Here’s an artificial intelligence that comes up with its own product ideas. Researchers at Carnegie Mellon University (CMU) and the Hebrew University of Jerusalem developed a method for teaching computers to generate creative new products or new ways to solve problems.

It does this by using language as much as engineering. By mining data about how engineers find analogies, they trained a computer system that could develop new ideas.

"After decades of attempts, this is the first time that anyone has gained traction computationally on the analogy problem at scale," said Aniket Kittur, associate professor in CMU's Human-Computer Interaction Institute, in a press release posted Thursday.

"Once you can search for analogies, you can really crank up the speed of innovation," said Dafna Shahaf, a CMU alumnus and a computer scientist at Hebrew University. "If you can accelerate the rate of innovation, that solves a lot of other problems downstream."

They took inspiration from historical discoveries such as the Wright Brothers’ work on powered flight, which stemmed from the Wrights’ experience building bicycles. In order to develop a database of ideas, CMU Ph.D student Tom Hope and CMU post-doctoral researcher Joel Chan lead a group of workers sourced from Amazon Mechanical Turk in organizing groups of products found on the product design community. They put groups of similar words together – for example, braking down a yogurt maker into terms like “food” and “concentrate” and then finding products with similar keywords. This taught the computer program how to link product descriptions. Then, the crowd workers demonstrated how to recombine those terms in order to generate new products.

"We were able to look inside these people's brains because we forced them to show their work," Chan said.

While the CMU release didn’t give examples of any of the “innovative” ideas the program came up with (Uber for Yogurt, anyone?) they are confident that the approach itself is a useful step in teaching computer programs to find analogies. People have tried to teach computers how to link terms on a semantic level before through handcrafting data structures or inferring structures from large amounts of text, but CMU says that their way has yielded the most innovative – but also sensibly analogous – results.