An artificial intelligence system has outperformed humans on one of the world's most challenging reading comprehension tests, the Stanford Question Answering Dataset (SQuAD).
This test, developed at Stanford University (USA), contains 100 thousand questions based on more than 500 Wikipedia articles, and is considered the most difficult in the field of cognitive word processing. It attracts developers from universities and companies such as Google, Facebook, IBM, Microsoft at Carnegie Mellon University, Stanford University and the Allen Research Institute. Competing teams must create machine learning models that can provide the most accurate answers to all questions. The highest score achieved by a person on this test was 82.304. Alibaba's deep neural network system (developed by the Alibaba Institute of Data Science of Technologies) scored 82.44 points, becoming the first artificial intelligence system to outperform a human in this test. Artificial intelligence developed by Microsoft also managed to get ahead of humanity. Its results are 82.65 points, but this became known only the day after Alibaba’s results were announced. The accuracy of Alibaba's system comes from its ability to find potentially accurate answers among paragraphs, sentences, and words in text. The technology also uses NLP - natural language processing, during which machines imitate human understanding of words and sentences. For several years, Alibaba's machines have been responding to large volumes of incoming customer queries using an artificial intelligence system, while learning at the same time. The researchers believe the SQuAD tests will help lead to the creation of more advanced robots and automated systems capable of solving complex problems and answering complex questions posed by humans.
GO TO FULL VERSION