Virtual Assistants Need Machine Learning, But They Need People Too

Despite the power of machine learning, it still needs a human touch


Anyone who’s ever tried to use Siri to find a local taxidermists and ended up booking a month-long cruise around the Balkans could be forgiven for thinking that virtual assistants are just a graveyard of empty promises and broken dreams. Tech companies don’t see them that way though, and the biggest are investing heavily as they compete to become the first to be truly ubiquitous in people’s daily lives.

On average, people who use Siri every day or week only do so to ask it three or four questions. The questions they ask are fairly basic, and usually amount to little more than a voice-activated web search. In 2015, the virtual assistant market saw tremendous growth in the number of innovations - with new features, functions, and integrations. Much of this is being enabled by major advances in machine learning and natural language processing, subsets of artificial intelligence (AI), that help computers to understand speech and people’s habits, as well as developing their own personalities that should, hopefully, help overcome the natural human discomfort with interacting with a machine on such a familiar level.

To give a basic example of how machine learning works, it can take you through the entire restaurant booking process. It can learn the kind of restaurants you like by analyzing your past behavior, and reviews on various websites like Trip Advisor, as well as other people’s recommendations. It can then book the restaurant you choose from the options it gives you, and if you book an Uber through it, it would eventually pick up on the pattern by itself so that it offers to book an Uber for you from the outset. This is just the most basic iteration, there is far more space for the technology to move into, learning all of your wants and needs before you have to ask.

Facebook has been testing its own chat-based digital assistant called M inside its Messenger app, which is currently being pushed as the future, not just for the Facebook website, but for the entire way we interact with our smartphones. M is exceptional because it blends AI and human trainers so that it can do everything from buy movie tickets to research restaurants for dinner. Obviously, using humans trainers to help teach the app means that it is only in a testing stage, with just a few thousand beta users in California, and it is unlikely to be rolled out on a large scale purely because of the logistics involved. However, the cyborg design means that M can deal with far more complex requests than its rival mobile app assistants like Microsoft’s Cortana and Apple’s Siri.

Humans behind the scenes are vital for helping virtual assistants to learn how to interact with people as if they were real people themselves. At the moment, one of the main things holding adoption of virtual assistants back is the public’s discomfort with communicating machines using their own voice. Amazon’s Alexa is working to humanize its responses by adding in ‘hmms’ and ‘ums’ into her responses. Apple’s Siri assistant too is renowned for making wry jokes. To do this, companies are hiring writers to help build the assistant’s personality. By using machine learning algorithms, the bot can learn to produce authentic responses on cue from the basis the writer creates. Former Hollywood screenwriter Robyn Ewing is one of those who is currently working to form their personalities. She notes that users could easily do everything that an assistant does simply by going online and getting the information themselves, albeit perhaps slower. It is the personality that sets it apart and makes people want to use the technology. As Ewing notes, ‘if the character doesn’t delight you, then what is the point?’

Would you like more information on Machine Learning? Join the Machine Learning Innovation Summit in San Francisco in May
University lecture small

Read next:

How Are Higher Education Institutions Using Analytics?