In Inventing the Medium I devote a chapter to the Companion Model as one way of scripting the interactor. I talk about the recurring fantasy of the computer as a perfect mind-reading servant. Siri, the voice-activated app on the iPhone, is the most ambitious attempt so far at creating a servant companion and its successes and failures illustrate the principles in this chapter.
The most dangerous part of Siri’s design is its use of spoken natural language, which leads to the high expectations similar to communication with a human being and therefore risks frustrating the interactor. Siri can be surprisingly responsive, remembering the subject from one utterance to the next as in this example where it remembers that I am talking about the weather:
But it can also be annoyingly dense. For example it seems unaware that it is part of a particular phone and service network so it cannot answer questions like “How do I set up my voicemail?” except with a generic “voicemail” web search. And since it lives in the cloud, it is unable to set a local timer if it can’t connect to the network. Siri also lacks the kind of semantic context that makes variant utterances recognizable to human beings. It does not recognize “dermatologist appointment” can be on the calendar as “dermatology doc,” nor will it let me teach it that one is the same as the other.
The design team have compensated for Siri’s inevitable misunderstandings by the visibility of its processing: It transcribes your utterance into text and it tells you what it is doing. It also explicitly scripts the interactor in computer-accessible phrases, suggesting commands it understands, like play a piece of music or do a web search.
These interaction design strategies and the pretty good effectiveness of its voice recognition are enough to make it viable as a product, but it remains risky as a source of consumer frustration (like the 1993 Newton handwriting application) and — more seriously — of disastrous miscommunication since we have a natural tendency to think of it as more sentient than it is, a tendency that it is programmed to encourage by participating in our politeness rituals. The commercials encourage children and people with emergencies like a flat tire to rely on Siri, which is a potentially dangerous strategy. Since human beings have a strong tendency to think of computers as magical companions, it is the designer’s job not just to make them responsive and reliable, but also to remind us that they are only dumb machines.