Generative AI Prompting With Sensitive Data Abstraction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Generative AI systems face issues with data leakage, bias introduction, and high resource consumption, particularly when handling sensitive data, leading to reduced accuracy and increased costs.
Innovation Solution
A method and system that preprocesses sensitive data by reducing it to enumerated outputs, expanding these outputs into textual descriptions, and combining them with non-sensitive data to form prompts for the generative AI, ensuring secrecy and preventing bias while minimizing resource usage.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If the generative AI system provides detailed reasoning for its decisions, then the explainability and usefulness of the output is improved, but sensitive data may be leaked
Solution Approach 1:
The patent extracts only the necessary classification result from the sensitive data, discarding the actual sensitive content while retaining the classification outcome. This allows the system to provide useful output without exposing sensitive information that would otherwise be leaked through detailed reasoning.
Solution Approach 2:
The patent creates a simplified copy of the sensitive data that contains only the classification-relevant features (e.g., file extension, size category) without the actual sensitive content. This copy is then processed by the generative AI to produce reasoning based on non-sensitive attributes.
2Reliability
If large amounts of input data are provided to the generative AI, then the model has more context to work with, but resource consumption and costs increase
Solution Approach 1:
The patent extracts only the essential features from the input data that are relevant for classification (e.g., file extension, size category, name patterns), discarding unnecessary sensitive details. This reduces the input data volume while maintaining classification accuracy.
Solution Approach 2:
The patent applies different levels of data processing to different parts of the input: sensitive content is heavily processed/removed, while non-sensitive metadata is preserved in full detail. This localized approach optimizes resource usage by processing only what is necessary.
3Measurement precision
If sensitive data is provided to the generative AI for classification, then the classification accuracy is improved, but bias may be introduced into the model
Solution Approach 1:
The patent extracts and removes bias-prone sensitive attributes (e.g., personal identifiers, sensitive metadata) from the input data before processing, while retaining the features necessary for accurate classification. This eliminates bias introduction while preserving classification capability.
Solution Approach 2:
The patent introduces an intermediary processing step that transforms sensitive input data into a bias-free representation. This intermediary layer (the data processing component) acts as a mediator between the raw sensitive data and the generative AI, ensuring the model receives only neutral, classification-relevant features.
Data Source
AI summary
A method for evaluating data by a generative artificial intelligence (AI) model to determine for an original query containing sensitive data a result and a reason for that result without leaking substantially any of the sensitive data, the method comprising: separating non-sensitive data of the original query and at least one type of the sensitive data; reducing each respective one of the at least one type of sensitive data to one enumerated output selected from a prescribed number of options for that respective type of sensitive data; expanding each enumerated output to a respective expanded form that is usable by the generative AI model; combining the expanded forms with the non-sensitive data to form a prompt; and submitting the prompt to the generative AI model.