MOSCOW, 27 Nov — RIA Novosti. The scientific staff of the MIPT and the company Facebook will hold in early December, a kind of “championship” for artificial intelligence systems, able to conduct a meaningful conversation, and take the most “humane” of them. This was reported by the press service of the Institute.
“Our goal is to raise the quality of communication between man and machine to a new level. In the project iPavlov we develop the ability of machines to conduct meaningful dialogue with a person. To achieve this goal we combine the efforts of scientific schools in Russia and the world. We work with leading universities in Canada and many other”, — said Mikhail Burtsev, a leading developer of systems for machine learning and artificial intelligence in MIPT.
Last summer, scientists from the MIPT announced the creation of a system of “conversational” artificial intelligence iPavlov the ability to mimic a living person, and invited everyone to test it and work together to make it better. Using the help of volunteers, the researchers hope over the next three years to create a voice assistant that is able to communicate with the person as well as a lively conversationalist.
In fact, these ideas weren’t invented today — modern voice assistants from Google, Apple, Amazon, and other IT companies rooted in the deep past, to the beginning of the computer era. The first “talking machine”, named ELIZA, was created in 1966, and was, in fact, a joke, a parody of a psychotherapist, giving useless advice to the patient.
In the subsequent years and decades, programmers have been creating increasingly complex and “live” communication with the computer. The most advanced of these systems can recognize the mood of the owners, to remember his old desires and preferences, and solve a part of the routine and domestic tasks for him, ordering food or goods in the store or playing the role of the operator in the call centers.
The creation of such systems, as the scientist, today is limited by the fact that they need to actually “manually” to teach communication skills, passing them through large data sets and the processed dialogue. The quality and content of these texts depends on how “humane” is a chatbot, and how well it will respond to the queries of the public.
In September of this year experts Facebook and the University of Montreal produced one of the largest datasets of this kind, Persona-Chat, which the company Mark Zuckerberg plans to teach a new version of “conversational” AI. They will add to the existing system Zo, imitating the manner of communication and behavior typical of a teenage girl.
The dialogs library contains 162 thousand phrases and 11 thousand conversations on different topics, which were anonymous web users, who did not know, with whom they communicate. It includes nearly 1200 different “persons”, styles of dialogue, which car to learn how to imitate, with only a small number of examples.
The quality of this dataset Burtsev and his colleagues will test in early December of this year at the conference NeurIPS, meeting leading experts in the world of AI, which will be held this year in Montreal. Experts from MIPT and the representatives of Facebook will hold a competition Challenge 2 Conversational Intelligence. In its framework participants will be able to use Persona-Chat for training your own chatbot and test it.
As scientists hope that this championship “chatbot” will help them not only to complement the dataset, but also to improve the work iPavlov and make it more “human” companion.
Read more •••