Salvato in:
Dettagli Bibliografici
Autori principali: Wang, Charles L., Dorchen, Keir, Jin, Peter
Natura: Preprint
Pubblicazione: 2025
Soggetti:
Accesso online:https://arxiv.org/abs/2510.04399
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866918318427865088
author Wang, Charles L.
Dorchen, Keir
Jin, Peter
author_facet Wang, Charles L.
Dorchen, Keir
Jin, Peter
contents As systems trend toward superintelligence, a natural modeling premise is that agents can self-improve along every facet of their own design. We formalize this with a five-axis decomposition and a decision layer, separating incentives from learning behavior and analyzing axes in isolation. Our central result identifies and introduces a sharp utility-learning tension, the structural conflict in self-modifying systems whereby utility-driven changes that improve immediate or expected performance can also erode the statistical preconditions for reliable learning and generalization. Our findings show that distribution-free guarantees are preserved iff the policy-reachable model family is uniformly capacity-bounded; when capacity can grow without limit, utility-rational self-changes can render learnable tasks unlearnable. Under standard assumptions common in practice, these axes reduce to the same capacity criterion, yielding a single boundary for safe self-modification.
format Preprint
id arxiv_https___arxiv_org_abs_2510_04399
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle On The Statistical Limits of Self-Improving Agents
Wang, Charles L.
Dorchen, Keir
Jin, Peter
Artificial Intelligence
Machine Learning
As systems trend toward superintelligence, a natural modeling premise is that agents can self-improve along every facet of their own design. We formalize this with a five-axis decomposition and a decision layer, separating incentives from learning behavior and analyzing axes in isolation. Our central result identifies and introduces a sharp utility-learning tension, the structural conflict in self-modifying systems whereby utility-driven changes that improve immediate or expected performance can also erode the statistical preconditions for reliable learning and generalization. Our findings show that distribution-free guarantees are preserved iff the policy-reachable model family is uniformly capacity-bounded; when capacity can grow without limit, utility-rational self-changes can render learnable tasks unlearnable. Under standard assumptions common in practice, these axes reduce to the same capacity criterion, yielding a single boundary for safe self-modification.
title On The Statistical Limits of Self-Improving Agents
topic Artificial Intelligence
Machine Learning
url https://arxiv.org/abs/2510.04399