Where Current QA Systems Fail

Video produced by Steve Nathans-Kelly

Andreas Blumauer, CEO, Semantic Web Company, discussed six areas where contemporary QA systems fail during his presentation at KMWorld Connect 2020.

Text only QA systems learn typically only from documents, he said. QA systems don't work well with niche, or specific questions and finding the answers. 

"QA systems typically, produce non-expert tenable, answers. So you don't know why the answers were generated. And in many domains, in the meantime, it's very important to have explainable AI. So QA system should provide explainable answers," Blumauer said.

Most of the QA systems may work very well in English, but lots of corporations have many, many languages to have to deal with, he explained. So it's good to have a QA systems which can learn from other databases and so on.

"QA systems, need a lot of data to get trained. And yes, you can probably add more data to it. You can train your machine endlessly and you still do not reach the level of precision you really would like to deliver. So we need a QA system which can also get trained with less data," he said. 

•