It's more difficult to teach AI how to read than it seems
A Google research team developed a large language AI model called BERT. It wasn't made to take a test, but it did just as well as humans in answering questions.
In recent months, researchers have demonstrated that computer systems can grasp the complexities of language in general terms and then utilize this knowledge to complete a variety of specific tasks. These systems, constructed rapidly one after the other by various independent research organizations such as Google and the Allen Institute, could improve digital assistants like Alexa and Google Home along with software that automatically examines documents inside businesses such as law firms, hospitals, banks etc.
If researchers give BERT questions with their answers, it learns to answer similar questions by itself.
Sebastian Ruder, a researcher based in Ireland who collaborates with Fast.ai, sees the arrival of systems like BERT as a "Wake-up call" for him and other AI researchers because they had assumed language technology had hit a ceiling.