Machine Readings Understanding (MRC) is an artificial intelligence capability that understands the specific knowledge embedded in different sources - a necessary skill for many real world scenes. For example, in a search application, it helps the certification authority to give a precise answer rather than the URL of the page, and in the future, MRC can even help doctors find information in thousands of documents, reduce time-consuming tasks and possibly Improve the medical profession.
But the current machine reading system is usually based on supervised training data, which means that they not only use the article they should understand the training, but also need to manually mark the problem of these articles, and give the appropriate answer. However, this approach is not extensible because the labeling process must be completed for any knowledge domain and for each knowledge domain. For example, it is necessary to create an MRC for each disease in order to help physicians to establish artificial intelligence, and each disease should be continually updated due to the increasing number of articles being produced in the literature.
Microsoft is now taking SynNet, Microsoft's new "two-phase integrated network" training model, SynNet first from a field to learn key knowledge points or semantic concepts, and based on available monitoring data to learn, and then learn to form their own natural language, Given the context of the article, give the answer to the question.
The most interesting aspect of SynNet is that once trained, it can be used in a new field to generate false questions and answers for a given article. This approach enables it to create the training data needed to train specific MRCs, eliminating the need for manual labeling and making SynNet an AI teacher. SynNet is still in its infancy, and not everyone is happy with the direction in which AI is developing. Tesla boss Elon Musk warns the US president that they need to be alert to AI and make adjustments.