Latent visual mechanisms as computational phenotypes: Toward a precision science of vision
Abstract
Vision science has traditionally sought mechanisms shared across observers, treating individual differences as noise. Yet factor-analytic and latent-variable studies show that variation among observers is highly structured, recovering reproducible dimensions in contrast sensitivity, color, stereopsis, motion, and face recognition, among others. What are these latent dimensions? I propose they be treated as candidate computational phenotypes: a term borrowed from computational psychiatry and re-operationalized for vision as a stable, measurable pattern of visual information processing recovered from the covariance structure of individual differences. A computational phenotype is not a latent factor, mechanism, endophenotype, or biomarker; it occupies Marr's algorithmic level, between behavior and biology, and earns the name only by meeting stringent criteria of reliability, replication, convergent validity, and mechanistic relevance. I review candidate phenotypes drawn from many laboratories, recast development, aging, and disease as perturbations of phenotypic structure, and turn the strongest objection (that vision has no common factor) into support, since the relevant phenotypes are narrow and mechanism-specific. The payoff spans audiences: diagnostic profiles for clinicians, a non-invasive way to count and test visual mechanisms for basic scientists, and performance-predicting traits for human-factors work. The framework moves vision science from the average observer toward a precision science of visual computation.
Related
- Global knowledge structure and thematic evolution in inflammation-associated gastric cancer epigenetics: a multidatabase science-mapping study with nested latent Dirichlet allocation and independent NMF validation
- Macrophage reprogramming by the ultraviolet filter benzophenone-1 promotes malignant phenotypes in breast cancer cells: integrated computational and experimental evidence
- Joint–brain communication in rheumatoid arthritis: glial mechanisms, pain phenotypes, and translational evidence
- Interindividual Variability in Functional Food Clinical Trials: Statistical Challenges, Responder Phenotypes, and Implications for Precision Nutrition
- Burnout- and Depression-Like Functional Collapse States — A Within-Person Biomarker Framework for Precision Psychiatry: State-Dependent Multisystem Regulation in High Processing Cost Phenotypes (HPCPh)
BibTeX
@misc{peterzell2026latent,
title = {{Latent visual mechanisms as computational phenotypes: Toward a precision science of vision}},
author = {David Henry Peterzell},
howpublished = {PsyArXiv},
year = {2026},
month = oct,
doi = {10.31234/osf.io/xq82m_v2},
url = {https://osf.io/preprints/psyarxiv/xq82m_v2/},
note = {CC BY 4.0},
}
RIS
TY - UNPB
TI - Latent visual mechanisms as computational phenotypes: Toward a precision science of vision
AU - David Henry Peterzell
PY - 2026
DA - 2026/10/05
PB - PsyArXiv
DO - 10.31234/osf.io/xq82m_v2
UR - https://osf.io/preprints/psyarxiv/xq82m_v2/
AB - Vision science has traditionally sought mechanisms shared across observers, treating individual differences as noise. Yet factor-analytic and latent-variable studies show that variation among observers is highly structured, recovering reproducible dimensions in contrast sensitivity, color, stereopsis, motion, and face recognition, among others. What are these latent dimensions? I propose they be treated as candidate computational phenotypes: a term borrowed from computational psychiatry and re-operationalized for vision as a stable, measurable pattern of visual information processing recovered from the covariance structure of individual differences. A computational phenotype is not a latent factor, mechanism, endophenotype, or biomarker; it occupies Marr's algorithmic level, between behavior and biology, and earns the name only by meeting stringent criteria of reliability, replication, convergent validity, and mechanistic relevance. I review candidate phenotypes drawn from many laboratories, recast development, aging, and disease as perturbations of phenotypic structure, and turn the strongest objection (that vision has no common factor) into support, since the relevant phenotypes are narrow and mechanism-specific. The payoff spans audiences: diagnostic profiles for clinicians, a non-invasive way to count and test visual mechanisms for basic scientists, and performance-predicting traits for human-factors work. The framework moves vision science from the average observer toward a precision science of visual computation.
ER -