Conversations can be tough. Whether you’re chit-chatting with a coworker or having an important talk with your partner, it’s easy to misinterpret, say the wrong thing, or accidentally offend someone. Now imagine teaching a computer how to avoid those minefields. That’s even tougher—and Googler Cathy Pearl knows exactly how difficult it is.
Cathy has made a career out of teaching computers how to talk to humans. She’s worked in the field of conversation design for decades, and now works in outreach at Google, where she helps spread the word about her field both within and outside of the company. She also served as a judge for this year’s Webby Awards, which is introducing a category for voice user interfaces for the very first time. (Google ended up winning several awards, too, in categories Cathy didn't judge.)
For this installment of The She Word, Cathy tells us about the challenges of teaching computers to talk to humans, and what that’s taught her about her own conversations:
Designing conversations is trickier than you think. That’s because human conversations are really complicated.
“Basically, conversation design is about teaching computers how to communicate like humans, not the other way around. We all know how to talk from a young age, so now we need to build computers that can understand us where we are, instead of forcing people to speak some foreign computer language.
People may not realize how complex it really is. Think about something that seems like a simple yes or no question: What if you asked me, ‘Do you want a cup of coffee?’ Let’s say I replied, ‘Coffee will keep me awake.’ Is that a yes, or a no? Well, if you asked me first thing in the morning and I have a big presentation to write, it’s probably a yes. Ask me right before bed, and it’s probably a no. People say things like this all the time, but it’s hard for computers to understand.”
Voice recognition used to seem like the stuff of fiction. It's come a long way.
“I learned how to program when I was a kid, and I was really interested in learning to get the computer to talk back to me. I was really into movies like ‘War Games’ and TV shows like ‘Knight Rider’ that had these talking computers. Now, there was no such career at the time really, unless you were a researcher at Bell Labs or something like that. Coming out of grad school, I didn’t know of any jobs I could take in that field.
So really it was in 1999 when I saw a job opening for a company and they said, ‘Come work on speech recognition!’ And I said, ‘Well, that stuff doesn’t work, it’s still a science fiction thing.’ But they had a demo line you could call, and it was this fake banking demo where you could move money from checking to savings. It’s all you could do, really, but it worked. I was astounded. I spent eight years at the company learning the ins and outs of building voice user interfaces for phone systems for companies.”
When you find yourself at a career crossroads, don't limit your options.
“If you do something like IVF, it takes over your whole life. It’s a constant thing. That’s why I quit my job. You can’t plan vacations, you can’t plan work meetings, because you have to go to the doctor’s office. And it’s so disruptive. After nearly 3 years of trying, I had my son. I spent the next three years as a stay-at-home mom.
I think what was hardest for me was the point where I thought, I absolutely want to go back to work now, which was earlier than those three years, but I didn’t know what I was going to do. I didn’t know what resources to use to try and figure out what I should do to get back into a great career. I felt very alone in that way.
I went to a career counselor, and I just tried to start saying yes to more things. So when somebody asked me to give a talk, even if I didn’t think I was necessarily qualified, I said yes. I said yes to writing a book, which was just a terrifying prospect. It expanded my worldview of what was out there, and it opened a lot of doors to opportunities I wouldn’t have had otherwise. I think as women we often undersell ourselves.”
Teaching computers how to talk to us can teach us a lot about ourselves.
“So much of the time when we communicate, we want to be acknowledged. We don’t want you to try to solve problems. When I’m saying I had this really hard day, I don’t want my friend to say, ‘You know what you should do next time?’ No! I want you to say, ‘That sounds frustrating.’
That applies to voice user interfaces. With the Google Assistant, there’s a lot of stuff we can’t do yet. But it’s better to acknowledge the things we can’t do then just say, ‘I don’t understand.’ If someone says, ‘I want to rent a car,’ and we can’t do that, can we say, ‘I’m sorry, I can’t rent cars yet?’ That’s more satisfying at a basic, human, primitive level, because at least they understood me.”