The Stanford webinar “Human-Centered AI: Designing Systems People Trust” featured Professor James Landay and moderator Vanessa Parli, both from Stanford’s Institute for Human-Centered Artificial Intelligence (HAI). The session began by exploring what “human-centered AI” means, emphasizing that it goes beyond simply designing for user needs or ethical considerations. Landay explained that human-centered AI requires a holistic approach, considering not just direct users but also communities and society at large. He highlighted the need to move past traditional user-centered design to include those indirectly affected by AI systems, such as people impacted by healthcare or public safety decisions, and to anticipate broader societal effects like those seen with social media platforms.
A significant portion of the discussion focused on the cultural and societal implications of AI. Landay shared research showing that large AI models often reflect the dominant cultures present in their training data, which can lead to a mismatch between AI outputs and the values or perspectives of users from less-represented cultures. He illustrated this with an example about how different cultures conceptualize a “tree,” demonstrating that AI models tend to default to Western perspectives unless specifically prompted otherwise. This led to a conversation about the emerging concept of “sovereign AI,” where countries seek to control AI infrastructure, data, and models to protect national security, economic interests, and cultural identity.
The webinar also addressed practical applications of human-centered AI in fields like health and education. Landay described projects from his lab, such as an AI-powered fitness coach app called Bloom, which uses motivational interviewing techniques to help users set and achieve health goals. The app provides personalized feedback and encouragement, aiming to foster long-term behavior change. He emphasized that effective AI solutions require integrating best practices from relevant domains—like coaching or education—rather than relying solely on generic AI models.
Audience polls and survey data were used throughout the session to gauge public trust in AI and perceptions of its benefits and risks. The results showed a mix of optimism and skepticism, with concerns about privacy, job displacement, bias, and transparency being prominent. Landay noted that optimism about AI tends to be higher in countries with lower per capita incomes or where access to services is limited, while Western countries are more cautious, often due to concerns about privacy and surveillance. He argued that building trustworthy AI requires openness, transparency, and international collaboration, such as open-source models and global partnerships.
In closing, the speakers discussed the future of AI and how individuals and organizations can prepare. Landay predicted that AI interfaces will evolve rapidly, becoming more multimodal and context-aware, and that education systems will need to adapt fundamentally to the presence of generative AI. He advised individuals to actively engage with AI tools to understand their strengths and limitations, and encouraged ongoing education through resources like Stanford HAI’s seminars and online courses. The session concluded with a call for continued collaboration between academia, industry, and policymakers to ensure that AI development remains human-centered and aligned with societal values.
