Asimovian Adaptive Agents

The goal of this research is to develop agents that are adaptive and predictable and timely. At rst blush, these three requirements seem contradictory. For example, adaptation risks introducing undesirable side eeects, thereby making agents' behavior less predictable. Furthermore, although formal veriication can assist in ensuring behavioral predictability , it is known to be time-consuming. Our solution to the challenge of satisfying all three requirements is the following. Agents have nite-state automaton plans, which are adapted online via evolutionary learning (perturbation) operators. To ensure that critical behavioral constraints are always satissed, agents' plans are rst formally veriied. They are then reveriied after every adaptation. If reveriication concludes that constraints are violated, the plans are repaired. The main objective of this paper is to improve the eeciency of reveriication after learning, so that agents have a suuciently rapid response time. We present two solutions: positive results that certain learning operators are a priori guaranteed to preserve useful classes of behavioral assurance constraints (which implies that no reveriication is needed for these operators), and eecient incremental reveriication algorithms for those learning operators that have negative a priori results.