The AI Genome: Model Evolution, Lineage Tracking, and Cross-Family Comparison
Modern foundation models increasingly evolve through complex networks of fine-tuning, knowledgedistillation, model merging, and synthetic data generation, rendering informal naming conventionsinadequate for rigorous comparative analysis. This paper introduces the AI Genome, astandardized research framework and symbolic sequence taxonomy designed to represent,compare, and trace the evolutionary lineage of artificial intelligence models across open-weightsand proprietary ecosystems. The framework organizes model attributes into four functional chromosomes: Structural Genotype (CHR1), Ancestral Lineage (CHR2), Epigenetic Modifiers(CHR3), and Behavioral Phenotype (CHR4)—comprising 21 fixed positional loci encoded via aFASTA-style single-letter allele alphabet. To handle corporate opacity and undisclosed specifications, the schema incorporates explicit null and unknown locus markers alongside maskedevolutionary distance metrics. We formalize the normalized Genomic Similarity Score (SG) andintroduce pairwise Consensus Archetypes to isolate conserved traits across model generations. An empirical pilot study across six Google DeepMind models (five proprietary Gemini releases) demonstrates how symbolic genomic encoding quantifies intrafamily evolutionary drift, differentiates model tiers, and extracts an 8-locus Universal Cohort Core Archetype. Keywords: AI Genome, Foundation Models, Model Lineage, Symbolic Taxonomy, Sequence Alignment,Phylogenetic Analysis, Large Language Models
Authors
- Rashid Suleiman Galiev (ORCID: https://orcid.org/0009-0006-9333-7869)
Institutions
- Boston Children's Hospital (US)
Publication Details
- Journal
- Zenodo (CERN European Organization for Nuclear Research)
- Published
- 2026-11-16
- DOI
- https://doi.org/10.5281/zenodo.24284132
- Primary Topic
- Topic Modeling
- Type
- article
- Field-Weighted Citation Impact
- 0.00