Intelligent Systems

AI engineers seek to automate mental processes. Of these processes, only some have a deliberative component based on declarative mental representations. The two commonest declarative forms are visuo-spatial images and propositions. Below introspectahle level are found the tacit skills of the highly trained expert. In real-time control tasks these tend to dominate. Since they are also largely subarticulate, the road to knowledge acquisition at first sight seems blocked. Modern machine learning tools can, howE'ver, recover symbolic models of skills from behavioural traces. The resulting data-derived "clones" show transparency and run-time dependability beyond the capacities of their human exemplars. An aerospace application is reviewed.