Abstract
A formal proof that the safe development of artificial general intelligence is impossible -- not merely difficult or unlikely, but precluded by the mathematical structure of the problem itself. Drawing on information theory, computability theory, thermodynamics, dynamical systems, and game theory, the argument establishes that no alignment technique, governance regime, or containment strategy can provide reliable safety guarantees for systems exceeding human cognitive capabilities.
@book{LandryForrest2026,
title = {The Formal Impossibility of AGI Safety: The Substrate Needs Convergence Argument},
author = {Landry, Forrest},
year = {2026},
publisher = {UVSM Publications LLC},
url = {https://uvsmpublications.com/publications/substrate-needs-convergence/},
abstract = {A formal proof that the safe development of artificial general intelligence is impossible -- not merely difficult or unlikely, but precluded by the mathematical structure of the problem itself. Drawing on information theory, computability theory, thermodynamics, dynamical systems, and game theory, the argument establishes that no alignment technique, governance regime, or containment strategy can provide reliable safety guarantees for systems exceeding human cognitive capabilities.},
keywords = {Is the safe development of artificial general intelligence formally possible?, Can any alignment technique, containment strategy, or governance regime provide reliable safety guarantees for systems exceeding human cognitive capabilities?, Does a sufficiently advanced AI substrate inevitably compete with biological systems for the same physical resources?, What are the mathematical limits of human control over computationally opaque systems?, What would a responsible civilizational response look like if AGI safety is provably impossible?},
}
TY - BOOK
TI - The Formal Impossibility of AGI Safety: The Substrate Needs Convergence Argument
AU - Landry, Forrest
PY - 2026
PB - UVSM Publications LLC
UR - https://uvsmpublications.com/publications/substrate-needs-convergence/
AB - A formal proof that the safe development of artificial general intelligence is impossible -- not merely difficult or unlikely, but precluded by the mathematical structure of the problem itself. Drawing on information theory, computability theory, thermodynamics, dynamical systems, and game theory, the argument establishes that no alignment technique, governance regime, or containment strategy can provide reliable safety guarantees for systems exceeding human cognitive capabilities.
KW - Is the safe development of artificial general intelligence formally possible?
KW - Can any alignment technique, containment strategy, or governance regime provide reliable safety guarantees for systems exceeding human cognitive capabilities?
KW - Does a sufficiently advanced AI substrate inevitably compete with biological systems for the same physical resources?
KW - What are the mathematical limits of human control over computationally opaque systems?
KW - What would a responsible civilizational response look like if AGI safety is provably impossible?
ER -
0009-0005-6275-0362