Wednesday, February 28, 2018

Emory team vies for best social bot via Amazon's Alexa Prize

Faculty advisor Eugene Agichtein (far right) with the Mathematics and Computer Science Alexa Prize team (clockwise from top left): Ali Ahmadvand, Mingyang Sun, Jason Choi, Sergey Volokhin, Zihao Wang and Harshita Sahijwani. (Photo by Ann Borden, Emory Photo/Video)

By Carol Clark

“Alexa, when will you learn to chat with me like people I might meet at a party or a pub?”

“I couldn’t say.”

Alexa may be a popular talking bot, but she has not yet acquired the “social” skills to turn that query into a conversation.

A team of Emory students from the Department of Mathematics and Computer Science are trying to help her develop those skills sooner, rather than later. They are among eight university teams selected from around the world to create a social bot and compete for this year’s Alexa Prize. Amazon is sponsoring the $3.5 million university challenge in order to advance the conversational capabilities of bots such as Alexa — Amazon’s “personal assistant” software that responds to voice commands through a growing list of devices.

“Conversational AI is one of the most difficult problems in the field of artificial intelligence,” says Zihao Wang, a graduate student and the leader of the Emory team. “Human language is so rich. We use combinations of words to form different expressions and idioms. It’s difficult to represent them in computer language.”

Wang’s teammates include Ali Ahmadvand, Jason Choi, Harshita Sahijwani and Sergey Volokhin — all graduate students — and senior Mingyang Sun. The team’s faculty advisor is Eugene Agichtein, an associate professor of Mathematics and Computer Science.

Each of the university teams received a $250,000 research grant, Alexa-enabled devices, and other tools, data and support from Amazon. A $500,000 prize will be given next November to the team that creates the best social bot, while second- and third-place teams will receive $100,000 and $50,000.

Additionally, a $1 million research grant will be awarded to the winning team’s university if their social bot achieves the grand challenge — conversing coherently and engagingly with humans for 20 minutes with a user rating of 4.0 or higher.

“The contest is a wonderful way for students to get hands-on experience developing a social bot using state-of-the-art technology,” Agichtein says. “Their work will be tested out by millions of real-world consumers through Amazon. And Amazon provides support and training so they can get experience with data and computing environments that are usually only accessible to those within major corporations.”

Agichtein’s IR Lab is developing new techniques for intelligent information access, including Web search and automated question answering. Conversational search capabilities are a key emerging trend, he says.

He notes that his children love asking Alexa trivia questions or about music and sports. “It’s natural for them to talk to devices instead of having to type in a question because they’re growing up amid this technology,” Agichtein says. “And as time goes on, it’s clear that voice-based communication devices are going to keep improving and become more ubiquitous.”

Wang is a native of China who earned his master’s in civil engineering at Carnegie Mellon University. A robotics project sparked his interest in information retrieval powered by machine learning, leading him to Emory and Agichtein’s lab to work on his PhD.

“Machine learning is widely applied in the real world,” Wang says. “It’s changing peoples’ lives in every way.”

Autonomous vehicles, drones, online shopping mechanisms and robots designed to detect and remove dangerous objects are just a few examples of how machine learning is being applied.

 “The idea is to train an algorithm to ‘learn’ patterns embedded in data,” Wang explains.

While a machine learning algorithm to simulate natural, human conversation is a difficult challenge, Wang says it’s one well worth pursuing.

Possible healthcare uses for conversational social bots include providing companionship to isolated seniors, serving as therapeutic agents for people suffering from depression and conducting patient interviews to streamline admissions to a medical clinic.

Wang also led an Emory team in the inaugural Alexa contest last year, but the team did not make it to the finals. “We learned a lot from the experience,” he says.

The working title for the Emory social bot this year is IRIS, which stands for information retrieval and informative suggestion agent. “Our focus will be on the accuracy and usefulness of information that we provide to users,” Wang says. “And we will add conversational functionality to our design to make the responses as natural and engaging as possible.”

IRIS will incorporate “ideas from each member of the team,” he adds. “That’s one of the most fun things about the contest, is working as a team.”

Starting in May, the public can access competing bots to provide feedback and rate them by saying, “Alexa, lets chat,” to an Echo device, or to the Amazon mobile app. The bots will be randomly assigned and remain anonymous, so that people providing feedback cannot identify the university that generated them.

By August, Amazon will have used this feedback to winnow the contestants down to three finalists that will continue to get more consumer feedback until the winner is announced in November.

Other university teams competing this year include: Heriot-Watt University in Edinburgh, Scotland, Czech Technical University in Prague, Brigham Young University, UC Davis, KTH Royal Institute of Technology in Stockholm, Sweden, UC Santa Cruz, and Carnegie Mellon.

Related:
Raising IQ of web searches
Mouse trail leads to online shoppers

No comments:

Post a Comment