News / Utah / 

BYU team could win $500K in Amazon's Alexa chatbot challenge

(Brigham Young University)



This archived news story is available only for your personal, non-commercial use. Information in the story may be outdated or superseded by additional information. Reading or replaying the story in its archived form does not constitute a republication of the story.

PROVO — The smart speaker was the undisputed champion of 2017 tech, and Amazon’s Alexa was the industry’s main contender.

Now, Amazon wants to make Alexa even smarter. The retail giant issued a challenge late last year to universities around the globe, calling on programmers to create a bot that will help Alexa chat with users.

Out of nearly 300 applicants, Brigham Young University became one of eight schools to win a spot to compete. The team of 10 students received $250,000 from Amazon to invest in their research lab and will win half a million dollars to split amongst themselves if their bot wins.

“I like to think that (the reason we were chosen) is because of the central concept of our idea,” said Nancy Fulda, the project's leader and a computer science grad student at BYU.

At the moment, Alexa is more like a voice-activated taskbot — a system with preprogrammed responses. Amazon wants its smart speaker, instead, to be able to answer questions like, “Alexa, why is it important for me to get an education?” or “Alexa, how do I cheer up my sister?”

“Most machine-learning researchers tend to think of conversation as a mapping from text to text. You take text in, you manipulate it, and then you try and figure out what’s the right text or the right set of words to send back out,” Fulda said. “Our view comes at it with the idea that a conversation isn’t about mapping text to text, it’s about communicating between two people.”

Fulda’s team is trying to create a bot — which they’ve since dubbed “Eve” — that will model the human with whom it’s talking and fill that user’s need. To do so, the BYU students are using a combination of probabilistic programming and neural networks.

Eve will use a series of estimates about the person with whom it’s speaking, then calculate how it should respond. The bot may detect there’s a 25 percent likelihood that the person is happy, a 30 percent likelihood that they’re wanting more information and a 20 percent likelihood that they’re trying to give information.

“Based on what they say, we’re constantly adjusting those probabilities, and we use those probabilities to guide the neural networks that produce those responses,” Fulda said.

The team began working on the project in February and has until July before a semifinal round where five schools will be eliminated and the remaining teams will compete for the top spot. The final winner will be announced in November.

Because of the nature of the project, Fulda said it is difficult to know whether the bot is on track or not, though she believes it’s “progressing very nicely.”

“It’s like raising children,” she said. “You never see the results until you’re already done. So you never know if you're doing it right until you see the finished product.”

And much like a child, the bot is still learning and growing, Fulda said.

“You have to view Eve as being in her infancy. … She’s just barely learning how to say things, so half the stuff she says doesn’t make sense yet, and the other half is wildly inappropriate to the current situation — and that’s totally fine.”

SIGN UP FOR THE KSL.COM NEWSLETTER

Catch up on the top news and features from KSL.com, sent weekly.
By subscribing, you acknowledge and agree to KSL.com's Terms of Use and Privacy Policy.

KSL Weather Forecast