Collaborative Research: Advancing Generative AI for Engineering Design through Neurocognitive Alignment

The goals of this project are to: (1) understand and model the alignment between generative artificial intelligence and humans during early-stage engineering design; and 2) use this knowledge to develop brain-informed generative AI systems for design concept generation. Generative AI systems are increasingly being used to generate design concepts, explore alternatives, and support creative decision-making. However, these systems are typically trained and evaluated using large-scale language or image data, rather than direct empirical evidence about how designers actually think, represent ideas, and move through the design space. As a result, current AI design tools may produce outputs that do not reflect human design intent or support the neuro-cognitive processes involved in concept generation and exploration. This project addresses this gap by comparing human behavioral and neural representations of design concepts with internal representations from large language models. The work advances foundational and use-inspired AI research by developing new methods to evaluate generative AI systems against empirical measures of human cognition in engineering design. It also contributes to human-AI teaming by identifying when and how AI representations align with designers during creative technical work. The resulting knowledge will support more human-centered AI tools for engineering design and provide reusable methods for studying AI systems in open-ended creative engineering design contexts. The research will combine generative AI development, controlled design experiments, and functional magnetic resonance imaging. First, the team will develop a language-based AI platform that supports conceptual engineering design by identifying and presenting exploratory structures from human-generated design ideas. Second, behavioral studies will examine how designers use these AI-generated structures to reinterpret, extend, and regenerate concepts, and will compare human and AI representations using semantic-distance measures, design-quality ratings, and qualitative analysis of think-aloud protocols and interviews. Third, an fMRI study will measure brain activity as participants complete concept generation, exploration, and regeneration tasks. These neuroimaging data will be compared with LLM representations of the same design concepts using representational similarity analysis and encoding models. Together, these studies will produce new knowledge about human-AI alignment in engineering design, reusable behavioral and neuroimaging datasets, task materials, and analysis code. The project will train graduate students at the intersection of artificial intelligence, engineering design, and cognitive neuroscience, contributing to an AI-ready engineering workforce. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.