🔬 physics

Toward a new AI winter? How diffusion of technological innovation on networks leads to chaotic boom-bust cycles

This paper proposes a unified mathematical model combining innovation diffusion on networks with supply-demand-investment dynamics to demonstrate how excessive investment or diffusion can trigger chaotic boom-bust cycles, offering a quantitative framework to explain historical patterns like NFT volatility and predict the potential emergence of a new AI winter.

Original authors: Sabin Roman, Francesco Bertolotti

Published 2026-02-04

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: Sabin Roman, Francesco Bertolotti

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine the world of technology not as a straight line going up, but as a giant, interconnected web of friends at a party. This paper builds a mathematical model to understand how these "friends" (different technologies) influence each other, how money flows in and out, and why the party sometimes turns into a wild, chaotic dance floor that eventually crashes.

Here is the breakdown of the paper's ideas using simple analogies:

1. The "Friend Group" Effect (Innovation Diffusion)

Think of technology like a group of friends. If your best friend learns a new skill, you are likely to learn it too.

The Idea: Technologies don't grow in isolation. A computer chip gets better because the software gets better, which gets better because the data tools get better. They feed off each other.
The Surprising Twist: The authors found that if one technology tries to grow too fast on its own (high growth rate) without sharing knowledge with its neighbors (low diffusion), it actually hits a wall and stops growing sooner.
The Analogy: Imagine a runner sprinting alone. They burn out quickly. But if that runner runs with a team, sharing tips and pacing each other, the whole group can keep going much longer and reach higher heights. The paper suggests that in the current AI market, companies are sprinting alone, which might lead to an early burnout.

2. The "Feast and Famine" Cycle (Business Cycles)

The paper also looks at the money side: Supply (making the product), Demand (people wanting it), and Investment (money poured in).

The Mechanism:
- When people want a product, companies make more.
- When companies make more, investors get excited and pour in more money.
- But if investors pour in too much money too quickly, the system gets unstable.
The Result: Instead of a smooth, steady growth, the market starts to oscillate like a pendulum. It swings wildly between a "Boom" (everyone is buying, prices are high) and a "Bust" (everyone stops buying, prices crash).
The Analogy: Think of a restaurant. If the owner sees a line out the door, they hire 50 new chefs and buy 50 new stoves. Suddenly, they have too much food and no customers left to eat it. The restaurant goes bankrupt. Then, when the market is empty, no one invests, and the cycle waits to start again.

3. The "Chaotic Dance" (The Combined Model)

When the authors combined the "Friend Group" idea with the "Feast and Famine" idea, they found something scary: Chaos.

What Happens: If the network of technologies is very connected and investors are pouring in too much money, the system doesn't just swing back and forth; it goes haywire. It becomes unpredictable.
The NFT Example: The authors tested this against data from Non-Fungible Tokens (NFTs). NFTs had a massive, wild boom and then a massive crash. The model showed that this wasn't just bad luck; it was a mathematical certainty when investment and diffusion get too high. The market becomes so sensitive that a tiny change can cause a massive crash.
The "AI Winter" Warning: The paper suggests that Artificial Intelligence (AI) is currently in a similar spot.
- There is massive investment (hype).
- There is rapid growth, but perhaps not enough "sharing" or open diffusion between different parts of the AI ecosystem.
- The Risk: If the investment gets too high and the growth hits a wall (like a computer chip hitting a physical limit), the whole system could snap. This "snap" is what the authors call a potential "AI Winter." This isn't just a slow-down; it's a sudden, chaotic collapse where interest and funding dry up, similar to what happened to AI in the 1980s.

The Takeaway

The paper argues that more money and faster growth aren't always better.

If you push a system too hard without letting ideas spread and share (diffusion), you risk a chaotic crash.
To avoid a new "AI Winter," the authors suggest we need to slow down the investment race, encourage open sharing of ideas (like open-source code), and support research even when the hype dies down, rather than only funding it when everyone is excited.

In short: The paper uses math to show that if we keep pouring gas on the AI fire without building a better structure for how those technologies talk to each other, we might end up with a fire that burns out and leaves us in the cold.

Technical Summary: Toward a new AI winter? How diffusion of technological innovation on networks leads to chaotic boom-bust cycles

Problem Statement
The paper addresses the challenge of modeling the complex techno-social landscape of artificial intelligence (AI) and technological innovation. While foundational theories like Rogers' diffusion curve exist, they often assume independent adoption trajectories or linear market dynamics, failing to capture AI's general-purpose nature, rapid recursive advancement, and deep interdependencies between data, compute, and protocols. The authors posit that current models cannot adequately explain the uneven diffusion of AI or the conditions under which the market might collapse into a new "AI winter." Specifically, there is a lack of quantitative models that unify innovation diffusion, business cycle dynamics (supply, demand, investment), and network effects to explain the emergence of chaotic boom-bust cycles.

