Monojit Choudhury

Microsoft

Session: AI: Feats, Limits and Caveats

Research papers

Artificial Intelligence is changing the way we live, think and feel. It has instilled hope for a better world, as well as fear of being taken over by robots (even if not literally). How much of this is hype, and how much reality? In this talk, we will present my personal perspective on AI, based on our two decades of research experience in Natural Language Processing (NLP), AI and cognition, primarily as a scientist, but also from what Iwehave seen happening at Microsoft’s development centres and many start-ups with whom we have closely interacted with. Taking the case of Natural language understanding and dialogue systems as a running example, on one hand we will try to explain why Artificial General Intelligence is still a dream quite far away; on the other hand, we will try to argue that the most crucial aspect of building practical and useful AI systems is deep understanding of the domain, the user and the problem that one is trying to solve. Techniques like deep learning and platforms like Tensorflow, AzureML or MS Cognitive Services have improved the accuracy of AI systems and drastically reduced the time to production and scale; nevertheless, there are problems, seemingly very simple ones, that are much harder to solve not because of technology, but for a variety of other circumstantial reasons. In particular, we will highlight some unique and interesting challenges for the Indian market and user