GRADIENT DESCENT TRAINING FOR DEFENSIBLE ARTIFICIAL INTELLIGENCE (RFT 653)

Invention Summary

Artificial Intelligence (AI) systems are widely used in various sectors for critical decisions but often lack transparency in their internal decision-making processes. This technology employs a gradient descent training method within expert systems to optimize network structures for AI applications. The method focuses on identifying and weighting rules and facts, significantly improving decision-making transparency. By training AI to recognize contributions of specific rules and facts, the system ensures that each decision point within the network can be audited and understood by humans, thereby making the AI defensible and reliable.

Benefits

Enhanced Transparency: Provides clear visibility into AI decision-making processes
Increased Reliability: Ensures decisions are based on valid and human-approved rules and facts.
Reduced Legal Risk: Minimizes liability by avoiding decisions based on biased or non-causal associations.
Scalability: Adaptable to various fields with requirements for defensible decision systems.

Applications

Healthcare: Diagnosing diseases based on transparent decision pathways
Finance: Transparent credit scoring and loan approval processes.
Legal: Assist in evidence-based legal analysis such as law applicability, law/precedent searching, and providing sentencing recommendations
Social Media: Monitoring and managing content with defensible filtering rules

Patent

This technology has a U.S. Patent Pending US 2023/0281/452 A1 and is available for licensing/partnering opportunities.

Contact

NDSU Research Foundation
info(at)ndsurf(dot)org
(701) 231-8173

NDSURF Tech Key

RFT, 653, RFT653

Inquire about this technology >