For A.I. researchers, machine reading comprehension (MRC) has been a challenging goal, but an important one. MRC makes it possible for systems to read, infer meaning and immediately deliver answers while sifting through enormous data sets.

Last year, China’s Alibaba outperformed humans when tested by the Microsoft Machine Reading Comprehension dataset (or MS MARCO for short), which assessed its ability to use natural language to answer real questions posed by humans. Alibaba’s system delivered answers to search queries posted by people to Microsoft’s Bing, like “how many carbs are in an English muffin?” and “how do you grow hops?”

One practical application of MRC on the consumer side: When you perform a search query, wouldn’t you rather have a system offer you a precise answer than just a list of URLs where you can go to hunt down more specifics—even showing you where, on the page, that information comes from?

If you are an airline mechanic and you’re trying to troubleshoot a tricky engine problem without further delaying a flight, it would be easier if you had a computer read all of the technical documentation for you and suggest likely fixes. Or, better yet, let the machines figure out what’s wrong on their own, by making all technical manuals and documentation available to them for reading and analysis.

That’s the promise of MRC, which represents a necessary step in realizing artificial general intelligence, and in the near-term could potentially turn everything from technical manuals to historical maps to our medical records into easily searchable repositories of information.