Since June 2017, MIPT’s Neural Networks and Deep Learning Lab has been involved in iPavlov, a project that aims to create conversational artificial intelligence. The laboratory has now partnered up with Mobile Medical Technology (MMT), a developer of telemedicine solutions. The company will provide the researchers with depersonalized data for training and testing dialogue models and state its own requirements for a functional dialogue system. This collaboration will contribute to the development of the laboratory’s public library, DeepPavlov, which is intended for the creation and training of conversational agents.
Photo. Left to right: Mikhail Burtsev, the head of the Neural Networks and Deep Learning Lab, and researcher Mikhail Arkhipov. Credit: Evgeniy Pelevin/MIPT Press Office
“Chatbots have been around for quite some time now. They are increasingly being used by companies for communicating with customers,” says Olga Kairova, deputy head of the laboratory. “However, most of the available dialogue systems are still unable to have a meaningful conversation with the user: identify intentions, request the necessary information, and achieve the goals appropriate to a given dialogue.”
“Developing complex conversational agents capable of using natural language at a reasonable level is difficult and not always feasible,” she adds. “Project iPavlov aims to provide developers with a set of universal building blocks for creating dialogue solutions. DeepPavlov is a library of algorithms and components that can be used to build an intelligent conversational agent. We make all the results of our work immediately publicly available: The developers are encouraged to use DeepPavlov to build a chatbot from scratch or create a ‘brain’ for their already existing system.”
Like its analogs used by European and American companies, the conversational agent of MMT requires that the users answer its questions. A similar system is used by the British Babylon project, helping customers make sense of the disease classification used by doctors. By now, 40,000 conversations between patients and the staff of Pediatrician 24/7 and Online Doctor — two Russian web-based services — have already been analyzed.
“We decided to abandon the conventional chatbot configuration that we tested last year,” explains Denis Yudchitz, the CEO of MMT. “Based on the data we obtained, it is possible to manage the workload of doctors and gauge the efficiency of instant consultations. Artificial intelligence clearly has a lot of potential for applications in many areas, and this one is not an exception. AI obviously is not going to provide medical consultations on its own, but it can collect all the necessary information in advance to put a doctor in a better position to address the problem immediately.”
The researchers of the neural networks lab see their partnership with MMT as a stepping stone that brings them closer to an industry aware of conversational intelligence and its vast potential. With that object in mind, the laboratory aims to collaborate with other leading companies on Russia’s emerging NeuroNet market, which is prioritized by the country’s National Technology Initiative.