By Pushti Rana - 21st August, 2025
In 1959, John McCarthy, the father of artificial intelligence, helped lay the groundwork for AI research at MIT and served as a catalyst for the institute’s early role in the field. Such contributions by pioneers have led to pivotal advances in research today. At the 2025 MIT Generative AI for Design Workshop, hosted at MIT Media Lab, researchers and industry-leading professionals across the country came together to share their knowledge on artificial intelligence, generative design, media, and machine learning models.
Six teams showcased their work on generative design, computation, and fabrication. A research team from Texas A&M University introduced a framework based on Bayesian optimization, a method that optimizes functions that have complex input and output relationships to create interfaces in materials inspired by nature. Using inverse design, they took their desired result and worked backwards to find the optimal solution. Another group consisted of professors around the country who presented a method using CAD image prompting to help text-to-image models create more feasible designs. They tested this idea by having AI generate seven different bicycle designs with varying amounts of influence from the CAD image. The resulting images were feasible and novel suggesting that the use of CAD images helps AI generate more realistic designs. This study reinforces that AI is not only useful in creative industries but with engineering as well. A third group from Arizona State University proposed creating CadQuery code directly from text, using language modeling to produce parametric 3D models. Another team focused on bridging the gap between 3D generative AI and physical models, by utilizing a robot and 3D generative AI to make physical objects.
Anna Huang is a professor at MIT and researcher at Google. Huang is designing Generative AI systems that explore new ways to interact with music. She proposed using neural networks as a lens to explore our understanding of music. She also wants to understand how AI systems interpret music and aims to improve their use for musicians. Rethinking generative AI as a means of interaction will create a collaborative environment between musicians and AI making it easier to examine their relationship and enhance it.
Another panelist and MIT professor, Caitlin Mueller, researches digital design synthesis and manufacturing. She has led the research group, Digital Structures, since 2014, focusing on new computational and fabrication methods to develop high-performance structures. The next panelist, Karl D.D. Willis, is a senior researcher at Autodesk Research. Through methods in machine learning and computational design, Willis led the Model Operations team to transform AI models into usable features for Autodesk.
An assistant professor at MIT, Mina Konaković Luković, leads the Algorithmic Design Group, using algorithms to design and fabricate architectural materials more efficiently. Wojciech Matusik, our last panelist, is also a professor at MIT. He leads the Computational Design and Fabrication Group, using methods involving additive manufacturing and design simulations to advance computational design.
The event concluded with a talk from the keynote speaker Mark Fuge. He is a professor and Chair of Artificial Intelligence in Engineering Design at ETH Zurich. He studies how we can optimize the relationship between humans and computers through tech-enhancing engineering systems and works on projects ranging from the micro scale to full-scale ships. Fuge highlighted the concept of having “machines that learn how to design other machines.” He proposed a new perspective on engineering that does not compromise flexibility or simplicity when designing. This approach captures the most important differences in designs which enables for more efficient design exploration. He presented his recent project, “Inverse Design for Aerodynamic and Heat Transfer Surfaces,” about making energy use more efficient. He and his research group created a new technique called “least volume” that improves designs for airflow and heat transfer by using the simplest design space while preserving the original data’s shape.
These conversations bring light to alternative solutions that make artificial intelligence more efficient. This event showcased how AI is expanding into unconventional industries and deepening the ways we collaborate with it.
