Despite all the advances and benefits of artificial intelligence (AI), integrating the human capabilities of creativity and visual knowledge when it comes to problem-solving have always been a missing element. A Carnegie Mellon University (CMU) research team, which is part of a larger project involving the study of human and computer teams sponsored by the Defense Advanced Research Projects Agency (DARPA), recently published its work on an AI framework that observes human design data to generate its own designs without defined rules.
“The AI is not just mimicking or regurgitating solutions that already exist,” said Jonathan Cagan, professor of mechanical engineering and interim dean of CMU’s College of Engineering. “It’s learning how people solve a specific type of problem and creating new design solutions from scratch.”
The team developed a two-step framework as a way for the AI to learn purely from human design strategies. The goal was to have the technology imitate and then develop and implement a unique design. For initial testing, the researchers selected the complexity of a bridge truss. The AI observed the progression of design sequencing via screen pixels based on what a human engineer would see without any additional parameters or biases.
“We were trying to have the agents create designs similar to how humans do it, imitating the process they use: how they look at the design, how they take the next action, and then create a new design, step by step,” said Ayush Raina, a Ph.D. candidate in mechanical engineering.
The team’s framework includes multiple deep neural networks that are prediction based. After the AI went through five of the sequential images, the neural network assisted it in mapping the next design sequence based on the previous steps, eventually leading to the AI imagining its own design.
Once designs were generated, they were compared with data from human teams facing the same design issues. The results were efficient and feasible designs. The researchers also tested the AI framework on other engineering structures, which produced similar outcomes. Surprisingly, some of the tests produced better outcomes than those created by the human counterparts, although the testing used only visual learning and not other necessary elements such as weight considerations or feedback.
While this research is a step toward creating potential symbiotic human and AI teams for design and engineering, which could eventually produce better results than either could do on their own, that future has not quite arrived.
“It’s tempting to think that this AI will replace engineers, but that’s simply not true,” said Chris McComb, an assistant professor of engineering design at Pennsylvania State University. “Instead, it can fundamentally change how engineers work. If we can offload boring, time-consuming tasks to an AI, like we did in the work, then we free engineers up to think big and solve problems creatively.”
Interested in more AI innovations? Check out New Technique Helps AI Learn Almost 70 Percent Faster and Robots Use AI to Learn from Their Mistakes.