AI/ML Data Protection via Multi-Stage Validation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The use of third-party AI/ML services for business operations risks exposing proprietary data and raises security, privacy, and competitive concerns, as unrefined AI/ML models can produce hallucinations, and frequent model updates require costly retraining processes.
Innovation Solution
A multi-stage approach utilizing third-party AI/ML services to answer queries, validate answers through code execution within a proprietary system, and refine responses to ensure security, privacy, and accuracy, without sharing proprietary data, by parsing and reformulating answers to mitigate hallucinations and maintain control over proprietary information.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If proprietary data is used to fine-tune a baseline AI/ML model using third-party services, then the model accuracy and business-specific performance is improved, but security, privacy, and competitive concerns worsen due to data exposure risks
Solution Approach 1:
The patent extracts only the necessary model architecture and training methodology from third-party services, while keeping all proprietary data within the business's own infrastructure. This allows the business to benefit from advanced AI/ML capabilities without exposing sensitive information to external providers.
Solution Approach 2:
The patent introduces an on-premises AI/ML service as an intermediary layer between the business's proprietary data and the fine-tuning process. This intermediary enables model customization while maintaining data security by preventing direct data transmission to third-party services.
2Object-affected harmful factors
If a fine-tuned model is hosted on a business server to avoid data exposure, then security and privacy are improved, but the model may lack completeness and desirable features of third-party services
Solution Approach 1:
The patent performs preliminary fine-tuning of the AI/ML model using third-party services during the development phase, creating a pre-configured model that incorporates desirable features and completeness. The fine-tuned model is then deployed to on-premises servers, ensuring both security and model capabilities are maintained.
Solution Approach 2:
The patent separates the model development dimension from the model deployment dimension. Third-party services are used in the development dimension for comprehensive model creation, while on-premises servers handle the deployment dimension for secure operation, thus achieving both completeness and security.
3Productivity
If third-party AI/ML services are used directly to answer business questions, then service efficiency and productivity are improved, but hallucinations and wrong answers increase
Solution Approach 1:
The patent applies local quality by customizing the AI/ML model specifically for the business's domain and data characteristics through fine-tuning. This localized adaptation improves the model's understanding of business-specific contexts, reducing hallucinations while maintaining high service efficiency.
Solution Approach 2:
The patent implements feedback mechanisms where the fine-tuned model's performance is continuously monitored and evaluated against business-specific criteria. This feedback loop enables ongoing optimization of answer accuracy while preserving the efficiency gains from automated AI/ML service deployment.
Data Source
AI summary
Systems and methods are provided for utilizing AI/ML services to solve business problems while protecting proprietary information. Third-party AI/ML services are utilized in a multi-stage approach to answer a question and/or or solve a problem, create code to validate the answer, execute that code inside a proprietary system, and then use that answer to create a better answer without exposing proprietary data to a third-party.