Implementation and Evaluation of a Gradient Descent-Trained Defensible Blackboard Architecture System

Jordan Milbrath,Jonathan Rivard,Jeremy Straub

2024-04-18

Abstract:A variety of forms of artificial intelligence systems have been developed. Two well-known techniques are neural networks and rule-fact expert systems. The former can be trained from presented data while the latter is typically developed by human domain experts. A combined implementation that uses gradient descent to train a rule-fact expert system has been previously proposed. A related system type, the Blackboard Architecture, adds an actualization capability to expert systems. This paper proposes and evaluates the incorporation of a defensible-style gradient descent training capability into the Blackboard Architecture. It also introduces the use of activation functions for defensible artificial intelligence systems and implements and evaluates a new best path-based training algorithm.

Artificial Intelligence

What problem does this paper attempt to address?

The main problem this paper attempts to address is the application of gradient descent training methods to a Blackboard Architecture with practical execution capabilities, in order to enhance its interpretability and defensibility. Specifically, the paper proposes the following points:

1. **Combining Gradient Descent Training with Blackboard Architecture**: Traditional expert systems and neural networks each have their own advantages and disadvantages. Expert systems are easy to understand and interpret but lack adaptive capabilities; whereas neural networks, despite their powerful learning abilities, lack transparency and interpretability. The paper proposes a new system that combines the advantages of both by introducing gradient descent training methods into the Blackboard Architecture.

2. **Adding Activation Functions**: To further enhance the system's performance, the paper introduces activation functions, enabling it to handle more complex tasks and data.

3. **Optimal Path Training Algorithm**: The paper proposes an optimal path-based training algorithm to optimize the weights in the rule-fact network, thereby improving the system's accuracy and efficiency.

4. **Improving System Interpretability and Defensibility**: Through the aforementioned improvements, the paper aims to enhance the system's interpretability and defensibility, making it more reliable and trustworthy in practical applications.

In summary, the goal of the paper is to improve the Blackboard Architecture so that it can not only provide decision support but also execute decisions practically, while maintaining a high degree of transparency and interpretability during the training process.