Human-centered AI is a must-read for anyone in charge of designing a software system involving a significant artificial intelligence (AI) component. However, this book is not about how to implement an AI technique to solve a problem. It is about the issues around allowing humans to control such systems. As the author says, “human-centered AI ideas open up new possibilities for design of systems that offer high levels of human control and high levels of computer automation” (p. 43).
The concept of human-centered AI (HCAI) is not formally defined. However, it is explained with numerous examples of systems that could have been implemented better if human issues had been considered during the design process. In fact, this book could be used as a history of computer-related problems, from simple problems with autonomous vacuum cleaners to airplane disasters to moon missions. Many application areas are extensively analyzed.
Part 2 is about a HCAI framework. The classification of levels of autonomy is presented. A 2D model of levels of human control versus levels of computer automation is included. Understanding where a system fits on this 2D model helps clarify what control is appropriate.
Part 3, “Design Metaphors,” discusses four different design metaphors. Each metaphor influences how you would design the system. Issues with autonomy are discussed with many examples, including lethal autonomous weapons.
Part 4 is “Governance Structures.” These structures go from team-level guidelines to government agencies. The recommendations are intended to ensure reliability, safety, and trustworthiness.
The book consists of 280 pages of text, 46 pages of notes, a 31-page bibliography, a 7-page name index, and an 11-page subject index.
More reviews about this item: Amazon