Eve is optimistic, supportive and friendly. She doesn’t like bananas — but she could be swayed if they were in a banana cream pie. She likes chips and salsa, and has a list of favorite books.
For now, Eve, a conversational bot that would be incorporated into Amazon’s Alexa, can’t remember your name. But she’s working on that.
“We spent a lot of time making sure she doesn’t say bad words or give legal or financial advice,” said William Myers, one of 10 Brigham Young University computer science students working on a project to make Alexa converational.
It’s harder than it sounds.
The BYU team is one of eight teams chosen worldwide to compete in the second annual Alexa Prize Challenge by Amazon. The team received a $250,000 grant to work on Eve, short for Emotive Adversarial Ensembles, and work on cracking conversational artificial intelligence.
The team will work on the project into the summer for the chance to win $1 million for the university and $500,000 to split among the team.
The ultimate goal, set by Amazon, is to have Eve be so engaging, a human will continue a conversation with her for 20 minutes.
A BYU group had previously submitted an application for the challenge that wasn’t accepted.
The group is approaching the problem by looking at conversations as a collaborative process where each party is trying to say things relevant and useful to the other’s life. If that doesn’t happen, then the conversation drifts into what Nancy Fulda, a Ph.D. student who leads the team, calls “bad date syndrome.”
“Our whole idea is based on the idea of, let’s not treat this as a learning problem, but a cooperative effort,” Fulda said.
Eve will have a base personality, but her personality will change based on who she interacts with. The team doesn’t want her to agree with everything a user says, but they don’t want her to be controversial. Her core personality is made up of a list of likes and dislikes gathered from the 10 team members, but her personality can’t contain any information that could identify the team that programmed her; so don’t expect her to spout off a “go Cougs” anytime soon.
The team is using conversational datasets, like Reddit (minus the foul language and trolls), to train Eve on. They are also running the BYU Chit-Chat Challenge to get sample conversations for Eve.
For the team, their drive comes from the idea of being the classic inventor and solving previously unsolved problems.
“That process and the ability to do things that have never been done before in a conversation AI setting is fascinating in and among itself,” Fulda said.
The small answers they are solving now could have greater implications.
“Cracking the conversation AI is a huge part of cracking AI, so if you can solve conversation AI, in my opinion, you can solve a lot of AI problems,” Myers said.
Eve also has the potential to provide a tangible impact on the world, as seniors could use her to remember where they put their keys or a student could use her to help with writing an essay.
“It is pretty cool that Amazon is giving us this platform to experiment with this technology,” said Tyler Etchart, a member of the team.
David Wingate, a computer science professor and the team’s faculty advisor said the project has been student-run and fueled by the idea of making the world a better place and solving problems.
“What is great is they are driven by more than just winning a prize,” Wingate said.
For Zachary Brown, a sophomore on the team, he knows not all the questions will be answered this summer.
“That excites me,” Brown said. “That gives me a sense of what is going to be happening for years down the road.”