The Undecidability of Artificial General Intelligence (AGI) Alignment

Jose Pascual Gumbau Mezquita · 2026

Read The Undecidability of Artificial General Intelligence (AGI) Alignment by Jose Pascual Gumbau Mezquita (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2606-28639.txt.

Description

Jose Pascual Gumbau Mezquita argues, in a solo theoretical paper, that the deepest obstacle to aligning a future artificial general intelligence is not building an aligned system but ever verifying one. He presents two claimed impossibility results, which he calls the Unverifiability Theorem of Alignment and a theorem of finite structural unverifiability. He grounds them at what he terms Trakhtenbrot's Wall.

On his account, defenses built on finite hardware or halting designs cannot escape the logical blocks. Open domains fall to undecidability results in the line of Rice and Gödel. Universal finite checking collapses into incomputability, and bounded settings trap the supervisor in worst cases that cannot be solved in reasonable time. A claimed corollary is a trilemma among soundness, completeness, and tractability. Workable containment plans, he concludes, are forced give-ups of logical expressivity rather than stopgap patches.

Catalog record

CREATORJose Pascual Gumbau Mezquita
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE40 KB
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2606.28639

Browse the full library · About the Tower of Babel