🔬 physics

Autocatalytic knowledge dynamics in the AI era

This study proposes a macrodynamic model of autocatalytic knowledge production that incorporates AI-driven acceleration alongside counteracting forces like resource scarcity and information noise, demonstrating how their balance determines whether the system grows, degrades, or collapses, with empirical validation using WIPO patent data.

Original authors: Boris M. Dolgonosov

Published 2026-08-03

📖 6 min read🧠 Deep dive

CC BY 4.0

Original authors: Boris M. Dolgonosov

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

The Great Knowledge Engine: Why We Can't Just Keep Speeding Up Forever

Imagine the history of human discovery not as a straight line, but as a snowball rolling down a hill. At first, it's tiny and barely moves. But as it rolls, it picks up more snow, gets bigger, and rolls faster, gathering even more snow. This is the idea of autocatalysis: the past helps create the future, and the more you have, the easier it is to get more. In the world of science and technology, this means that every new discovery gives us the tools to make the next one.

However, snowballs don't roll forever without hitting a wall. They eventually run out of snow, or they get so big they start to wobble and break apart. In the real world, our "snowball" of knowledge faces similar limits. We need energy to think and build, and our planet has a limit on how much heat it can handle before things get too hot to function. Plus, there's the problem of "noise"—so much fake or useless information that it becomes hard to find the good stuff.

This brings us to a fascinating question: Can our civilization keep accelerating its learning forever, or are we heading for a crash? A researcher named Boris M. Dolgonosov has built a mathematical model to answer this. He treats human knowledge like a living system that grows, slows down, or sometimes even collapses, depending on the balance between our ability to learn and the obstacles we face.

The Paper's Big Idea: The Equation of Learning

In his paper, "Autocatalytic knowledge dynamics in the AI era," Dolgonosov tries to write the "rules of the road" for how fast humanity can learn. He suggests that the speed at which we create new knowledge depends on a tug-of-war between three main forces.

1. The Gas Pedal (Autocatalysis)
Think of this as the "snowball effect." The more knowledge we have, the easier it is to make new knowledge. If you know how to build a wheel, you can build a cart; if you have a cart, you can build a train. The paper argues that this force makes our learning speed grow incredibly fast, potentially leading to a "singularity"—a point where things get so fast they seem to happen all at once.

2. The Brakes (Resource and Environmental Limits)
You can't just press the gas pedal forever. Eventually, you run out of gas (resources like money, energy, and materials) or the engine overheats (environmental limits). The paper points out that making knowledge takes energy, and using that energy creates heat. If we produce too much heat, the planet gets too hot to live on. This acts as a "resource-environmental brake" that slows us down before we can reach that impossible singularity.

3. The Static (Information Noise and Loss)
Imagine trying to listen to your favorite song while someone is shouting random words next to you. That's what happens when there is too much "noise"—fake news, outdated ideas, or useless data. The paper suggests that as we generate more information, we also generate more "trash" that we have to filter out. This wastes energy and slows us down. Also, old knowledge sometimes gets forgotten or becomes useless, which is like losing parts of your map.

What the Model Predicts: Three Possible Futures

Dolgonosov uses a mathematical equation to simulate what happens when these forces fight against each other. He finds that the future of human knowledge depends on the balance of these forces, leading to three distinct scenarios:

The Happy Path (Growth): If the "gas pedal" (autocatalysis) is strong enough to beat the brakes and the static, our knowledge keeps growing. The rate of discovery speeds up until it hits a steady, high level where we are constantly inventing new things.
The Slow Fade (Degradation): If the brakes get too strong, our speed of discovery slows down. We don't stop completely, but we stop getting faster. We keep learning, but at a much slower, manageable pace. The system survives, but it doesn't thrive like it used to.
The Crash (Collapse): If the brakes and the static are too strong, or if we start with too little knowledge to get a good run, the speed of discovery drops to zero. The system stops learning entirely, and the civilization effectively collapses because it can no longer solve new problems.

