Source: Carnegie Mellon University College of EngineeringResearchers from Carnegie Mellon University have enabled artificial intelligence (AI) to create new designs.

Typically, engineers call on AI to assist with design issues with definite rules and guidelines and not to problem-solve design issues. Creative problem-solving is where humans tend to excel over AI. However, Carnegie Mellon researchers set about enabling AI to autonomously create its own designs based on observing human design behaviors.

To accomplish this, researchers applied an AI framework to the design of a truss, which is a network of rods, generally found on bridges. The AI framework was trained on sequential designs via pixels on a screen. However, that information was out of context and without a clearly defined outcome. When tasked with designing, the AI followed a sequence similar to that followed by humans.

The AI framework, which is composed of several deep neural networks working together in a prediction scenario, observed five sequential images and used that data to make predictions about future designs. The researchers determined that the AI performed better than the human engineers, designing a truss without guidance, thereby demonstrating that AI can be taught to effectively design.

The researchers, however, were quick to add that the AI was developed not as a replacement designer for engineers but as a complement to engineers.

The research appears in the ASME Journal of Mechanical Design.