U.S. – Amazon Inc. (NASDAQ: AMZN) launched the Alexa Prize Competition with a $2.5 million reward. Winning the contest will require improving their voice assistant software Alexa, so the e-commerce giant is challenging college students to create a chatbot with the company’s AI platform.
Student teams will be building a social bot that can get Alexa talking to people about popular topics, news, and recent events. The winners will get a $500,000 prize, and if the software can handle a coherent conversation with a person for 20 minutes, their university will receive 1$ million.
Current top-end voice AIs can talk an average of 5 minutes with a person, and as the company stated on the AWS blog (Amazon Web Services), their dream is to make their Alexa as smart as LCARS, the operating system used in Star Trek.
What are the rules of the Social Bot Challenge?
Alexa Machine Learning Vice-president Rohit Prasad explained the objective was to create a bot that can talk about everyday issues. The software should gather information and opinions from the web and express them in context as a human.
The enterprise will sponsor ten teams and give them $100,000 each. Anyone interested can gather their friends and submit an application on the social bot challenge web page before Oct. 28, 2016.
Groups will have until November 2017 to build the conversational app based on Alexa. The contest concludes with the cloud-computing event “Amazon Web Services (AWS) re: Invent,” where the company will announce the winners.
Amazon will deliver full databases to the participants and will offer support via their Alexa team. The work groups can start building their software using the Alexa Skills Kit (ASK), a collection of self-service APIs, tools, documentation and code samples for developers.
The students will also receive immediate feedback on their ideas from millions of engaged Alexa customers. The customers will be able to test the results by chatting with Alexa about different topics like gossip, entertainment, business, sports, and technology, to name a few.
Amazon’s popular voice assistant is getting better thanks to machine learning technologies, but it still has a long way to go before the AI can talk like humans do.
Can machines think?
Alan Turing, the forefather of computing and artificial intelligence, said in 1950 machines might one day think as humans do. However, language remains difficult for machines because of its ambiguity, and complexity.
On June 2014, a computer chatbot named Eugene Goostman was the first AI to pass the Turing Test in a competition held at England’s University of Reading.
33 percent of the judges thought they were talking with a 13-year-old Ukranian boy. The machine could even avoid questions by joking around and change subjects. The judges forgave its mistakes because the AI was a teenager who spoke English as his second language.
Shortly after, the Canadian researcher, Hector Levesque, updated the Turing Test to unveil how much of human speech an AI truly deciphers. The Winograd Schema Challenge asks machines 60 questions that would require common sense to solve.
For example: “The women stopped taking pills because they were pregnant. Which individuals were pregnant?” Machines would have to reason that women can get pregnant, but pills cannot.
AIs are 40% accurate in the Winograd Schema Challenge. The test proves common-sense is not something engineers can code (for now), and computers must rely on analyzing statistics and data to understand the real world.