The Role of AI: The Turbocharger

The paper looks closely at Artificial Intelligence (AI) as a game-changer. AI acts like a massive turbocharger for the "gas pedal." It helps us process information faster, test more ideas, and connect dots we couldn't see before. The author suggests that AI is already pushing us into a new "era" of knowledge.

However, the paper warns that AI also tightens the brakes. Because AI needs huge amounts of electricity and creates massive amounts of heat, it might make the environmental limits hit us sooner. Furthermore, AI can generate so much "noise" (fake or unverified ideas) that we might get stuck trying to sort the good stuff from the bad.

What the Data Says (and Doesn't Say)

To test his ideas, the author looked at real-world data, specifically the number of patent applications filed with the World Intellectual Property Organization (WIPO) and scientific papers posted on the arXiv server.

The Patent Plateau: The data shows that patent filings grew for a long time but recently hit a "plateau" (a flat line) around 2020. The paper suggests this is because we hit a temporary limit in our current technology. However, it predicts that as AI becomes more integrated into engineering, we might see a new jump in growth—a "phase transition"—where the curve goes up again.
The Preprint Explosion: In the world of scientific papers, there was a huge spike in 2023–2024, driven by AI tools helping researchers write and share ideas faster. This supports the idea that AI is currently boosting the "gas pedal."

The Bottom Line

The paper concludes that we are standing at a crossroads. We have a powerful engine (AI) that can accelerate our learning, but we are also hitting a wall (energy limits and heat). The future isn't guaranteed to be a straight line up. Instead, our civilization might go through a series of "phase transitions"—sudden shifts where we either find a way to overcome the limits and keep growing, or we get stuck in a slower, degraded state.

The author suggests that the most critical challenge isn't just making more knowledge, but figuring out how to verify it. If AI can generate a thousand theories in a second, humans (or future AI systems) will need to spend a hundred years just checking which ones are true. Until we solve that "verification bottleneck," the brakes on our knowledge engine will remain tight. The paper doesn't promise a utopia or a doom; it simply offers a map of the possible routes, showing that our survival depends on keeping the balance between our drive to learn and the limits of our world.

Technical Summary: Autocatalytic Knowledge Dynamics in the AI Era

Problem Statement
The paper addresses the macrodynamics of knowledge production within intelligent systems, specifically examining the accelerating rate of knowledge growth in the AI era. The central problem is to mathematically model how knowledge accumulates, identifying the driving forces that accelerate this process and the counteracting forces that inhibit it. The study seeks to determine whether the current trajectory of knowledge production leads to sustainable development, system degradation, or collapse, and to elucidate the specific role of Artificial Intelligence (AI) in altering these dynamics. The author posits that while knowledge production is inherently autocatalytic (existing knowledge generates new knowledge), it is constrained by resource scarcity, environmental limits, knowledge obsolescence, and information noise.

Methodology
The study employs a mathematical modeling approach to construct the macrodynamics of knowledge.

Master Equation Formulation: The author defines the volume of knowledge as $q(t)$ $q (t)$ and the rate of production as $v = dq/dt$. The core of the methodology is the derivation of a "master equation" for the acceleration of knowledge production ($dv/dt$). This equation incorporates:
- An autocatalytic term ( $av^2$ ), representing the positive feedback loop where existing knowledge catalyzes new discoveries.
- A resource-environmental braking term represented by the factor $(1 - bv)$ multiplying the autocatalytic term, yielding $av^2(1 - bv)$ . This represents physical limits such as energy scarcity, infrastructure constraints, and planetary heat dissipation.
- A knowledge loss term ($-cv$), accounting for the inactivation of knowledge due to obsolescence, the filtering of information noise, and the energy dissipation required to verify unreliable data.
Dimensionless Analysis: The system is normalized using dimensionless variables to reduce the number of parameters. Key dimensionless parameters include $\mu$ (the ratio of inhibitory factors to autocatalysis intensity) and $y_0$ (the ratio of the initial production rate to a planetary threshold rate).
System Decomposition: The model is analyzed using an adiabatic approximation, separating the system into a "fast subsystem" (where coefficients are treated as constant) and a "slow subsystem" (where coefficients evolve as knowledge accumulates).
Empirical Calibration: The theoretical model is applied to historical data, specifically international patent filings from the World Intellectual Property Organization (WIPO) and scientific preprints from arXiv, to estimate model parameters and validate the dynamics.

Key Contributions

The Master Equation of Knowledge Dynamics: The paper establishes a differential equation that unifies autocatalysis, resource constraints, and information loss into a single framework. It formalizes the transition from hyperbolic growth (singularity) to logistic constraints.
Phase Transition Analysis: The study identifies three distinct dynamic modes based on the balance between autocatalysis and inhibition:
- Mode 1 (Strong Autocatalysis): The system can survive and develop, either reaching a high plateau or stabilizing at a lower non-zero rate depending on initial conditions.
- Mode 2 (Balanced Forces): A critical threshold where the system may survive at a constant rate or collapse if initial rates are too low.
- Mode 3 (Strong Inhibition): Inhibition dominates autocatalysis, leading inevitably to system collapse (rate drops to zero) regardless of initial conditions.
Integration of AI into Dynamics: The model explicitly treats AI not just as a tool but as a factor that modifies the coefficients of the equation. AI increases the autocatalysis coefficient ( $a$ ) but simultaneously increases inhibitory factors ( $b$ and $c$ ) due to higher resource consumption and information noise.
Demographic Correlation: The paper extends the knowledge dynamics model to population growth, demonstrating that population size is directly linked to knowledge production rates, thereby applying the same phase transition logic to demographic stability.

Results

Dynamic Scenarios: The analysis reveals that the fate of an intelligent system depends on the parameter space of $\mu$ and $y_0$ . If the initial rate of knowledge production is too low relative to inhibition, the system collapses. If inhibition is too strong relative to autocatalysis ( $\mu > 1/4$ ), the system collapses regardless of the initial rate.
Plateau and Thresholds: The model predicts a "planetary barrier" or threshold rate ( $v_b$ ) determined by resource and environmental limits. Current WIPO patent data shows a plateau around 0.3 million applications per year, interpreted as the system approaching this threshold in the absence of AI-driven acceleration.
AI Impact: The application of the model to patent and preprint data suggests that the current stagnation in patent filings represents a pre-AI plateau. The integration of AI is expected to trigger a "phase transition," shifting the system to a new state with higher productivity. However, the long-term analysis of variable coefficients suggests that if the growth of inhibitory factors (energy, heat, noise) outpaces the growth of autocatalysis (exponent $\epsilon > 0$ ), the system will eventually face a fatal decline.
Barriers: The study identifies three specific barriers to future growth: the energy barrier (Landauer's limit), the thermal barrier (planetary heat dissipation), and the data density barrier. The thermal barrier is identified as the most immediate constraint.

Significance
The paper claims to provide a mathematical framework for understanding the "intelligent capital" of civilization. Its significance lies in shifting the perspective of knowledge from a linear accumulation of facts to a dynamic system governed by a balance of forces.

Predictive Capability: The model offers a way to predict whether a civilization is on a path to sustainable development, degradation, or collapse based on the balance of autocatalytic and inhibitory forces.
AI Contextualization: It frames AI development not merely as a technological breakthrough but as a fundamental shift in the system's dynamics, capable of inducing phase transitions.
Sustainability Warning: The study highlights that without overcoming physical barriers (specifically thermal and energy limits), the acceleration of knowledge production via AI may lead to system collapse. It suggests that the ultimate bottleneck for knowledge growth may shift from human verification to the physical limits of the planet's ability to dissipate heat and process information.
Evolutionary Outlook: The paper concludes that the future of knowledge production may involve a symbiosis where AI becomes a core architectural element of a "collective individual," potentially displacing humans from the direct production circuit while remaining essential for goal formulation.