REFLECTIONS ABOUT ADAPTIVE SYSTEMS FOR OP ERATOR ASSISTANCE

The automation of complex systems can lead to situations of man-machin e mismatch, propitious to human errors and accidents. Many types of co unter-measures are foreseen like operator training, man-system interfa ce improvement and use of on-line expert systems. Improving the collab oration between operators and automation is another kind of measure. T he philosophy of assistance considered in this article consists of pre serving or supporting the operators control and risk management strate gies and of maintaining opportunities for learning-by-doing, while dev eloping safety envelopes aimed at filtering any erroneous command and allowing automated responses in the absence of human command The pract ice of such a philosophy requires a technological intelligence capable of adapting itself to operators. Systems with such properties belong to the class of adaptive systems. Without claiming to be exhaustive, t his article reviews research and Practical applications which use this paradigm Adaptive systems use models of their environnment or of thei r users (including their behaviours or intentions) in order to adapt t hemselves dynamically to the actions and strategies of the users. Most of them also demonstrate anticipation capabilities. Adoptive assistan ce takes advantage of these properties, by creating conditions propiti ous to a human-like collaboration between agents. This paradigm featur es indeed some properties of human interactions, in which each actor b uilds and uses models of his partners in order to understand and antic ipate their activity. The intelligibility, the predictability and fina lly the reliability of the interactions however depend on these models ' coherence and expressiveness. As any assistance system is itself fal lible or of limited application, the last part of the article discusse s some side-effects of adaptive assistance, in terms of induced risks and user acceptance. It concludes by recognising the promising but als o very demanding character of the line of research.