Paper Number

ICIS2025-2480

Paper Type

Short

Abstract

Explainability is critical for human-AI decision-making, but methods centered on faithful and plausible explanations fail for complex Large Language Models (LLMs). This paper proposes a new paradigm that replaces the goal of model transparency with the goal of interaction verifiability. We frame human-LLM collaboration as a deliberation and introduce a formal Interactive Human-LLM Deliberation Protocol that incorporates human cognitive limitations. We then prove that this protocol functions as an Interactive Proof Protocol and can be extended to an Interactive Zero-Knowledge Proof (ZKP) Protocol. This framework shows that trust in LLMs can be established through rigorous, structured dialogue and recasts the human user's role as an "Effective Verifier" responsible for verifying and challenging the model's claims.

Comments

12-GenAI

Download

Share

COinS
 
Dec 14th, 12:00 AM

Human-LLM Deliberation as an Interactive Zero-knowledge Proof Protocol

Explainability is critical for human-AI decision-making, but methods centered on faithful and plausible explanations fail for complex Large Language Models (LLMs). This paper proposes a new paradigm that replaces the goal of model transparency with the goal of interaction verifiability. We frame human-LLM collaboration as a deliberation and introduce a formal Interactive Human-LLM Deliberation Protocol that incorporates human cognitive limitations. We then prove that this protocol functions as an Interactive Proof Protocol and can be extended to an Interactive Zero-Knowledge Proof (ZKP) Protocol. This framework shows that trust in LLMs can be established through rigorous, structured dialogue and recasts the human user's role as an "Effective Verifier" responsible for verifying and challenging the model's claims.

When commenting on articles, please be friendly, welcoming, respectful and abide by the AIS eLibrary Discussion Thread Code of Conduct posted here.