“Why Did You Help Me?”: Making Virtual Assistants Understandable

What is it about?

Voiced virtual assistants, also known as Intelligent Personal Assistants (IPAs), have become increasingly ever-present in recent years with commercial examples including Amazon Alexa, Google Assistant, and Apple Siri. However, these assistants remain used almost exclusively for simple tasks such as setting alarms and providing information about the weather. Moreover, their use-case is reactive and the systems require a cue word (e.g. "Hey, Siri", "Okay Google", etc.) to start assisting the user. In contrast, this paper proposes an envisioned IPA that would be proactive -- actively anticipating the user's needs and desires -- and initiating conversations or actions without an explicit cue. An example would be reminding the user that she will be late unless she leaves home soon or opening a window when the quality of air indoors drops. Moreover, the envisioned system would be delegative, capable of contacting others on the user's behalf. We then argue that such a system would have to be scrutable, i.e. it must allow the user to study the modelled knowledge.

Featured Image

Photo by Clay Banks on Unsplash

Photo by Clay Banks on Unsplash

Why is it important?

The envisioned IPA's capabilities have the potential to cause broader consequences for the user (including social and financial). Therefore, we argue that scrutability -- i.e. the user's ability to study or scrutinise the system towards tunning the system's personalised behaviour -- would emerge as an essential element of the envisioned human-assistant interaction. Without it, the risk of misunderstanding the user's goals and environment would be higher. Moreover, a scrutable IPA might help the user trust the system more.

Perspectives