Optimal Neurocontrol: Practical Benefits, New Results and Biological Evidence

摘要

From late 1993 to March 1995, five groups have presented -- for the first time -- serious engineering accomplishments in simulations of control designs which meet four criteria for what is a "brain-like" intelligent system. In addition to demonstrated, intelligible engineering functionality, a "brain-like" system should contain at least three major general-purpose adaptive components: (1) an Action or Motor system, capable of outputting effective control signals to the plant or environment; (2) an "Emotional" or "Evaluation" system or "Critic," used to assess the long-term costs and benefits of near-term alternative outcomes; (3) an "Expectations" or "System Identification" component, which serves as a model or emulator of the external environment or of the plant to be controlled. This paper will: (1) summarize the practical advantages of moving from more classical designs to more brain-like designs, in today's engineering applications; (2) summarize the recent accomplishments of the 5 groups mentioned above; (3) argue that these designs (with some enhancement) do constitute a worthwhile theory of how intelligence works in the brain, generating testable predictions and proposed experiments; (4) scope out some key areas of research needed to implement the required enhancements in engineering.