IIC Journal of Innovation 11th Edition | Page 84

AI Trustworthiness Challenges and Opportunities Related to IIoT data science teams while also ensuring critical content stays safe and protected from the outside world. characteristics (security, privacy, reliability and resilience). To address the trustworthiness challenges, it helps to partition the AI use cases that are emerging into two categories: AI IN I NDUSTRIAL S YSTEMS Industrial operation is facing a shrinking decision timeline, so when it comes to the application of AI in industry, it is not enough for AI to simply pass the proverbial Turing Test. 5 This is because human-like performance can at times be immoral and lead to unacceptable outcomes, as evidenced by malware, ransom-ware and terrorism to mention a few. This means that in IoT, a naive approach to AI is unacceptable, especially since people place higher expectations on automated systems 6 . 1. The use of AI to improve the efficiency, reliability and effectiveness of processes and tasks that can be fully automated with little risk. These are processes and tasks that are generally mundane, repeatable, static with few variations, or tasks that are very specific and/or localized to specific components in system. 2. The use of AI in processes that are critical, consequential 8 9 and non- mundane. When the level of risk is high enough, humans must maintain the ultimate decision-making capacity – this is referred to as the “human-in-the-loop” approach or HIL. Just as with other systems, designers of AI must address regulations, laws and established best practices, especially with regards to safety, privacy and security. They need to consider the need to be able to explain the decisions of systems and how they are reached, not only to avoid inappropriate bias, but also to create systems that can be trusted, through evidence and audit. Such an approach is essential to addressing safety concerns, 7 as well as other IoT Trustworthiness Consider these two categories are part of designing for trustworthiness. The basic principle is that trustworthiness characteristics (safety, security, privacy, reliability and resilience) cannot easily be 5 The Turing Test is a test of a machines ability to exhibit intelligent behavior equivalent to or indistinguishable from that of a human. 6 Car accidents of autonomous cars get much higher attention in society than car accidents caused by humans. 7 “Key Safety Challenges for the IIoT”, IIC White Paper, 1 December 2017, https://www.iiconsortium.org/pdf/Key_Safety_Challenges_for_the_IIoT.pdf 8 Patient X-rays analysis, autonomous driving, etc. 9 US DoD Directive 3009.09: Establishes DoD policy and assigns responsibilities for the development and use of autonomous functions in weapon systems, and establishes guidelines to minimize the probability and consequences of failures in such autonomous systems - https://www.hsdl.org/?abstract&did=726163 IIC Journal of Innovation - 80 -