AI-Generated Software Redundancy for Safety-Critical Control
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In safety-critical applications like vehicle control, existing methods for increasing software reliability, such as N-version programming, are complex and often fail to achieve sufficient diversity in AI-generated software versions, necessitating multiple versions to ensure reliability.
Innovation Solution
A method involving human programmers creating a primary computer program and multiple AI-generated programs to process inputs, with a safety measure initiated only if a predetermined threshold of AI-generated programs contradict the human-generated program's results, thereby weighting human-generated results more heavily and optimizing resource use based on safety requirements.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If N-version programming is used to increase software reliability, then reliability is improved, but device complexity increases significantly
Solution Approach 1:
The patent creates multiple copies of software programs (first program by human programmers, second programs by AI) to achieve redundancy for reliability. Instead of manually creating N versions, the system generates multiple AI-generated copies that can be compared against the human-created original, reducing the complexity of managing multiple versions while maintaining reliability through comparative validation.
Solution Approach 2:
The patent introduces an intermediary evaluation mechanism that compares results between the human-created program and AI-generated programs. This intermediary comparison process acts as a mediator to validate reliability without requiring direct management of complex N-version programming, simplifying the overall system architecture while ensuring software reliability through result verification.
2Reliability
If multiple AI-generated software versions are created to ensure diversity, then reliability is improved, but the effort and resources required increase
Solution Approach 1:
The patent employs AI systems that automatically generate multiple software versions and perform self-evaluation through result comparison. The AI-generated programs serve themselves by creating diverse versions and participating in the validation process, eliminating the need for extensive manual development effort while maintaining reliability through automated diversity generation and comparative assessment.
Solution Approach 2:
The patent implements a feedback mechanism where AI-generated programs are evaluated based on their results compared to the human-created program. This feedback loop allows the system to automatically assess reliability without requiring extensive manual verification, improving productivity by using automated feedback from the AI systems themselves to determine software quality and diversity.
3Productivity
If AI-generated programs are used to reduce manual effort, then productivity is improved, but manufacturing precision of software quality decreases
Solution Approach 1:
The patent uses the human-created program as a counterweight to balance the potentially lower quality of AI-generated programs. The human program serves as a reference standard against which AI-generated versions are compared, providing a quality anchor that compensates for AI limitations while allowing productivity benefits of automated generation to be realized.
Solution Approach 2:
The patent introduces an intermediary comparison process that mediates between AI-generated programs and the human-created program to ensure quality standards. This intermediary evaluation mechanism verifies that AI-generated software meets acceptable quality thresholds by comparing results against the human reference, maintaining manufacturing precision while benefiting from AI productivity.
Data Source
AI summary
A method for carrying out data processing. The method includes: creating, by one or more human programmers, a first computer program for a predetermined data processing task; creating a plurality of second computer programs for the predetermined data processing task, wherein each of the computer programs is created by an artificial intelligence; processing an input by the first computer program to ascertain a first processing result, and processing the input by each of the plurality of second computer programs to ascertain a respective second processing result; ascertaining a number of computer programs among the second computer programs whose second processing result contradicts the first processing result; checking whether the number of computer programs among the second computer programs whose second processing result contradicts the first processing result is greater than a predetermined threshold, which is greater than or equal to two; and initiating a safety measure.