From a human-computer interaction (HCI) perspective, the arrival of virtual assistants (for example, Alexa, Siri, Google Assistant, and Bixby) has led to a shift in interaction mode from visual to speech-based: instead of manipulating objects displayed on a screen, users instruct or query these assistants. Beneath the surface, however, the use of artificial intelligence (AI) methods has also influenced how users interact with these assistants and many other computational devices. In this paper, Amershi et al. use the term “AI-infused systems” to describe devices that expose AI capabilities to the user. A consequence of this shift is the difficulty of using design guidelines rooted in graphical user interfaces (GUIs), and to some degree in touch-based interaction, with limited AI capabilities in the devices.
Starting from a collection of AI-related design recommendations obtained through an extensive literature review, the authors use a systematic process to distill, revise, and validate a set of 18 human-AI interaction design guidelines. Ranging from initial expectations about the device (“make clear what the system can do”) to guidance during interaction (“show contextually relevant information”) to long-term aspects (“learn from user behavior”), it is my impression that the guidelines can be helpful in shaping user experience when interacting with such AI-infused systems. We are in the early stages of interacting with computational devices that have AI capabilities, and such a set of guidelines will hopefully lead to better solutions.
As an instructor of both HCI and AI courses, I am planning to incorporate the guidelines in two ways: as a basis for the thorough usability testing of devices with AI capabilities, and for the design, development, and assessment of team projects in these classes. While such projects rarely result in widely used products, the use of these human-AI design guidelines will at least raise student awareness of the underlying issues.