Virginia Tech HokieBot competes in Alexa Prize SocialBot Grand Challenge 5 to develop advanced AI technology that enhances user conversation
Photo of four graduate students on the Alexa Prize SocialBot Grand Challenge team with faculty advisor Lifu Huang
A team of four graduate students at the Sanghani Center for Artificial Intelligence and Data Analytics is one of nine international university teams selected to compete in the Alexa Prize SocialBot Grand Challenge 5 sponsored by Amazon.com Services LLC. Each participating team will receive up to a $250,000 research grant to build a skill that can help Alexa converse with users on popular topics and current events for at least 20 minutes while achieving a user rating of at least 4.0/5.0. Top finishing teams will also be eligible for various prizes.
All applications for the challenge were reviewed by a panel of Amazon employees and selected on the criteria of potential scientific contribution to the field, technical merit of approach, novelty of the idea, and an assessment of the team’s ability to execute the plan.
Lifu Huang, assistant professor in the Department of Computer Science and faculty at the Sanghani Center, serves as advisor to the Virginia Tech team that will build a HokieBot that integrates functions ranging from everyday chit-chat to answering both open-domain and domain-specific questions to provide a seamless user experience, similar to chatting with real humans.
“To meet our goal of creating an engaging user experience,” Huang said, “we also have to ensure that our HokieBot understands the user’s intents, reactions, and feelings in between the dialogue.”
By designing a new method of modeling a dialogue system that enables long-term memory, user adaptation, and more controllable generation, the team will address the challenges associated with enforcing coherence and increasing human-machine conversation.
Huang offered the following example to show how conversational artificial intellingence (AI) might be combined successfully with coherent and natural responses that include historical context and retrieved relevant knowledge:
HokieBot: Hi, how are you doing today?
User: Well, not bad. It’s getting cold now.
HokieBot: Yeah, it’s October. Winter is coming.
User: That’s pretty different from my hometown.
HokieBot: Where are you from?
User: Arizona.
HokieBot: Cool! I heard the tree-like cactus there can grow over 40 feet tall.
User: Exactly. Deserts surround my city and cacti live better than the tree.
HokieBot: That must be a fantastic and different experience living there. Do you hike? I know there are several mountains over 10,000 feet in Arizona.
User: Yes, a lot! I love hiking with my friends. Oh, and my dog.
HokieBot: Please also take me next time!
User: No problem.
HokieBot: I’m looking forward to that. However, today is not good for outdoor activities. How about we watch some movies?
User: Good idea. Do you know "I Am Sam?"
HokieBot: Yes. It was released on Dec. 28, 2001, a very touching family drama, and I remember it’s very long.
User: How is the box office of the movie?
HokieBot: I am not sure, but I know it's been nominated for a number of awards, including the Young Artist Award in 2002 and a number of nominations, including the 2002 Oscars Awards.
User: Awesome. Who is the director?
HokieBot: The director is Jessie Nelson. You know, she has a unique eye for casting actors.
User: Yeah, Sean Penn did a great job.
HokieBot: Well, he has won many awards, including the Best Actor Awards at the 55th and 60th Venice International Film Festival and the Golden Globe Award for Best Actor. He is amazing.
User: So cool. I cannot wait to watch it.
HokieBot: Let’s do it!
The four computer science Ph.D. students working on HokieBot are: Ying Shen, who is serving as team leader, and Minqian Liu, Zhiyang Xu, and Barry Menglong Yao. Each has extensive research experience in natural language generation, information extraction, text understanding, and deep learning.
“We believe our HokieBot project could potentially surpass existing systems and provide more complete and reliable insights into conversational AI and its potential and limitations,” said Huang, who is also leading a Virginia Tech team from the Sanghani Center competing in the Alexa Prize SimBot Challenge, which is focused on advancing the development of next-generation virtual assistants that continuously learn and gain the ability to perform common sense reasoning to help humans complete real-world tasks.
The Alexa Prize SocialBot Grand Challenge 5 began on Nov. 14 and final results will be announced in August.
