The Alexa Prize by Amazon – Towards Conversational AI

Amazon chose 12 university teams from around the world for the inaugural edition of The Alexa Prize. The goal of the contest is to create a socialbot capable of maintaining a coherent conversation with a human for 20 minutes. The winning team will receive a $500,000 prize.

The Alexa Prize

Each sponsored team will get a $100,000 funding for the duration of the project. Apart from the main prize, there’s a $1,000,000 bonus for the winning team’s university if their bot will be able to reach the contest’s goal – chat with a human in a meaningful way for 20 minutes. As the project aims at enhancing Alexa’s capabilities, the teams will also receive Alexa-enabled devices, access to the Amazon Web Services, and support from the Alexa Skills Kit developers.

Who, or What, is Alexa?

Alexa is the voice of Amazon’s Echo speaker. It’s one of the most popular AI personal assistants out there, being constantly improved through third-parties apps that make it smarter. Thanks to the so-called “Alexa Skills”, she can order a ride on Uber, a pizza from Domino, read news headlines, remind you to turn off the oven, manage your to-do lists or inform you about the weather forecast. The list is quite long and it’s getting longer with each week. Recently, it features more and more apps that integrate Echo with other smart domestic devices. However, Alexa is still far from being a chat companion.

Conversational AI

The Alexa Prize aims to change it and accelerate the progress of conversational AI. In order to achieve it, the teams will work on several tricky technologies at once, such as machine learning, natural language understanding & generation, context modeling, and dialogue planning. The socialbots will need to be able to converse with humans about news, popular topics or scientific breakthroughs, switching smoothly to related topics if need be, up to a point when the human chooses to stop. The key is to keep the exchange interesting and engaging at all times. Contrary to the famous Turing test, the socialbots don’t need to trick humans into thinking they’re talking to another person. This would only create unnecessary limitations to the bots’ potential.

Mass-Scale Customer Feedback

Interestingly, the teams will be able to share their work with Amazon’s customers. In turn, they’ll be providing immediate feedback that will help the participants to improve their bots even more. This two-part process will allow the teams to polish their creations before the final. We will know the winning team in November 2017, almost a year from now. We’ll keep you updated as the competition progresses.