Brain-Like Intelligent Control: From Neural Nets to True Brain-Like Intelligence
Paul J.Werbos
This past year has seen great progress in understanding stability and developing practical applications for new learning control designs, originally developed in the neural network area but applicable to a wider class of designs. However, there is still a big gap between the best designs implemented so far, and the kind of intelligence or higher-order decision-making capability seen in mammal brains, and required for multilevel control structures such as the management of entire factories or battlefields. This paper will briefly cite some of the prior work, but mainly summarize a new hybrid neural-AI design emerging from the underlying mathematics, rather than ad hoc synthesis – which has the potential to close that gap.[1,2].
The new design draws heavily on concepts of hierarchy and temporal chunking from AI, and on relational representation of objects and space, and a fuzzy goal representation.
[Contents]
|
