The World of SPLectrum
Home > Positioning > Subjects > Assembly Theory > The Algorithmic Information Theory Debate
The Algorithmic Information Theory Debate
The most technically pointed challenge to assembly theory is the claim that it is mathematically subsumed by algorithmic information theory (AIT) — that the assembly index is not a new measure but a restricted version of an existing one, and that the theory’s empirical findings are reproducible with established tools.
The debate is active across multiple peer-reviewed venues and is not resolved.
The critique
Hector Zenil and collaborators — Felipe Abrahão, Santiago Hernández-Orozco, Narsis Kiani, Jesper Tegnér, Allen Uthamacumaran — have pursued the subsumption claim through multiple papers since 2024.
The central paper: Abrahão et al. (2024), PLOS Complex Systems: “Assembly Theory is an approximation to algorithmic complexity based on LZ compression that does not explain selection or evolution.” Further papers: Uthamacumaran, Abrahão, Kiani, Zenil (2024), npj Systems Biology and Applications: “On the salient limitations of the methods of assembly theory and their classification of molecular biosignatures.” Earlier work appeared in Parallel Processing Letters.
The core technical claim: the assembly index is mathematically equivalent to a restricted version of LZ compression, which is itself an approximation to Kolmogorov-Solomonoff-Chaitin algorithmic complexity. The assembly index calculation is equivalent to the size of a compressing context-free grammar. Assembly theory “constitute[s] a weak version of algorithmic complexity” (Abrahão et al.).
The empirical claim: the same molecular discrimination achieved by assembly index — separating biological from abiotic samples — can be reproduced using Shannon entropy and standard compression algorithms. The critics published this finding using a substantially larger chemical dataset than the assembly theory authors’ original sample.
The framing claim: assembly theory makes broad unification claims that are unsupported, and relies on definitions from complexity theory used without attribution.
The defence
Cronin, Walker, and the Glasgow / ASU teams responded formally in “Assembly theory and its relationship with computational complexity” (npj Complexity, 2025). Three main lines:
LZ compression fails at the interesting cases. The equivalence claim is technically correct for the restricted regime but does not hold where Kolmogorov complexity is most informative. LZ compression is computable and bounded; Kolmogorov complexity is not. Claiming that assembly theory is “just” LZ compression misses what the theory is doing with the measure.
Mechanism, not measure. Assembly theory is mechanistically grounded in physical construction rather than algorithmic compression. The point of building measures in physics is to capture the mechanisms that produce objects. The assembly index is defined through physical joining operations, not through abstract symbol manipulation. The distinction between a physically grounded measure and a purely formal one is substantive, not cosmetic.
Measurability. The assembly index is experimentally measurable via mass spectrometry. Algorithmic complexity is not directly measurable for physical objects. The empirical apparatus — mass spec, infrared, NMR — gives assembly theory something that abstract complexity measures do not provide: a bridge to the laboratory.
The state of the debate
The exchange is ongoing through 2025–2026. Zenil’s group continues to publish and comment; the defenders prepare further responses. The debate has not converged.
Whether the assembly index is mathematically subsumed by Kolmogorov complexity, or whether assembly theory is a mechanistically distinct framework that happens to share formal structure with compression-based measures, remains an open question in the peer-reviewed literature. The technical overlap is real; the question is whether the physical grounding makes a difference.