Distributed Ledger AI Model Verification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Organizations face challenges in verifying the results of computations made by third-party AI models without disclosing the inner workings of these models, especially in scenarios where discrimination based on personal identifiable information (PII) is a concern, and legal restrictions apply.
Innovation Solution
A method and system that utilize a distributed ledger to represent and store AI models, allowing users to verify computation results without accessing the model's inner details, using zero-knowledge proof models and smart contracts to ensure immutability and secure execution within the ledger.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If organizations spend large resources on creating and maintaining AI models to maintain competitive advantage, then the model's accuracy and decision-making capability are improved, but the model's secrecy and inability to verify results deteriorates
Solution Approach 1:
The patent segments the AI model into two distinct parts: a public representation stored on the distributed ledger that enables verification of computation results, and the private original model that maintains competitive advantage. This segmentation allows organizations to share verification capabilities while preserving model secrecy.
Solution Approach 2:
The patent creates a copy of the AI model's computation logic and stores it on the distributed ledger as a verifiable representation. This copy enables users to verify results without accessing the original model, thus maintaining both reliability and secrecy.
2Loss of information
If organizations make AI models opaque to protect competitive advantage, then model secrecy is improved, but the ability to verify compliance with legal restrictions deteriorates
Solution Approach 1:
The distributed ledger acts as an intermediary between the private AI model and the public verification process. It stores a verifiable representation of the model's computation logic, enabling compliance verification without exposing the original model's inner workings.
Solution Approach 2:
The patent separates the compliant verification function from the proprietary model logic by storing the computation representation on the distributed ledger. This allows verification of legal compliance while maintaining model secrecy through the segmentation of public and private components.
3Ease of operation
If users are provided access to the full AI model to verify computation results, then verification capability is improved, but the model's competitive advantage and secrecy deteriorate
Solution Approach 1:
The patent provides users with access to a copy of the model's computation representation stored on the distributed ledger, rather than the original proprietary model. This copy enables verification operations while preserving the original model's secrecy and competitive advantage.
Solution Approach 2:
The patent segments access rights by providing users with the public verification representation on the distributed ledger while keeping the private original model inaccessible. This segmentation enables ease of verification without compromising model secrecy.
4Productivity
If AI models are deployed without distributed ledger representation, then deployment speed is improved, but the ability to provide auditable trace and verify results deteriorates
Solution Approach 1:
The patent performs preliminary action by storing the verifiable representation of the AI model's computation logic on the distributed ledger before the model is deployed for actual use. This enables auditable trace and result verification to be established in advance without significantly delaying deployment.
Data Source
AI summary
Examples relate to a method, a computer program and a system for enabling a verification of a result of a computation obtained from a third party. The method comprises providing a model on a distributed ledger, the model representing the computation used by the third party. The method comprises providing access to the model for one or more users, to enable the users to verify the result of the computation.