Can AI become a man’s best friend? A team of Stevens graduate and Ph.D. students competing to redefine chatbot technology says yes. The annual Alexa Prize SocialBot Grand Challenge (SGC5) places university computer science teams in an arena of multifunctional user-based chatbots that rival Amazon’s Alexa. The stakes? Tech publicity from the Amazon supergiant and a $1 million research grant. Under the rigor of competition and the push for innovation, the Stevens team rose to the top five finalists in the global competition held in early August.
With the phrase, “Alexa, let’s discuss,” Amazon users can access the Stevens bot, coined NAM—an abbreviation of Never Alone with Me—through the disembodied voice of Amazon Alexa products. Faculty Advisor, Professor Jia Xu, alongside first-year Stevens masters students João Luís Lins Rodrigues Cruz, Sai Nikhil Reddy Maligireddy, Abhijeet Gusain, and Yeshwanth Reddy Peddamallu developed NAM from a friendly inspiration to the final rounds as the Stevens bot measured up against Amazon guidelines.
In 2016 Amazon launched the Alexa Prize Competition to improve natural language learning in AI, typically developed by exposing the system to source information and then assigning the newly acquired pool of “empty” information common meaning by applying parameters, or a set of specific conditions used to establish recognition. This process, known as neural response generation, offers the exigence of ChatGPT as well as a semblance of machine-learned humanity in which Stevens’ NAM is Alexa’s virtual assistant protégé.
The SGC5 applies criteria for overall social delivery and scientific innovation, in which NAM gains its edge through an emphasis on empathy and connection to enhance the user-chatbot experience. NAM models active listening through large language models, a form of machine learning that statistically predicts a series of words to generate natural language, in order to target open dialogue as opposed to strictly task-oriented responses.
In previous rounds, the team bots were eliminated based on customer reviews, live feedback, and the scientific merit of technical research papers published by each team. NAM advanced to the final judge evaluations as the bot was tasked with holding a meaningful, engaging conversation for at least two thirds of a 20 minute period while earning a composite rating 4.0 on a 5.0 scale to win the final prize.
How did NAM perform? The Stevens bot exceeded competition standards as users ranked the bot in the top 3 for standalone conversation time, double that of other teams. Xu said in a Stevens Research and Innovation article, “On average, our bot’s conversation time doubles other bots’ time. For some users, our bot reaches 56 minutes of chat duration, compared to about 10 to 20 minutes typically.”
In addition to prolonged conversational stamina, NAM’s original inspiration to provide friendship and support is embedded within its language model. NAM takes open dialogue one step further by asking follow-up questions to clarify and or strengthen its conversation with the user. Once the user activates NAM, the bot becomes an active player in the course of the conversation as NAM listens, clarifies, and validates with training to provide explanations after offering suggestions.
NAM’s demonstration of user adaptability in chatbots of the future is just the beginning as Xu says, “Most of our team members at this point are fresh students without prior natural language processing backgrounds, and they became chatbot experts through the competition.” NAM’s success speaks to the drive of Xu’s team as the Stevens bot makes strides in open dialogue systems, language models, and the user relationship with technology to steer the next chapter of AI. “We realized our unique social bot based on imagination, innovation, and technology, and we are happy to be in the final.”
