AI Taking Actions in the Physical World - Strategies for Establishing Trust and Reliability
Sirko Straube, Ryuji Yamazaki-Skov, Raul Hakli
Frontiers in Neurorobotics, volume 17, Apr/2023.
Since making AI reliable and establishing trust is so important in all kinds of possible applications, we therefore addressed a wide field of research activities. This kind of broadness is in retrospect both the strength and the weakness of this Research Topic. We see from the results that the intended theme indeed shows that we are looking from very different perspectives on the same issue and it illustrates the heterogeneity of the approaches that are currently under research. On the other hand, we need more integrative and unified AI approaches that are also interoperable, so that one application can interface another one. A high level of reliability of and introspection on AI would be a fundamental building block for such a unification of AI approaches, since communication of uncertainties or even reasoning about failures would support interfacing with AI systems and, most importantly, strengthen the trust a human user would establish in the application (in contrast to a black box functionality). However, such a unified AI concept or strategy is still not common sense, and we can also not derive the overall tendencies from the current issue.Still, by looking at the contributions in the Research Topic from a higher perspective, some principles of current research directions are seen very nicely. Reliability is always depending on knowledge when an AI (or a robot) encounters unknown situations.