7 p.m.
AI is often presented to the public as magic: you type a question, and an answer appears. But there is nothing magical about it. Every AI system is built from human choices: what data to use, what to measure, what to optimize, what risks to take seriously, and who gets access to the results. Today, many of the most powerful AI systems are available mainly as closed products. That isn’t good enough for science, education, public service, or democratic accountability. Researchers and communities need more than access to a chatbot. They need open AI infrastructure: the data, code, training methods, documentation, evaluations, and intermediate models that make it possible to understand how these systems work and to build new ones for public purposes. In this talk, I will describe examples of building AI in the open. I will argue that open infrastructure is essential if universities, nonprofits, governments, and scientific communities are going to use AI responsibly and creatively. The goal is not to reproduce commercial AI, but to make room for a broader, healthier, more public-minded AI ecosystem.
Biography:
Noah A. Smith builds systems to augment human intelligence. His work advances language models, natural language processing, music processing, machine learning, and multifaceted AI system evaluation, with an emphasis on making AI more scientifically useful, efficient, environmentally sustainable, transparent, and broadly accessible. He mentors junior researchers, leads open language modeling efforts, develops infrastructure for open multimodal AI, and writes for both technical and general audiences, including Language Models: A Guide for the Perplexed.
Department Sponsors: The Graduate School, UW Bothell