The search giant debuted an initiative on Monday that brings together various Google researchers to study how people interact with software powered by AI technologies like machine learning.

Companies like Facebook (FB) and Google (GOOG) have been using AI to improve tasks like quickly translating languages and recognizing objects in pictures. But the technology has the potential to be able to do more.

The problem for companies like Google is to figure out more uses for AI beyond simply improving existing products and create entirely new products based on AI.

Get Data Sheet, Fortune’s technology newsletter.

One way Google hopes the project, called PAIR (short for People plus AI Research), will lead to more compelling uses of AI is to “focus on the ‘human side,’” Google researchers Martin Wattenberg and Fernanda Viégas wrote in a blog post. They want to figure out how and where to best use it from a human standpoint—and not just simply create AI-powered software for its own sake.

“We don’t have all the answers—that’s what makes this interesting research—but we have some ideas about where to look,” the two researchers wrote.

Some of PAIR’s goals include looking at how professionals like doctors, designers, farmers, and musicians could use AI to “aid and augment” their work. The researchers did not mention how exactly PAIR will do accomplish this in the Monday announcement, but Google has been already looking at how AI can aid specific industries like healthcare through its DeepMind business unit, for example.

The initiative also hopes to discover ways to “ensure machine learning is inclusive, so everyone can benefit from breakthroughs in AI.” Left unsaid is the fact that big companies like Google and Facebook are hiring many of the top leaders in areas like deep learning, which has led to some academics questioning whether big companies are hoarding AI talent and failing to share breakthroughs in AI to increase their own profits.

The researchers also wrote that PAIR would create AI tools and guidelines for developers that would make it easier to build AI-powered software that’s easier of troubleshooting if something goes wrong. One of the ways AI-powered software is different from traditional varieties is that conventional testing and debugging methods fail to work on AI software that constantly changes based on the data it ingests.