The Deterministic Horizon: Impossibility Results as Design Specifications for Trustworthy AI Systems
The paper discusses the limitations of AI systems, particularly large language models, and proposes design specifications based on impossibility results. It introduces the concept of a 'Deterministic Horizon,' which sets a ceiling on accuracy determined by architecture. The findings suggest that every fundamental limit of AI can serve as a design rule for developing trustworthy AI systems.
- ▪The research identifies a critical reasoning depth beyond which no training can improve accuracy in AI models.
- ▪The Deterministic Horizon is quantified between nineteen and thirty-one across twelve transformer architectures.
- ▪The study provides a catalogue of sixteen specifications that link computable boundaries with design rules for AI.
Opening excerpt (first ~120 words) tap to expand
Computer Science > Artificial Intelligence arXiv:2605.23024 (cs) [Submitted on 21 May 2026] Title:The Deterministic Horizon: Impossibility Results as Design Specifications for Trustworthy AI Systems Authors:Dongxin Guo View a PDF of the paper titled The Deterministic Horizon: Impossibility Results as Design Specifications for Trustworthy AI Systems, by Dongxin Guo View PDF Abstract:Large language models now write software, draft legal documents, and produce clinical notes, yet fundamental limits, from Turing and Arrow to the No Free Lunch theorems, shape what computation can do. This thesis turns such impossibility results from curiosities into design rules.
…
Excerpt limited to ~120 words for fair-use compliance. The full article is at arXiv cs.AI.
Discussion
0 commentsMore from arXiv cs.AI