Methodology
The authors propose a theoretical framework consisting of three progressively complex dynamical systems, modeled as ordinary differential equations (ODEs) and solved numerically using Julia (DifferentialEquations library). The study utilizes random regular graphs (and tests robustness on scale-free and star networks) to represent the network of interdependent technologies.

Pure Diffusion Model: This model captures the growth of a technology $y_i$ and its market capacity $u_i$ . Growth is logistic, limited by capacity. Crucially, the capacity $u_i$ is not static; it increases via a diffusion mechanism driven by the development of neighboring technologies in the network. The model uses a ReLU function to ensure capacity only grows (non-negative diffusion), reflecting that related technologies reinforce one another.
Business Cycle Model: This is a three-variable system representing Supply ( $x$ $x$ ), Demand ( $y$ $y$ ), and Investment ( $z$ $z$ ).
- Supply grows with investment but decays over time.
- Demand recovers logistically but is reduced when met by supply (modeled via a Lotka-Volterra style interaction term $-\alpha xy$ ).
- Investment increases based on the product of supply and demand (market success) but decays over time.
Techno-landscape Market Model: This integrates the previous two models. It applies the business cycle dynamics to each node in the technology network, while the market capacity ( $u_i$ ) for each node evolves according to the diffusion model. This creates a coupled system where economic feedback loops interact with technological interdependencies.

Key Contributions

Unified Framework: The paper introduces a mathematical framework that simultaneously models technological diffusion, market capacity expansion via network effects, and economic business cycles.
Identification of Regimes: The study identifies distinct dynamical regimes based on critical parameters: the investment sensitivity ( $\alpha$ ) and the diffusion rate ( $\sigma$ ).
Chaos and Instability: The authors demonstrate that when investment or diffusion rates exceed specific thresholds, the system transitions from stable equilibria to limit cycles and, in the networked context, to transient chaotic boom-bust cycles.
Network Topology Effects: The research shows that network heterogeneity (e.g., hub-dominated scale-free networks) lowers the threshold for chaos, making systems with highly connected hubs more sensitive to instability.

Results

Diffusion Dynamics: In the pure diffusion model, high intrinsic growth rates combined with low diffusion rates lead to earlier stagnation at lower development levels. Conversely, lower growth rates with high diffusion rates sustain growth longer and achieve higher overall development, suggesting that knowledge sharing is more critical than raw growth speed for long-term stability.
Business Cycles: The business cycle model exhibits a supercritical Hopf bifurcation. Below a critical investment threshold ( $\alpha_c$ ), the system stabilizes at a fixed point. Above this threshold, it enters a stable limit cycle (oscillations).
Chaotic Boom-Bust Cycles: In the combined network model, high values of both investment ( $\alpha$ ) and diffusion ( $\sigma$ ) lead to chaotic dynamics. The system exhibits transient chaos characterized by erratic boom-bust cycles where demand and supply oscillate with increasing amplitude before eventually collapsing.
Empirical Correlation: The model's chaotic regime qualitatively matches the transaction data of Non-Fungible Tokens (NFTs), which display sharp, asymmetric boom-bust patterns. The authors also note that the model reproduces the stagnation and renewal patterns seen in computing hardware (e.g., the shift from single-thread performance plateaus to multi-core growth) and Large Language Models (LLMs).
Lyapunov Exponents: The chaotic regime is confirmed by positive finite-time Lyapunov exponents, indicating sensitivity to initial conditions and deterministic chaos.

Significance and Claims
The paper claims that these quantitative considerations provide a structural explanation for potential market collapses in AI. The authors argue that:

Uninterrupted expansion can be fragile: A persistently growing market capacity can create an illusion of boundless potential, leading to overinvestment and speculation.
Conditions for an AI Winter: The model suggests that a new "AI winter" is structurally plausible if the current regime of high investment and high diffusion (driven by hype and speculative expectations) continues without moderation. This could trigger a collapse similar to the 1980s AI winter, driven by the same structural mechanisms: excessive investment, limited diffusion (knowledge silos), and systemic fragility.
Policy Implications: The authors suggest that policy mechanisms should aim to moderate investment rates and enhance diffusion (e.g., through open standards, interoperability, and knowledge sharing). Counter-cyclical public funding could help stabilize long-term trajectories.
Modest Predictions: The authors explicitly state they are not predicting the timing or severity of a future AI winter. Instead, they delineate the conditions under which the likelihood of such a downturn increases, highlighting that the current trajectory is not inevitable but structurally risky.

The study concludes that while the models are theoretical, they capture qualitative features of real-world data and offer a conceptual foundation for understanding how the interaction between technological diffusion and investment can lead to instability, urging a shift toward more resilient technological ecosystems.

1. The "Friend Group" Effect (Innovation Diffusion)

2. The "Feast and Famine" Cycle (Business Cycles)

3. The "Chaotic Dance" (The Combined Model)

The Takeaway

More like this