Designing a Smart Car Virtual Assistant
Cognitive load, exit strategies, and standing on the shoulders of giants.
Design&AI is a monthly meetup exploring case studies and the nitty-gritty decisions designers face when “making” for an AI-augmented world. Last month, we were joined by TribalScale’s Jane Motz Hayes (Director of Design) and Geoffrey Hunter (Data Sciences Lead), who started off the conversation by sharing their work developing a smart car console that uses natural language processing. Below is a roundup of some reflections and topics discussed.
Smart Car Console: a search and parking assistant
TribalScale worked with a client partner to design a smart car console with a feature that enable drivers to search for directions and parking using voice commands while driving their car. The car console uses natural language processing — not unlike virtual assistants such as Alexa, Cortana, or Siri — to analyze voice utterances (“Find me parking at Yonge & Eglinton”) and turns them into computational meaning. Here are the top four reflections that Jane and Geoff shared with us, about the considerations that went into their design process.
1. What is the cognitive load of the user?
It’s important to consider where the user will be and what else they will be doing while using the technology. That’s because, the concentration level of someone hanging out at the cottage versus the concentration level of someone driving down the highway is substantially different. While at the cottage, you may be asking Alexa for the PGA scores while scraping food off your plate or making a sandwich. While driving down the highway, you may be asking the car assistant for directions while juggling arguing kids in the backseat and navigating busy traffic. In the driving use-case, the voice utterances (i.e. commands) need to be very simple and direct. If the car assistant doesn’t understand the user’s command, clarification prompts need to be binary or in threes so that the driver doesn’t get frustrated, confused, or distracted. When engaging with an assistant while driving, prompts and commands need to be incredibly simple and clear so that users aren’t put at risk of getting into an accident.
2. How do you enter and exit the modality?
The car assistant needs a signal to know when it’s being engaged. Are you asking the assistant a question? Or, rather, are you talking to one of your passengers, singing along to the radio, or letting another driver know what’s up with their driving? In the early days of Cortana, some Cortana users complained that the virtual assistant would randomly interrupt with a search result without having been prompted to search for anything in the first place — for example when they were playing video games or talking with a friend. Being interrupted in this way not only feels rude and intrusive, it can also feel uncomfortable to realize the assistant was listening in that whole time (great, now Microsoft knows I have jury duty next week). In creating the car assistant, the team chose a clear signal (for example, “Hey car”) to enter the modality, and to have a response from the assistant to signal to the user that it is ready for your command (for example, “How can I help you”). Having simple voice enabled utterances to signal when you want to enter and exit the modality creates a smoother user experience.
3. What are the different ways people ask for things?
It is no surprise that people in different parts of the country and world have different ways of asking for the same things. And these variations have implications for what short commands to program into the assistant. Jane shared that user testing uncovered a bunch of variations in wording, including: Stop vs. Cancel; Begin Again vs. Start Over; 4pm vs. 4 o’clock; providing a street addresses vs. an intersection. Furthermore, differences in the phrasing of queries also has to do with people’s mental models and habits. Designing a global product means baking in ways of handling regional customization and labelling, and it means continuing to user-test across regions where the product will be rolled out.
4. Where can you build on the work of others?
In designing the car assistant, Jane and Geoff explained that it was not the time to reinvent the wheel. The experience needed to be familiar to the people using it — so that users can concentrate on their main task: driving! Borrowing and building from the conventions that other companies already have in place — and aggressively user test at each stage of the new build — is how they created a user experience that is familiar, intuitive, and easy to use.
Are you designing a type of voice user interface? What design decisions are you grappling with? What applications of natural language processing are you excited about lately?
This blog post is the summary of a rich group discussion at the design&AI meetup, attended by a small group of Toronto-based design and technology leaders to chat about the nitty gritty decisions that go into designing for an AI-augmented world. While the event is currently invite-only, we will be opening it up to a larger group in the coming months. In the meantime, please do share any comments or questions in the comments section, or get in touch via hello@normative.com.
