In today's AI landscape, decision-making systems often act as mysterious black boxes, particularly in sensitive fields like healthcare and finance where transparency matters. One way to address this could be through an approach that combines manual knowledge engineering with modern database technologies, creating AI systems whose reasoning is fully traceable to human-defined rules and relationships.

Structured Knowledge for Clear Reasoning

At the core of this approach would be a relational database carefully designed by domain experts to capture all relevant knowledge in their field. For instance, a medical diagnostic system might have tables for symptoms, diseases, tests, and their interrelationships. The AI's decision logic would then be implemented as explicit rules that query this database, with every conclusion directly traceable to specific entries and rules. This differs from traditional black-box AI in several key ways:

• Instead of learning patterns from data, the system applies human-crafted logical rules
• Each decision can be explained by showing which database records and which rules led to it
• Domain experts maintain full control over how knowledge is represented

Addressing Regulatory and Practical Needs

This approach could be particularly valuable in regulated industries facing "right to explanation" requirements. For financial services, it could show exactly why a loan application was declined. In healthcare, it could explain diagnostic conclusions by referencing medical guidelines encoded in the system. The database-backed structure would offer advantages over older expert systems through:

• More sophisticated knowledge representation using modern database capabilities
• Better tools for maintaining and updating the knowledge base over time
• Clearer audit trails showing how information flows through the system

Implementation Pathways

Testing this concept might start with a narrowly defined domain where expert knowledge is well-established, like basic credit assessments. Initial versions could focus on capturing the most common decision rules and demonstrating how explanations would work. If successful, the approach could grow to include:

1) Mechanisms for handling partial or conflicting information
2) Tools to help experts maintain and expand the knowledge base
3) Optional learning components that suggest new rules while preserving explainability

While potentially less accurate than opaque machine learning systems in some cases, this transparent approach could find adoption wherever the ability to explain decisions is as important as the decisions themselves. By building on decades of expert system research but applying modern database techniques, it might offer a path to AI that professionals can actually understand and trust.