Bloomsbury AI simplifies the access to information through AI.
The London-based startup company developed algorithms that can learn to answer questions by automatically “reading” textual documents. The approach heavily uses deep learning and can scale to millions of datapoint and has been improved to perform real-time predictions. In this talk, we demonstrate the current ability of machine reading algorithms in different domains and for various degrees of difficulty. One of the most promising applications is our legal product, designed to help lawyers find the information they need much more easily than current approaches based on boolean search and keyword matching. We conducted quantitative experiments that show that the time spent searching for the right answer in a document can be drastically reduced by using machine reading algorithms. Finally, we discuss some insights about the next generation of machine reading algorithms that we can directly reason on legal documents, and even search for extra information when it is missing.