Training Data and the Maladaptive Mind
A Computational Framework for Developmental Psychology
Abstract
Childhood trauma reframed as maladaptive learned patterns from suboptimal training environments, with four developmental 'training data problems': direct negative experiences, noisy signals, absent positive experiences (linked to alexithymia), and limited exposure. Dissociation is modelled as meta-learned protective suppression, grounding the PTSD/CPTSD distinction. Toy PyTorch experiments illustrate the mechanisms: gradient magnitudes scale linearly with penalty weight (ratio 1,247 ± 93 at λ=1000), and limited caregiver diversity impairs generalization (p = 0.005, Bonferroni-corrected).
Suggested citation
Murad Farzulla (2026). Training Data and the Maladaptive Mind. Dissensus Working Paper DAI-2514. DOI: 10.21203/rs.3.rs-8634152/v1
Methodology
PyTorch simulation
Gradient analysis
Catastrophic forgetting
Topics
Philosophy
AI Safety
Psychology