AI Will Soon Be Able To Design And Solve Design Problems

Dhir Acharya - Nov 18, 2019


AI Will Soon Be Able To Design And Solve Design Problems

Humans have been great in solving major design problems that require exploratory and creative decision making. But AI will soon conquer that too.

Humans have always done a great job in solving major design problems that require exploratory and creative decision making. And over the years, AI hasn’t been observed to be able to match such skills.

Engineers have employed artificial intelligence to solve a problem using a certain set of rules instead of making it follow human strategies and come up with something new.

But now, according to a US study, we can train AI agents to have them adopt design strategies of the human for solving problems.

ai-will-soon-be-able-to-design-and-solve-design-problems-1

In this research, an AI framework can observe human data to learn the design strategies of humans. Then it can create new designs without specific guidance, bias, or goal information.

Co-authoring this study is Ayush Raina along with Jonathan Cagan from Carnegie Mellon University, and Chris McComb from Pennsylvania State University.

Cagan said that the AI wasn’t only regurgitating or mimicking existing solutions but it was also learning the way people address a particular kind of problem as well as generating new solutions for designing. He stated that an AI design agent is pretty good.

The main focus of the study is truss problems as they are representative of complex engineering challenges in design. A truss, mostly observed in bridges, is an assembly of rods forming a complete structure.

ai-will-soon-be-able-to-design-and-solve-design-problems-2

The team trained the artificial intelligence agents to observe a series of design modifications followed by engineers when they create a truss. The visual data that the observations of the agents were based on and that the engineers use are the same images on the screen, but the AI didn’t have additional contexts.

When the agents designed, they imagined the design progressions similar to what humans used and generated responses to realize their designs.

Raina says:

The AI agents were tested on similar problems, from which the researchers found that their performance, on average, better than humans’. Importantly, they achieved this success without having the same advantages as humans do when solving problems.

The agents are different from humans as they work without a specific goal, making something lightweight, and they don’t receive feedback on their work. They instead

Comments

Sort by Newest | Popular

Next Story