Engineering Intelligent Systems

1Artificial Intelligence, Science Fiction, and Fear

Artificial intelligence technologies such as machine learning and deep neural networks hold great promise for improving, even revolutionizing many application areas and domains. Curiously, experts in AI and casual observers line up on both sides of the are AI benefits worth the risks? question. Several books from prominent AI researchers paint dire scenarios of AI systems run amok, escaping the control of their human creators and managers, pursuing their “own” agendas to our detriment. At the same time, AI research races ahead, developing new capabilities that far surpass the performance of past systems and even humans performing the same tasks. How can we resist these advancements and the benefits they bring, even though there may be risks?

The way out of the dilemma is the application of systems engineering. Systems engineers have been addressing the issues of dangerous technologies for decades. Nuclear fission, like AI, is an inherently dangerous technology. Systems engineers can't make fission safer, so instead they build systems around the fission reaction, making the entire system as safe as possible. If a mishap occurs, the fault is not with fission itself, but with the design or implementation of the entire system.

This chapter looks at some of the main challenges in the development of intelligent systems – systems that include one or more AI‐based components to produce intelligent behavior – including reliability, ...