In the 2013 film Her, Spike Jonze offered his take on a society in which our closest relationships are not with humans but with technology. The movie tells the story of a lonely man who purchases a voice-controlled A.I. that evolves to meet his needs. Samantha, as she’s known, eventually becomes so in sync with him that they fall in love.
Oddly enough, the technology in Her is not as far off as you might think. At her core, Samantha is an interface capable of interpreting needs through conversation and contextual cues.
Today, we have chatbots that are similar in form. They are conversational systems that we use to play music, tell us the weather, order a cab, and much more. One of the more intelligent examples is Pana, a travel chatbot that monitors your flights and re-books when there’s a potential delay. And, yes, some of us have already fallen in love with her.
While these bots are ambitious, they’re not quite Her. They continue to rely on scripted, command-response interfaces, which can quickly get things wrong. Google “Facebook chatbots” and you’ll find headlines such as “Please, Facebook, don’t make me speak to your awful chatbots” or “Facebook Chatbots Are Frustrating and Useless.” Clearly, we’re not there yet.
Yet, help is on the way. With a number of advances, we should be able to make these interfaces work much better — and that has huge implications for companies that will use them to interact with customers. To succeed, however, they’ll need to overcome a number of challenges.
1. Natural language processing
Right now, chatbots require us to master their syntax. If you ask them “What’s the weather?” you’ll usually get an answer. If you ask “Could you check the weather?” you might not. To get around this problem, we need to enable bots to understand all the ways we say something — and how context changes what we mean. This is known as natural language processing and it’s an object of considerable focus for companies like Google with SyntaxNet and Facebook, with its DeepText A.I. We also have a range of open-source tools vying to make our machines understand us better.
2. Machine leaning
Getting natural language is one thing. Knowing what to do with it is another. To get truly natural responses, we’ll need to improve machine learning. With exposure to enough conversations, the computer on the other end of the line should gradually learn what the correct response should be. Of course, the challenges involved are far from simple, as illustrated by Microsoft’s short-lived experiment with its unintentionally racist chatbot Tay, but we should get there.
3. Human intelligence
Oddly enough, if you want to make bots seem more human, we have a good resource: ourselves. The technologies themselves are only as capable as the programmers who build them. Not surprisingly, in February, John Giannandrea, the head of Google’s artificial intelligence efforts, also took a position as Head of Google Search. This reflects a larger trend in which digital players are shifting their focus from traditional engineers to machine-learning experts. Rather than teaching humans to understand machines, we’re now teaching machines to understand us.
4. Providing value
While Samantha from the Her movie is all about technology, she ultimately provides a service. Companies also have to figure out what they can offer through A.I. Luckily, the possibilities here are great. If Pana can book a flight through a conversation, why can’t American Express Travel or KLM? If Uber recognizes the right time to offer a taxi, why can’t banks recognize and facilitate money transfers between friends?
5. Staying simple
It might seem like a “natural” chatbot needs to be infinitely vast in order to succeed. That’s not true. Most successful innovations, like iOS, are simple and easy to use. For companies to succeed, they’ll need to look for the least common denominator in their services. What are the small number of things you need to do in the most natural way?
6. Preparing an exit
Of course, this means they’ll also need to hand the experience off at some point. Currently, chat-like services such as Siri and Alexa have mixed track records when it comes to solving this problem. Most of the time, when they don’t understand you, they simply take your text and turn it into a web search. With intelligent A.I., we’ll need an elegant way to send people to a system that can understand them, whenever they’ve exhausted our possibilities.
Right now, intelligent chat is still a little ways off. But with work moving forward on all fronts, companies need to start thinking about what they can do with these new technologies. It’s time to keep an ear to the ground and start working on the future. Someday, Her will walk off the movie screen, and we want to be there to say hello when she does.
This article was written by Possible and Alex Whittaker from VentureBeat and was legally licensed through the NewsCred publisher network.