How Artificial Intelligence Can Improve Engineering Design
Note from the author: this is an adapted, shortened version of an essay I wrote for the NSF/ASME Student Design Essay Competition for IDETC 2023. You can read the full article here, in which I address how machine learning can improve product realization for sustainable development.
Artificial Intelligence (AI) and Machine Learning (ML) have become household terms in homes around the world. They refer to the effort to make programs, machines, and systems that demonstrate intelligent, self-learning behaviors. Have you thought about how AI will affect your design work? Maybe you’ve read our article - written by ChatGPT - about how ChatGPT can help designers. Here, I take a step back from talking about ChatGPT to think more about AI in general. I’m excited for the ways that AI can facilitate product realization work, and in this article I describe why I’m excited and two ways I’ve seen AI in product realization already.
Machine Learning (ML) is a subset of AI defined as the capability of AI systems to “acquire their own knowledge by extracting patterns from raw data” [1]. Using ML and AI has started many discussions around how these technologies should be used in creative work. For example, many content creators worry about AI impersonating their work. Or, a creator’s work could be used in training models and reference images without ever receiving credit, recognition, or compensation. Some valid questions include, how will artists compete and produce art in a world of AI? Or, How will consumers differentiate between AI generated art, and what has not been AI generated? Fortunately, several market leading companies are collaborating on standards for the use of AI and ML in digital, creative work [2].
Despite these challenges, consumers and designers are interested in using AI to speed up their workflow, spend less time on monotonous tasks, and spend more time rapidly iterating through ideas. For example, note how AI tools make graphic design more accessible to beginners without graphic design experience or experience using complicated software. Adobe advertises that their creative AI tools “unlock your creative superpowers” and “enhance the creative process” [3]. Images or scenes that used to take a long time creating can now be created or edited in seconds, which reduces the time between iterations of art pieces. The creator can spend less time on the creation of the artifact, and can spend more time ideating, evaluating, and iterating.
Similarly, AI tools can speed up or facilitate design engineering activities. Product development tasks that were typically time intensive and error prone could be replaced with quicker alternatives – such as using AI tools to create mechanical drawings, process diagram sheets, or 3D models. Furthermore, AI can lower the ability threshold [4] for many complicated engineering design activities, allowing more people to participate in product realization.
One way we are already seeing AI facilitate design activities is through program synthesis. Program synthesis is “an AI system that takes human-understandable instructions and transforms them into computer-understandable instruction” [5]. For example, chatGPT can write you a for loop if you give chatGPT the simple command, “write me a for loop in python.” Or, you could ask Firefly to generate “a tree with a bird in it.” Program synthesis takes commands intuitive to humans and turns them into computer commands, which can put novice software users on a more level playing field with experienced users. New users may know what they want the software to do, but don’t know which command to give the computer. For example, maybe you’ve had to ask your boss, professor, or a teaching assistant “How do I do this operation in CAD?” More than answering your question, a program synthesis tool would be able to complete the operation for you to some degree, as long as you can describe it well.
A ML researcher wrote, “to make good use of ML tools it is instrumental to understand its underlying principles at the appropriate level of detail. It is typically not necessary to understand the mathematical details of advanced optimization methods to successfully apply deep learning methods” [6]. As the mathematical details of product realization work get allocated to AI systems, new engineers and designers will only need to understand what the more fundamental principles of the design and manufacturing work are to make meaningful contributions in product realization. Similar to how program synthesis opens the door for more people to use complex software, removing the requirement to understand the mathematical details of engineering activities could let more people participate in product realization.
We will of course have to decide what processes and workflows for using AI in engineering design need to be created. Safeguards will need to be enacted to ensure the quality of the artifacts being produced and that safety, ethical, and technical standards are reviewed. In this capacity, expert review will always be required at some level. It must also be decided which “underlying principles,” and at what “appropriate level of detail,” are required for people to know before they engage in AI assisted product realization activities. Determining these safeguards and principles are important research areas for designers to consider before incorporating AI and ML into their regular workflow.
But I am optimistic to see how AI enables more people to participate in engineering design and improve the quality of the design work being done. In fact, we are already seeing benefits of using AI in product realization. For example, a research collaboration between researchers at SRI International, SwRI, Vanderbilt, CMU, and Purdue used ML design methods to create a diverse data set of 27,000 air vehicle designs [7] . What’s impressive about this data set is not only its quantity of designs, but its variety. Think about the variety of air vehicles you’re most common with. It probably consists of quadrotor consumer drones, helicopters, and some planes. But if you look around this data set, you’ll see six rotor vehicles, two rotor two wings, ten rotor drones… go check it out [8].
Figure 1. Three air vehicles from the AircraftVerse database [8].
I’m excited for the benefits that AI will bring to engineering design. Designers will be able to spend more time on creative work, and lowering the ability threshold will enable more people to participate in engineering design, leading to a greater variety of novel, quality ideas in the pursuit of sustainable solutions.
References
[1] Goodfellow, I., Bengio, Y., Courville, A. Deep Learning. MIT Press, 2016.
[2] Coalition for Content Provenance and Authenticity, accessed May 26th 2023. https://c2pa.org/
[3] Adobe Firefly, accessed May 26th 2023. https://www.adobe.com/sensei/generative-ai/firefly.html. Other AI softwares for graphic design or art include Midjourney and DALL-E.
[4] The ability threshold is a minimum level of skill required to participate in a given activity. It is also referred to as a competency threshold or minimum threshold. In this article, I am saying that AI lowers the skill to do a design activity because the program shoulders some of the technical burden, making it easier for a new user to get started.
[5] This definition was shared at a workshop I attended at the University of Maryland by Autodesk researcher Yewen Pu. Learn more at his blog, So, you want to know about program Synthesis? Accessed July 17th 2023. https://evanthebouncy.github.io/program-synthesis-minimal/what-is-synthesis/
You can also reference this article, which Pu is an author on:
Acquaviva, Sam, et al. "Communicating natural programs to humans and machines." Advances in Neural Information Processing Systems 35 (2022): 3731-3743. https://arxiv.org/abs/2106.07824
[6] Jung, A. Machine Learning the Basics. Springer Verlag, Singapore, 2022.
[7] Cobb, Adam D., et al. "AircraftVerse: A Large-Scale Multimodal Dataset of Aerial Vehicle Designs." arXiv preprint arXiv:2306.05562 (2023). https://arxiv.org/abs/2306.05562
[8] Aircraft verse, accessed July 17th 2023. https://aircraftverse.onrender.com/
