Eleni Stroulia and .Ashok K. Goel
Broadly characterized, learning can improve problem-solving performance by increasing its efficiency and effectiveness, and by improving the quality of produced solutions. TraditionalAI systems have limited the role of learning to the first two performance-improvement goals. We have developed a reflection process that uses a model of the system’s functional architecture to monitor its performance, suggest a quite broad range of modifications when it fails, and subsequently perform these modifications to improve its problem-solving mechanism. The modifications suggested and performed by the reflection process may result in performance improvement of all the above types.