born-again-trees-prove-interpretability-extractable-but-not-scalable

IN derived (depth 2)

Created 2026-06-21T14:03:57+00:00 · Reviewed 2026-06-21T15:37:01+00:00

Born-again decision trees demonstrate that interpretability can be extracted from black-box ensembles — yet this extraction path leads back to the interpretable model families whose inverse correlation with capability is already established, proving that interpretability recovery is possible in principle but constrained to the same capability ceiling that makes interpretable models insufficient.

Justifications

SL — Interpretability can be distilled out but only into the model families already known to trade capability for interpretability — the extraction doesn't escape the tradeoff.

Antecedents (all must be IN):

IN born-again-tree-recovers-interpretability — Model compression via 'born-again' decision trees can transform a random forest into a single minimal decision tree that faithfully reproduces the ensemble's decision function, recovering interpretability.
IN interpretability-inversely-correlated-with-capability — There is a tension between interpretability and model complexity in ML: easily interpretable model families (decision trees, linear models, rule-based models, attention-based models) tend to be simpler, while neural networks that achieve strong performance are 'black box' models requiring separate XAI research to explain. Even within a single family, scaling from a single decision tree to a random forest ensemble trades interpretability for accuracy.

Dependents

These beliefs depend on this one:

IN interpretability-extraction-confirms-crisis-closure — The demonstration that interpretability can be extracted from black-box models (born-again trees) but only at the cost of returning to capability-limited model families, combined with the permanent accountability vacuum, confirms that the interpretability-capability tradeoff is not a temporary engineering limitation but a structural feature of the crisis — every known path to accountability leads back through the same capability ceiling, establishing that the tradeoff is a closed loop rather than an open frontier.