Tom Simonite recently published a piece in Technology Review about the capabilities of Microsoft’s Cortana virtual assistant. Simonite spoke with Larry Heck, an engineer at Microsoft and a contributor to the Cortana project. According to the article, Cortana has some key capabilities that distinguish it from Apple’s Siri.
Heck contends that Cortana can track the meaning and context of conversations longer than Siri can. In the article, Simonite uses the example of someone asking Cortana to find a cheap Japanese restaurant close by. Once Cortana returns results, the virtual assistant is able to correctly respond to follow up queries such as “Which ones are still open?” and “How long will it take me to get there?” The ability to maintain the context of the conversation during these simple follow up questions would be a noticeable improvement over current virtual assistant technologies.
By using the capabilities of Microsoft’s Bing search engine, Heck says that engineers are also developing ways for Cortana to access knowledge from the web in real time to respond to people’s questions.
Simonite also spoke with Norman Winarsky, VP of SRI Venture at SRI International, for his insights. Winarsky makes two points that I find especially interesting. The first point is that user expectations are much higher for virtual assistants that are given a human personality. Users just naturally expect a virtual assistant in the persona for a “Siri” or a “Cortana” to be able to understand them and communicate with them according to the rules of natural human conversation. Simple mistakes in the dialog, which happen frequently with today’s limited conversational technology, aren’t easily forgiven and quickly cause user frustration. Simonite speculates that these unmet expectations may contribute to the fact that few people are actually using Siri. Anecdotal evidence seems to indicate that users are less frustrated by the personality-free Google Now.
The second point is that virtual assistants are much more effective when they’re implemented for a narrow, specialized area. They run into trouble when they try to go broad. Winarsky points out that Siri was originally designed to handle queries for travel and entertainment. It’s much easier for a virtual assistant to understand questions and respond appropriately if the context of the queries is limited to a specific domain. Once Siri was expected to handle a much broader range of inquiries, the performance of the technology suffered.
Although there’s no discussion of web self-service virtual assistants (along the lines of Nuance’s Nina) in Simonite’s article, the same two observations probably still apply. Intelligent virtual agents that act as extensions of a brand’s customer service can perform reasonably well, because the range of questions they’re likely to receive is limited. On the other hand, the more such virtual agents are designed to mimic human call agents, the quicker customers are likely to become frustrated by the technological limitations.
It will be interesting to keep an eye on Cortana and see if this relatively new personal virtual assistant can live up to the early hype. At the same time, we’ll be on the lookout for how vendors of personal assistants and web self-service virtual agents handle the dual challenges of personality and specialization (or the move away from it).