Abstract: The point of departure for this talk is the literature on High Reliability (human) Organizations. Human HROs provide us with examples of how a system can be robust and reliable in an open world. We will extract a set of requirements for AI systems and then focus on a few technical contributions. The first is to provide prospective guarantees for the future behavior of the AI system (under the assumption that the environment is stationary). The second is to detect anomalies through violated expectations. The talk will conclude with a few additional pieces of the puzzle including speculations about how an AI system might anticipate changes in the world and adapt in advance instead of reacting to changes once they occur.

Abstract: The talk will concern methods that allow robots to learn from, adapt to and forecast the perpetual changes occurring in natural environments. We will show that embedding cyclic, rather than linear, representations of time into traditional environment models used by robots provides them with the ability to forecast future states of changing environments. We will review the results of several long-term experiments that evaluated the impact of these methods on the efficiency of robots' autonomous operations over time. Furthermore, we will present the integration of these methods into systems created in response to the coronavirus crisis and rapid ecosystem decline.