At MIPT, the DeepHack.Turing sci-hack summer school has begun. This event is part of the effort to automate the Turing test, a classic way of telling if a machine can think. The participants will spend a week designing competing models that analyze text to differentiate between human and machine input in a dialogue. For those wishing to join in the fun: Anyone can help the researchers create better artificial intelligence by talking to chatbots.
Computer models capable of identifying the author of messages in a conversation could lead to a breakthrough in dialogue systems, producing conversation agents that can effectively engage in dialogue with the user. According to Mikhail Burtsev, the head of the Laboratory of Neural Networks and Deep Learning at MIPT, the availability of training data is critical to artificial intelligence research. "When it comes to chatbot development, training data means real dialogues," he explains. "That's why we're inviting everyone to have a conversation with the robot about some excerpts from Wikipedia in English. This will help us get hold of enough data to enable further training of algorithms and improve chatbots."
Those participating in the experiment will be matched with an unknown and unseen conversation partner that might turn out to be either a chatbot or an actual person. They will then be asked to grade the intelligence of the partner. To take part in this research, you should enter into dialogue — the messages are in English — with @ConvaiBot on Telegram. The data gathered in this experiment will power next-generation dialogue systems.
As before, the DeepHack summer school and hackathon event features lectures on deep learning in the context of natural language processing by the world's leading specialists from Facebook AI Research, Sentient Machines, Cambridge University, New York University, Carnegie Mellon University, KAIST, and other research centers. This time, particular attention will be devoted to the practical cases of applying AI technology in dialogue systems. Admission to the lectures is free, but registration at http://deephack.me is required.
The hackathon has been organized by Burtsev's laboratory alongside the iPavlov project. The latter brings the laboratory staff together with researchers from the major research centers in the field of machine learning on a three-year venture aimed at developing conversational intelligence technology. The results will eventually be published as an open-license library available to anyone who wishes to use them in research or applied solutions.
The main partners of the event are the National Technology Initiative's Fund for Project Support, Sberbank, MIPT, and Facebook. The hackathon was also supported by Flint Capital, an international venture capital fund, as well as 1C Company, a software developer; MTS, a mobile network operator; BKS Broker, a brokerage firm; Hostkey, a web services provider; and the Online Doctor project. This time, the science school plus hackathon event serves as one of the stages in the international contest known as The Conversational Intelligence Challenge. Its final round will be held this December during the biggest deep learning conference there is — the Conference on Neural Information Processing Systems (NIPS) in California. The winning team in the hackathon will travel to Los Angeles to compete in the final round of the contest.