AI that is pre-programmed is limited in its tasks and human bias. Learning systems offer richer decision-making behaviors where collaborative projects have led to the following three systems that require integration:
- A symbolic learning system that can continually learn Boolean classification problems as they are presented to it. But this needs to be extended to real-valued, noisy and uncertain classification problems.
A lateralized system that can consider an input at the constituent level and the holistic level simultaneously, which enables flexible and abstract learning. How lateralisation functions in continual learning system requires research, e.g., if we learn the concepts of ‘nose’ & ‘eyes’, how does this help in identifying types of animal?
A compaction algorithm exists that can remove redundant and irrelevant learnt knowledge, which helps with system scaling. But research is needed in preventing catastrophic forgetting in continual learning systems.