cognitive habits, their ways of framing problems, their intuitions about what is worth building—all are continuously shaped by the tools they use, which are simultaneously shaped by the data their use generates. The feedback loop is bidirectional, accelerating, and in aggregate, potentially narrowing: if millions of creators coevolve with a small number of AI platforms, the diversity of creative approaches may contract even as the volume of creative output expands. This is the network-science formulation of the concern that Byung-Chul Han’s smoothness critique expresses phenomenologically.

The fitness model’s most hopeful implication for the cycle is its prediction of rapid turnover: nodes with genuinely high fitness—judgment, taste, the capacity to identify problems worth solving and communicate a vision that attracts others—can overcome incumbency and trigger Bose-Einstein condensation events in which they capture a macroscopic share of the network’s connections. This is Google overcoming AltaVista. It is the basis for the cycle’s hope that the AI transition will not merely reinforce existing power structures but create genuine opportunities for builders whose fitness is high and whose network position is low. The caveat is that fitness cannot be outsourced—it is the irreducible human element that determines whether capability equalization translates into opportunity.

Origin

Born in 1967 in Csíkszereda (then Romania, now Miercurea Ciuc in the Székely Land of Transylvania), Barabási studied physics in Bucharest and Budapest before completing his doctorate at Boston University. He joined the faculty at Notre Dame in 1995 and assembled the research group that would produce, in 1999, the paper in Science that founded the modern study of scale-free networks. The 1999 paper with Réka Albert identified preferential attachment as the mechanism underlying the power-law degree distributions observed across the natural and social world. Its publication transformed network science from a mathematical subspecialty into an interdisciplinary field with applications in biology, economics, sociology, and now AI.

The fitness model, developed with Ginestra Bianconi in 2001, introduced the critical modification: nodes have intrinsic fitness that modifies their attractiveness independently of their current connectivity, allowing high-fitness newcomers to rise against established hubs. The model predicted the Bose-Einstein condensation phenomenon—winner-take-all dominance by the highest-fitness node—which has since been observed in market dynamics from internet platforms to streaming services. His 2018 book The Formula elaborated the science of success, demonstrating that success is not a function of individual performance alone but of the interaction between performance and the network’s capacity to recognize and reward it. In 2025, Barabási and colleagues introduced the human-AI coevolution framework, defining a new field at the intersection of network science and AI research.

Key Ideas

Scale-free networks and preferential attachment. In any growing network where new connections form preferentially to already well-connected nodes, the degree distribution follows a power law: a few hubs with enormous connectivity, many nodes with almost none. This is not a market failure or a historical accident. It is a mathematical inevitability—the same mechanism that produced the web’s topology produces the structure of the creative economy. AI equalizes the production side of this network. It does not equalize the distribution side—visibility, reputation, network effects, and institutional advantage—which remains governed by preferential attachment.

Fitness and the path from periphery to hub. The fitness model modifies preferential attachment by introducing intrinsic node quality: a node’s probability of attracting new connections is proportional to the product of its connectivity and its fitness. High-fitness nodes can overcome incumbency—Google’s algorithm was fitter than AltaVista’s, and the fitness differential triggered condensation. In the AI economy, fitness is judgment, taste, and the vision that directs capability toward problems worth solving. When AI equalizes technical execution, fitness becomes the only differentiator—and fitness, unlike capability, cannot be outsourced.

Where innovation lives. Innovation does not occur uniformly; it clusters at hubs because hubs occupy positions of dense information flow and cross-domain exposure. AI tools transform every user into a potential bridge between previously disconnected knowledge clusters—functioning as structural hole closers of unprecedented reach. But bridging information does not automatically translate into bridging influence: the user who can access cross-domain information through AI still must traverse the old topology—publication, distribution, institutional endorsement—to contribute innovation that the network can receive.

Human-AI coevolution. Humans and AI algorithms continuously reshape each other through feedback loops: users’ choices train models; models shape users’ preferences; reshaped preferences generate new data; new data refines the models. This coevolution is faster than any previous tool-cognition coevolution and more personalized. Its cumulative effect on human cognitive habits, creative diversity, and the capacity for independent deep thinking is the most consequential empirical question of the AI age—and it is being answered, in real time, across millions of knowledge workers simultaneously, without controls.

Phase transitions and the Death Cross. Network phase transitions occur when a control parameter crosses a critical threshold and the system reorganizes qualitatively. The software industry’s Death Cross was such a transition: the cost of code production dropping below the threshold that sustained the old topology, forcing rapid reorganization around ecosystem rather than code as the structural principle. The transition is not gradual; it is sudden, as phase transitions always are, and the financial markets registered it as such.