LLMs + Humans >> LLMs or Humans

While generative AI has the potential to help people be smarter and more creative, current approaches often risk hallucinating, missing crucial context, or homogenizing creative work. Even if successful, LLM systems risk atrophying the critical thinking and creative processes we value as humans. Our research explores a future of composite cognition, in which humans, LLMs, and other computational agents work together in partnership, each augmenting the other.

Instead of replacing the deep contextual nuances of human cognition, diverse computational elements, and rich visual interfaces with one-size-fits-all LLMs and chat, we instead aim to augment cognition, computation, and interfaces with diverse LLMs and agents. In doing so we are unlocking new ways of combining human and LLM intelligence resulting in better decisions and more creativity than using either humans or LLMs alone. Some recent examples of our work include:

🐢BioSpark (CHI2025)
Using LLMs to help people find analogical inspirations from nature and adapt them to solve complex design and engineering problems
[coming soon]

✏️InkSpire (CHI2025)
Using multimodal LLMs to power an interactive analogical sketching interface that increases designers' agency and expressiveness
[paper][website]

🌐Selenite (CHI2024)
Augmenting sensemaking on the web with LLMs by helping people build personalized overviews of unfamiliar topics while avoiding hallucinations
[paper][video]

🔬Synergi (UIST2023)
Combining graph mining and LLMs to help people explore threads of scientific literature instead of just individual papers.
[paper]