From software development to car design, engineers struggle every day with complex design situations. “Optimizing a technical system, whether it is more convenient or energy efficient, is a very difficult problem!” says Anti Vlaswirta, a professor of electrical engineering at Aalta University and the Finnish Center for Artificial Intelligence. Designers often rely on a combination of intuition, experience and trial and error to manage them. In addition to the fact that this process is inefficient, it can lead to “fixing the design”, focusing on familiar solutions, and new ways remain unexplored. The “manual” approach will also not scale to larger design issues and is highly dependent on individual skills.
Vlaswirt and his colleagues tested an alternative computer method that uses an algorithm to search in the design space, set possible solutions given multidimensional input and constraints for a particular design problem. They suggested that a managed approach could bring better projects by reviewing a wider range of solutions and balancing human inexperience and fixing design.
Together with researchers from Cambridge University, the researchers decided to compare traditional and assisted approaches to design, using virtual reality as a laboratory. They used Bayesian optimization, a machine learning technique that simultaneously explores the design space and leads to promising solutions. “We’ve included a Bayesian optimizer in a loop with a person who will try a combination of parameters. The optimizer then offers some other values and they continue in the feedback loop. This is great for developing virtual reality interaction techniques, ”explains Vlaswirta. “What we didn’t know until now is how the user feels about this optimization-oriented design approach.”
To find out, the Vlaswirty team asked 40 novice designers to take part in their virtual reality experiment. Subjects had to find the best settings to display the location of their real hand holding the vibration controller, to the virtual hand in the headset. Half of these designers were free to follow their own instincts in the process, and the other half received optimizer-selected designs for evaluation. Both groups had to select three final designs that would best capture accuracy and speed in the 3D virtual reality interaction task. Finally, subjects reported how confident and satisfied they were with the experience and how much they controlled the process and final design.
The results were obvious: “Objectively, the optimizer helped the designers find better solutions, but the designers didn’t like it when they were held in their hands and under command. It destroyed their creative potential and sense of freedom, ”Ulasvirta reports. The process led by the optimizer allowed designers to explore more design space compared to the manual approach, leading to more diverse design solutions. Designers who worked with the optimizer also reported less mental needs and effort in the experiment. On the contrary, this group also received a lower score on clarity, will, and mastery, compared to designers who conducted the experiment without a computer assistant.
“Certainly there is a compromise,” says Vlaswirta. “With the help of the optimizer, the designers came up with a better design and covered a wider range of solutions with less effort. On the other hand, their creativity and sense of ownership of the results have been reduced. ” These results are instructive for the development of AI, which helps people make decisions. Vlaswirta believes that people need to engage in tasks such as assisted design so that they maintain a sense of control, do not get bored, and get a better idea of how the Bayesian optimizer or other AI actually works. “We’ve seen that inexperienced designers can benefit from raising the AI if they participate in our design experiment,” says Vlaswirta. “Our goal is to make optimization truly interactive without compromising human activity.”
This document was selected for honorary mention at the ACM CHI Conference on the Human Factor in Computing Systems in May 2022.
Source of history: