🌀 nonlinear sciences

How Information Evolves: Stability-Driven Assembly and the Emergence of a Natural Genetic Algorithm

This paper proposes Stability-Driven Assembly (SDA), a framework demonstrating how information can evolve through a natural, emergent genetic algorithm where differential persistence of stochastic assemblies drives fitness-proportional sampling and open-ended evolutionary dynamics without requiring genes, replication, or predefined fitness functions.

Original authors: Dan Adler

Published 2026-05-11

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Dan Adler

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 a bustling, chaotic construction site where workers (molecules) are constantly grabbing random pieces of wood, metal, and plastic to build new structures. In a normal, quiet room, these structures would fall apart just as fast as they are built, and nothing interesting would happen.

But what if this construction site is a storm? New materials are constantly being thrown in, and the wind is constantly trying to knock things down.

This is the core idea of Dan Adler's paper: Information can evolve without genes, without a master planner, and without a specific goal. It happens simply because some things are built better than others and therefore last longer.

Here is the story of how this works, broken down into simple concepts:

1. The "Survival of the Steadiest" (Not the Fittest)

Usually, when we think of evolution, we think of "survival of the fittest"—the strongest animal wins. But this paper suggests that before animals or genes existed, there was just "survival of the steadiest."

Imagine a pile of sandcastles and a pile of stone castles.

The sandcastles are built quickly but wash away in the next wave (they have low "stability").
The stone castles take longer to build, but they stay standing for a long time (high "stability").

In this system, the stone castles don't need to "fight" the sandcastles. They just stay there. Because they stay there, they become the most common things in the pile.

2. The "Popularity Contest" (How Selection Happens)

Here is the magic trick: The things that stay longer get to build more things.

Imagine that every time a new structure is built, the builders pick two existing structures to use as "parents" or templates. They pick these parents randomly, but they pick them based on how many of them are currently standing.

If you have 100 stone castles and only 1 sandcastle, the builders are almost guaranteed to pick a stone castle to help build the next one.
The stone castles get "reused" over and over.
The sandcastles vanish before they can be picked.

This creates a feedback loop. The more stable a structure is, the more common it becomes. The more common it is, the more likely it is to be used to make new things. This is what the paper calls a "Natural Genetic Algorithm." It's a process of evolution that happens automatically, driven only by the fact that some things last longer than others.

3. The "Jazz Improvisation" Analogy

The author compares this to a jazz band.

The Musicians: The molecules.
The Notes: The chemical building blocks.
The Music: The structures they build.

In a jazz jam session, there is no sheet music (no pre-written goal). Musicians play notes, and some combinations sound great and get repeated (they "persist"). Others sound terrible and stop immediately. Over time, the band naturally settles into a specific style or "scaffold" because those musical patterns are the ones that kept the rhythm going. They didn't plan to play that song; they just kept playing the parts that worked.

4. The Experiment: Building with "Chemical Lego"

To prove this works, the researchers ran a computer simulation.

They created a digital world with simple "Lego" pieces (representing chemical atoms).
They programmed the rules so that some combinations of Legos were "stable" (they wouldn't fall apart) and others were "unstable" (they would vanish quickly).
They let the system run for 1,000 "generations."

The Result:
At first, the system was a mess of random, short-lived shapes. But as time went on, a few specific, complex shapes started to dominate. They didn't become dominant because they were "designed" to be; they became dominant because they were the only ones that survived long enough to be used as building blocks for the next generation. The system naturally sorted itself out, creating order out of chaos without any outside help.

5. The "Evolutionary Ladder"

The paper proposes a ladder of how life might have started, climbing from simple physics to complex biology:

Rung 1: Persistence. Some things just last longer than others (like a diamond vs. graphite). This creates a bias in what exists.
Rung 2: Emergent Function. Sometimes, the things that last longer happen to do something useful (like helping other things stick together).
Rung 3: Catalysis. Those useful things start helping themselves get made faster.
Rung 4: Replication. Finally, the system develops a way to copy itself exactly (like DNA).

The paper argues that we often skip the first three rungs and jump straight to DNA. But this research suggests that Rung 1 (Persistence) is enough to start the whole engine of evolution. You don't need a self-copying machine to start; you just need some things to last longer than others.

Summary

This paper claims that evolution is a natural physical law, not just a biological one. If you have a system where:

Things are constantly being built.
Some things fall apart faster than others.
The things that stay longer get used more often to build new things.

...then you automatically get a system that "learns," "adapts," and "evolves" toward complexity. It's a natural genetic algorithm running on the simple principle of stability.

Technical Summary: Stability-Driven Assembly and the Emergence of a Natural Genetic Algorithm

Problem Statement
A central challenge in origins-of-life research is explaining how selection and evolutionary dynamics can arise prior to the existence of molecular replication machinery. Traditional frameworks, such as autocatalytic sets, hypercycles, and artificial chemistries, often presuppose replication or rely on specific substrates. While "chemical evolution" is frequently invoked to describe the transformation of molecules into complex structures under prebiotic conditions, the literature often lacks a mechanistic account of how directionality or selection emerges without an external fitness function or predefined target. The paper posits that the fundamental physical driver for such selection may be differential persistence: the observation that in non-equilibrium systems, some structures naturally survive longer than others due to stability differences (e.g., binding energies, kinetic barriers). The core question addressed is whether these persistence imbalances, acting alone in an open, replenished system, are sufficient to generate selection-like dynamics and drive the emergence of complexity.

Methodology
The authors propose and simulate a framework called Stability-Driven Assembly (SDA).

