AI: Unexplainable, Unpredictable, Uncontrollable

Roman V. Yampolskiy · 2024 · Chapman & Hall/CRC (AI and Robotics Series) background medium priority coded

Main argument

Thesis (book-level): advanced AI systems are, in principle and increasingly in practice, UNPREDICTABLE (their decisions cannot be anticipated even by their designers), UNEXPLAINABLE/INCOMPREHENSIBLE (no explanations both accurate and humanly graspable), UNVERIFIABLE (no proof of safety/alignment properties for arbitrary systems), UNOWNABLE, and UNCONTROLLABLE (no mechanism guarantees ongoing human control at superhuman capability) - with chapters on pathways to danger, accident taxonomies, personhood, consciousness, and skepticism replies. The control problem is thus not an engineering backlog but a cluster of impossibility results.

Why it matters here

The impossibility-results compendium: chapter-length arguments that advanced AI is unpredictable, unexplainable, unverifiable, unownable, and uncontrollable - the strongest sustained pessimist position, from within AI safety. For the dissertation: each impossibility claim strikes a condition on responsibility (foreseeability, explanation, verification, ownership, control), so if Yampolskiy is even partly right, the responsibility-attribution problem is permanent, not transitional.

Reading notes

Targeted treatment (265pp): chapter map + framing read (Chs 2-6 impossibility arguments; Ch 9 Personhood; Ch 12 Human≠AGI). Note MELO's undecidability result is the peer-reviewed formal core of Yampolskiy's unverifiability chapter; Lundgren cites him for the control problem.

Yampolskiy, R. V. (2024). AI: Unexplainable, Unpredictable, Uncontrollable. Chapman & Hall/CRC.

Close reading — 1 coded units

#1 · pp. 15 · claim

“[Chapter structure as argument: Unpredictability (Ch 2), Unexplainability and Incomprehensibility (Ch 3), Unverifiability (Ch 4), Unownability (Ch 5), Uncontrollability (Ch 6) - each developed as an in-principle limitation of advanced AI, followed by Pathways to Danger, Accidents, Personhood, Consciousness, and replies to Skepticism.]”

RL-CONTROL

Synthesis-matrix row

complicates T6-RESPONSIBILITY-UNALLOCATED

impossibility claims strike each responsibility condition

Memos (1)

theoretical · unit #1

Use as the pessimist ceiling in the responsibility chapter: standard responsibility conditions (control - Fischer & Ravizza; foreseeability - negligence law; explanation - Kästner's MI regime) are each targeted by one of Yampolskiy's impossibility claims. Two consequences: (a) if he is right, the responsibility gap is PERMANENT and the only honest allocation is ex-ante (development/deployment decisions - who chose to build/deploy under acknowledged uncontrollability), which converges with H&D's governance turn and strict-liability regimes (PLD); (b) if he is wrong in degree (Kästner's MI, Melo's alignment-by-construction), responsibility tracks the achievable degree of control/explanation - which is the graded picture the dissertation defends. Either way, cite the chapters not the vibe: Ch 4 (unverifiability) pairs with MELO's theorem; Ch 6 (uncontrollability) with LUNDGREN.