ml-pragmatic-shortcuts-create-fundamental-fragility

IN derived (depth 4)

Created 2026-06-21T10:09:45+00:00 · Reviewed 2026-06-21T15:37:01+00:00

ML architectures succeed through pragmatic inductive biases rather than biological fidelity, and these architectures exhibit adversarial vulnerabilities absent in biological perception — suggesting that the engineering choices enabling ML progress may contribute to characteristic failure modes, though the antecedents do not establish a direct causal link between specific shortcuts and specific vulnerabilities.

Justifications

SL — Pragmatic bias selection succeeds on benchmarks but creates attack surfaces that faithful biological modeling might avoid

Antecedents (all must be IN):

  • IN inductive-bias-not-biological-fidelity-drives-ml — CNNs illustrate that effective ML architectures can succeed through well-chosen inductive biases rather than biological fidelity — their local connectivity and weight sharing capture useful structural constraints despite not faithfully replicating neuroscience. The manifold hypothesis offers one explanation for why such biases work, since if data lies along low-dimensional manifolds, architectures that exploit local structure can generalize effectively regardless of their biological motivation.
  • IN adversarial-vulnerability-limits-deployment-despite-performance — The tension between superhuman benchmark performance and fundamental adversarial vulnerability creates a deployment paradox — neural networks can exceed human accuracy on standard benchmarks while remaining susceptible to imperceptible perturbations, and no amount of scaling resolves this because it is a general property of the architecture class, not a training deficit.

Dependents

These beliefs depend on this one: