Ilya Gelfenbeyn is the CEO of Speaktoit, the top-rated virtual assistant on Android, and an authority on natural language processing. We caught up with him to ask why we still have to talk down to digital personal assistants, and when exactly we might acquire a more natural relationship—and tone—with them. He answered in the most natural tone he could muster.
FAST COMPANY: This is the year the digital personal assistant has really taken off. Speaktoit has been downloaded several million times since launching last year. And Siri is a household name. But my sense is that these assistants still have a long way to go.
ILYA GELFENBEYN: The field is still in a really early phase of development. It’s something like the search engines in the beginning of the '90s. There’s a long way to improve—there’s not yet something like the Google of this industry. We’re working on some functions to make an assistant much more helpful, by working with lots of different services. Natural language is a way to combine them in one place.
Can you explain natural language processing, in brief?
It means you can just tell the assistant whatever you want, and don’t have to remember particular commands.
But I still feel like we have to talk down to Siri. When will she be able to understand slang, so I can just talk to her like I talk to my homies?
There’s a lot of work to do. We’re working on a natural language processing engine that will take into account conversation context like location, dialogue history, and user preferences. Different users can mean different things with the same words, and the assistant should understand the context.
What’s an example?
A user can say, "I’m bored," and a personal assistant should understand what to provide to the user. For some people it can say, "Let’s go somewhere, I can find you a nice nightclub." For others, it will provide a joke. For others, it’ll say, let’s call some friends.
And how long till we can talk to Siri or Speaktoit as casually as we do to our friends?
You should compare it not to talking to friends, but to a real life personal assistant. With friends you can discuss general stuff, but a virtual assistant is not created for that. I would say in three years it will be much better—it will support all basic tasks, and definitely will not have problems with setting up appointments, working with your phone, working with basic features. Of course it still will not be able to replace a real personal assistant in some situations.
When will we get that Google of assistants? In the next decade?
Yes, I think so. In the next decade, we should have a good combination of technologies and the features they implement.
Is some sort of breakthrough needed, the way Sergey Brin and Larry Page figured out the significance of linking?
I wouldn’t say there’s a magic formula. It just really requires hard work. It may of course include some new algorithms, but there’s no one magic formula.
Who will support this research—universities, enterprise, military?
Enterprise is much more focused on this. Universities will be a good resource for fundamental knowledge.
The military had a hand in developing some of the tech behind Siri. Are they out of the picture now?
They can have some role. But the personal assistant will not be focused on military purposes.
So we won’t have generals saying, "Siri, please bomb this village"?
The cost of the mistake here is too high. Not in the next 50 years.