Nine teams advance to semifinals for the Alexa Prize SocialBot Grand Challenge

In November, Amazon announced that nine teams from around the globe were selected to participate in the Alexa Prize SocialBot Grand Challenge 5 (SGC 5), a university challenge focused on advancing human-computer interaction and open dialogue conversation. As of today, all nine teams have advanced to the semifinals based on their performance during the initial customer feedback period.

The teams selected for the challenge, which began in November, include five returning competitors — including the top two finishers in the most recent challenge — and four new universities.

“Since its inception in 2016, the SocialBot Grand Challenge has driven technical advances in neural response generation and the application of large language models to open-domain dialogue”, said Michael Johnston, an applied science manager in Alexa AI who leads the science and engineering teams supporting Alexa Prize. “This year SGC5 teams are applying and integrating a broad range of different large language models into their SocialBots, and it is super exciting to see the kinds of engaging interactive experiences they can enable for Alexa customers”

The competing teams also face a challenge newly introduced for SGC5: their socialbots must provide a compelling multimodal user experience, integrating speech with visuals. Teams are pursuing a broad range of approaches including emotive avatars, synchronized graphics and multimedia, image generation, and multimodal dialogue using hints and touch input.

“It will be incredibly interesting to see which approaches prove out to be most effective in the finals,” Johnston added.

“Creating a socially adept AI is a hard problem,” said Reza Ghanadan, senior principal scientist with Alexa AI and head of Alexa Prize. “This is because human-like social conversation is remarkably delicate and complex, and the open domain nature of the SocialBot dialogues makes it extremely challenging.

“You need to provide relevant and deep responses to a wide range of topics, awareness to differentiate between reality and imagination, maintain a natural and coherent exchange throughout a potentially long conversation, and accurately interpret the intent by correctly picking up on names, topics, places and products while considering the context of each conversation turn. You also need to make the interactions lively, robust and engaging, which is challenging given the diversity of topics and users.”

The Alexa Prize is a unique industry-academia partnership program which provides an agile real-world experimentation framework and tools for accelerating scientific discovery. University students have the opportunity to launch innovations online and rapidly adapt based on feedback from Alexa customers.

“Prize competitions provide data, AI tools, and an agile experimentation framework for researchers and students to innovate on advanced topics in creating socially intelligent digital assistants, encouraging them to explore transformational ideas at the boundaries of what is achievable in the real world,” Ghanadan said.

Alexa customers can interact with the university socialbots by saying “Alexa, let’s chat” on Amazon Echo or Fire TV devices. Customer ratings and feedback help the student teams improve their bots leading up to the competition finals.

The ultimate goal is to meet the Grand Challenge: earn a composite score of 4.0 or higher (out of 5) from a panel of judges, and have those judges find that at least two-thirds of their conversations with the socialbot in the final round of judging remain coherent and engaging for 20 minutes. The first team to meet the Grand Challenge will win a $1 million research grant for their university.

Updates to this year’s competition

As noted above, this is the first iteration of the SocialBot Grand Challenge to incorporate multimodal customer experiences. In addition to verbal conversations, customers with Echo screen devices or a Fire TV may be presented with images or text that enhance the conversational experience. Teams have the opportunity to improve their customer interactions by including additional text and images that provide more diverse and meaningful information.

This year there are also two sets of awards: one set for overall social interaction performance and one set for scientific innovation. Prizes for overall performance in the competition will be $250,000 for the first-place team, $50,000 for second, and $25,000 for third.

The new award for scientific invention and innovation allows teams to focus on advancing the field of conversational AI through a deeper study of the fundamentals of open dialogue conversations. Cash prizes for scientific contribution will be awarded to students on the winning teams — $250,000 for the first-place team, $50,000 for second, and $25,000 for third.

A unique challenge

The SocialBot Grand Challenge represents a unique opportunity for student researchers to experience and learn how their ideas work within a real-world environment.

“We have learned that success requires researchers to create generalizable AI techniques and incorporate knowledge in appropriate and engaging ways,” Ghanadan said. “It also involves addressing open research problems in natural language understanding and multimodal language processing, contextual understanding, natural response generation, empathy, and commonsense reasoning, to understanding social norms, and dialogue management.”

Each university selected for the challenge receives a research grant of up to $250,000, Alexa-enabled devices, free Amazon Web Services (AWS) cloud computing services to support their research and development efforts, access to Amazon scientists, the Cobot (conversational bot) toolkit, and other tools such as automated speech recognition through Alexa, neural detection and generation models, conversational data sets, and design guidance and development support from the Alexa Prize team.

In previous challenges, participating teams have improved the state of the art for open domain dialogue systems by developing improved natural language understanding (NLU) systems, neural response generation models, common sense knowledge modeling, and dialogue policies leading to smoother, and more engaging conversations.

The “Alquist” team from Czech Technical University won the fourth challenge, with teams from Stanford and the University of Buffalo earning second- and third-place prizes, respectively. The publications from that challenge can be found here.

Winning teams from previous years include Emory University, the University of Washington, and the University of California, Davis.