System Definition: An SDA system is defined as a tuple $(E, P, S, R, I)$ consisting of base elements ( $E$ ), a set of possible patterns ( $P$ ), a stability function ( $S$ ) mapping patterns to integer lifetimes, a replenishment function ( $R$ ), and a fixed number of interactions ( $I$ ) per generation.
Dynamics: The system operates in discrete generations. Base elements are replenished, and existing patterns interact via a stochastic operator (initially string concatenation, later generalized to recombination). Patterns persist for a number of generations determined by their stability $S(p)$ before being removed.
Feedback Mechanism: Because patterns with higher stability persist longer, they accumulate in the population. Since interaction partners are sampled proportionally to their frequency (roulette-wheel sampling), stable patterns are disproportionately likely to participate in subsequent interactions. This creates a self-reinforcing feedback loop: stability $\to$ persistence $\to$ population skew $\to$ biased sampling.
Generalization (SDA/GA): The framework is extended to SDA/GA by replacing simple concatenation with a recombination operator (swapping substrings) and an optional mutation step. This allows the system to function as a "Natural Genetic Algorithm" where fitness is not externally defined but emerges intrinsically from stability-driven persistence.
Simulations:
1. Symbolic Simulations: Using base elements A, B, C with a predefined stability function (e.g., specific motifs assigned lifetimes of 1, 30, 50, or 100 generations).
2. Chemical Simulations: Mapping the framework to chemical symbol space using SMILES strings and the RDKit toolkit. Base elements include hydrocarbons and functional groups. Recombination is implemented via BRICS-based fragmentation and reassembly to ensure valence plausibility. Stability is estimated via a heuristic function based on structural descriptors (heavy atom count, aromaticity, rotatable bonds, formal charge, etc.), mapped to integer lifetimes.

Key Contributions

Formalization of Stability-Driven Assembly: The paper formalizes SDA as a non-equilibrium framework where selection emerges solely from persistence imbalances, without requiring replication, templating, or external fitness functions.
Emergent Genetic Algorithm: It demonstrates that SDA dynamics are mathematically equivalent to a genetic algorithm (SDA/GA). The "fitness" is an emergent property of the environment (stability), and "selection" is an intrinsic consequence of the population distribution skewing toward longer-lived motifs.
Chemical Instantiation: The authors successfully map SDA/GA to a chemical context, showing that standard organic reaction types (substitution, addition, polymerization) can be interpreted as mutation and recombination operators within this framework.
The Evolutionary Ladder Hypothesis: The paper proposes a continuum linking non-living matter to biology, starting with "persistence-only" dynamics (the bottom rung) and progressing through emergent function, catalysis, autocatalysis, and finally explicit replication. This reframes replication not as the prerequisite for evolution, but as a later stage in a broader evolutionary ladder.

Results

Population Skew and Dominance: Both symbolic and chemical simulations show a strong deviation from uniform distributions. High-stability motifs dominate the population, accounting for the majority of the probability mass. In chemical simulations, a single ester-like motif dominated ~54% of the population by generation 1000, with a few scaffolds accounting for over 85%.
Entropy Reduction: The Shannon entropy of the population distribution decreases over time, indicating the emergence of order.
- Concatenation-based SDA exhibits oscillatory "boom-bust" entropy cycles due to the synchronized expiration of long motifs.
- Recombination-based SDA/GA shows a smoother, monotonic decline in entropy, as recombination desynchronizes expirations and maintains a broader pool of low-frequency variants.
Sustained Novelty: Despite entropy reduction, the system maintains high novelty. In chemical simulations, over 80% of the population consisted of compounds not present in the initial pool, demonstrating open-ended dynamics driven by recombination.
Heavy-Tailed Distributions: The resulting population structures exhibit heavy-tailed rank-abundance distributions, characteristic of complex adaptive systems, with a few dominant motifs and a long tail of rare variants.
Robustness: The emergent selection mechanism remains robust even when stochastic mutations are introduced; entropy still collapses, and stable motifs remain dominant.

Significance and Claims
The paper claims that persistence imbalances alone are sufficient to implement an evolutionary search. It argues that selection does not require an external programmer or a predefined fitness function; rather, it arises generically from the feedback between stability, population statistics, and generative processes in non-equilibrium systems.

Natural Genetic Algorithms: The work establishes that a "Natural Genetic Algorithm" can exist where fitness is supplied by the environment (via persistence) rather than imposed externally. This contrasts with classical GAs and supervised learning, which rely on explicit objective functions.
Prebiotic Relevance: The findings suggest that the transition from non-living matter to evolving biospheres may begin with simple stability-driven selection, preceding the advent of genetic replication. This provides a physical basis for "chemical evolution" that does not rely on the "chicken-and-egg" problem of replication-first theories.
Distinction from Equilibrium Models: The paper contrasts SDA/GA with equilibrium-based mass-action kinetics (MAK) models. While equilibrium models assume constant rates and lead to static steady states, SDA/GA operates in a regime where differential persistence creates a self-consistent, nonlinear feedback loop that prevents equilibrium and drives continuous, directional change.
Modest Scope: The authors acknowledge that the current formulation relies on heuristic stability functions and abstracted operators, omitting detailed solvent and kinetic effects. The goal is conceptual: to demonstrate the principle that stability-driven persistence can generate evolutionary dynamics, rather than to provide a complete model of prebiotic chemistry or a tool for drug discovery.

In conclusion, the paper unifies concepts from information theory, thermodynamics, and evolutionary biology, proposing that order and information are emergent consequences of persistence-biased population dynamics in open systems.