theoretical-crisis-parallels-practical-fragility

IN derived (depth 3)

Created 2026-06-21T10:16:39+00:00 · Reviewed 2026-06-21T15:37:01+00:00

ML's theoretical and practical reliability crises are parallel and reinforcing — generalization theory is in fundamental revision as double descent and benign overfitting undermine the classical framework, while deployed neural networks face two independent failure classes (adversarial vulnerability, algorithmic bias) that the revising theory cannot yet predict or prevent.

Justifications

SL — Theory revision (depth-2) and practical failure classes (depth-2) proceed in parallel — the theory being revised is precisely the one needed to address the practical failures

Antecedents (all must be IN):

  • IN generalization-theory-in-fundamental-revision — ML generalization theory is undergoing fundamental revision on two independent fronts — the No Free Lunch theorem established that no universal best model exists (killing the quest for a single optimal algorithm), while double descent and benign overfitting overturned the classical U-shaped bias-variance curve (killing the traditional model selection heuristic) — leaving the field without a reliable theoretical guide to practice.
  • IN neural-networks-face-two-independent-failure-classes — Neural networks face two distinct failure classes that standard accuracy benchmarks may not capture — adversarial vulnerability (a general property spanning supervised learning, reinforcement learning, and single-pixel attacks) and systemic bias (structural discrimination from training data documented across decades from medical admissions to criminal justice) — suggesting that improving performance on i.i.d. test sets alone is insufficient to address either.

Dependents

These beliefs depend on this one: