A New Approach to Voice User Interfaces
I was very much involved in the development of Mobile phone user interfaces 15 years ago. One of the struggles we had was the assumption, at the time, that you could just modify existing PC based interfaces by changing the size of the UI components, but keeping everything else the same. This incorrect approach lasted nearly 5 years. Eventually it was realised Mobile UIs had to be designed from scratch and hey presto, these simplified and easy to use interfaces could then be used on PC interfaces with few changes (and they were better than the original)— this approach was called Mobile First design. It really only became obvious once Mobile began to dominate usage.
I feel history is repeating itself with Voice UIs. This time people have already started to use the phrase Voice First, but there is a legacy of text chatbot interfaces playing a bad influence on the Voice design. Again text chatbots are, today, the dominant chatbot type. But it’s obvious, Voice (being the natural UI to computers) will become the dominant UI eventually. Large companies are incorrectly concentrating on text chatbot design, first, in the thought that the same design will work with voice — it won’t. A good example of how bad this approach is, is in Dialog Flows. A Dialog Flow is a hierarchy of questions and answers which take the user through a hierarchy of contexts to hone in on the issue at hand — its a slightly friendlier approach to press 1 for ‘x’ and press 2 for ‘y’ used in IVR phone systems. It’s clearly an incorrect approach and is frustrating for the users. Even in a text chatbot environment this is clearly a bad approach.
Since the phrase Voice First is already in use, we have created a concept of “Choiceless Voice First” UIs. This is where we remove the choices from the user and further go out of our way to minimise the things the user has to say in order to get the information he/she is looking for. Furthermore, we believe once you have designed a “Choiceless UI”, this UI will be applicable to both Voice and text. The end result is a very usable system.
But, how can we implement a choiceless system — surely we need mind reading! The answer is no, we have new AI techniques to help. AI can look at patterns in average users and in each individual user preference too. This is how we’ve been implementing Choiceless Voice First UIs.
An example of one of our new implementations using this choiceless technique is the “Business Elf” skill available on Amazon Echo in US and UK (so far). In its simplest form this is an RSS reader of business new stories using several feeds from different sources (BBC, CNBC, Reuters, Internet Retailing, Business Insider, Harvard Business Review and the Economist), providing hundreds of new stories per day. The user is not told its an RSS reader, they are told it has business stories and the system just reads a single story(no choice). The story it reads is specific to that user using weighting of their interests. We acquire their interest by seeing if they want to read the full articles. The full articles can be added to a basket for sending to an email for later reading on their phone. We envisage a use case of the user waking up in the morning and running Business Elf on their bed side Echo. They listen to the stories selected by Elf for them and add the interested ones to their basket. When they finish they say “send and stop” — its a single command which sends the email and stops the skill (part of our principle of minimising user activity).
To see a video of the user experience watch this video .
To run the skill on your Echo search for “Business Elf” in the Alexa app.
This article is part of a book I am writing about computer engineering techniques to create an AI capable of conscious thinking which will be available in 2019 and is in support of my Youtube Channel called “AI with Jargon” (starts May 1st 2018).
