The ten participants competing for the best conversational Alexa Prize bot were selected by Amazon last June out of 375 student teams from more than 15 countries. The only non-American teams are the CTU Alquist team lead by Jan Šedivý from CIIRC and the team from the Moscow Institute of Physics and Technology. Since the Alquist has been shortlisted for the third year in a row, it can be considered an “experienced guy” and a clear favorite.
In addition to the fact that the American organizers of the Amazon Alexa Prize contest last year completely “reorganized” the time schedule of this international battle compared to the last two years to correspond more closely to the academic year, the traditional meeting of all the selected teams where the semifinals are being announced – Alexa Prize Summit – was recently canceled. It was scheduled to take place in Seattle next week, but it was dismissed at the last minute due to concerns about the spread of the COVID-19 virus. Moreover, it is not possible to move it to another date because this year´s competition already ends at the turn of May and June.
What is the tactic of the team this year, are they working on something new, how they stand this year?
Somehow different Alquist
Since September last year, the team has been working on a new version of Alquist, which uses knowledge-based graphical databases to perform the dialogue. With these databases Alquist is able to remember user preferences and customize their communication style. Moreover, the last year´s version of Alquist had some reserves in terms of responses to unexpected user queries. Therefore the philosophy of microdialogues was created. In the new version of Alquist, the dialogue is being assembled from the atomic parts of the communication directly during its course so the Alquist is able to react better. The team is convinced that this new approach consisting of the knowledge-based databases and microdialogues, will bring unexpected use possibilities even in other conversational AI applications – such as AI mental health therapists or automotive voice assistants.
The course of the competition
We are now in the quarterfinals, which runs from February 3. At this stage, the teams are divided into two groups of five. The four best teams from each group will advance to the semifinals. As in previous years, the quarterfinals ratings go from users. Anyone who owns any Amazon Echo device with Alexa voice assistant can say “Alexa, let’s chat”, Alexa randomly selects one of the contestants and the user rates the conversation by one to five stars after it is over. The goal is to get the highest average score. According to backstage information, our team is doing great, so it will certainly not be missing amongst the eight best players in the semifinals that is to be announced on March 17. The semifinals phase will run from March 20 to April 29.
Winner will be known in June
The best three teams will advance to the finals, while two of them will be selected by the Amazon on the basis of scientific contributions and other solution qualities. In the finals, which will take place at the end of May and will result in the announcement of the winners in early June, the chatbots are judged by an expert jury. The winner will get USD 500 000, the second team USD 100 000 and the third will be awarded USD 50 000.
Team that manages to hold a twenty-minute continuous conversation in the finals will bring a million-dollar research grant to its university. But this is a “task of complexity comparable to landing on Mars”, said Rohit Prasad, research chief of Alexa AI from Amazon.
Follow news and more info