Pitfalls and Possibilities in Designing with Generative AI

Explore the complex relationship between human intuition and generative AI in this Allen School Colloquia talk by J.D. Zamfirescu-Pereira from UC Berkeley. Discover how language-based interfaces for creating text, images, and code often lead users to misapply human reasoning patterns, limiting their effectiveness. Through case studies in chatbot development and interactive programming, learn about the tensions that arise when users approach AI instruction based on human-human communication models, and examine innovative solutions that leverage AI's unique strengths like generating diverse alternatives and cross-domain connections. Zamfirescu-Pereira, a PhD student whose paper "Why Johnny Can't Prompt" became CHI's most-downloaded publication, presents a forward-looking vision for computational scaffolding that enhances our ability to use language as a creative tool. This 56-minute presentation offers valuable insights for anyone interested in the future of human-AI co-design and effective generative AI utilization.