Statistical Proof of Execution (SPEX)
Michele Dallachiesa, Antonio Pitasi, David Pinger, Josh Goodbody, Luis Vaello
TL;DR
This work formalizes verifiable computing and introduces SPEX, a sampling-based protocol that provides robust correctness guarantees with significantly lower overhead than traditional methods and remains applicable to non-deterministic AI workloads. SPEX achieves verification by representing computations as a set of states encoded in Bloom filters, with configurable confidence $\delta$ and false-positive control via $FPR$. It also introduces a comprehensive toolkit of hashing-based techniques—integer, floating-point arrays, semantic embeddings, and LLM-generated agentic plans—to support non-deterministic and semantically rich computation states. The proposed design includes concrete solver/ verifier interfaces, specialized verification strategies (including lazy and adversarial defense), and an open-source Python implementation, enabling practical deployment in cloud and blockchain environments. Overall, SPEX enables auditable, scalable, and cost-efficient verifiable computing for AI-driven decision-making.
Abstract
Many real-world applications are increasingly incorporating automated decision-making, driven by the widespread adoption of ML/AI inference for planning and guidance. This study examines the growing need for verifiable computing in autonomous decision-making. We formalize the problem of verifiable computing and introduce a sampling-based protocol that is significantly faster, more cost-effective, and simpler than existing methods. Furthermore, we tackle the challenges posed by non-determinism, proposing a set of strategies to effectively manage common scenarios.