Key Moments
Stanford CS547 HCI Seminar | Autumn 2026 | Designing AI for Societal Good
Stanford OnlineKey Moments
Designing AI for societal good requires shifting from creating harm-free apps to building systems that measure and mitigate harm, influencing public opinion and policy alongside technology development.
Key Insights
Current approaches to designing AI for societal good, which focus on creating less problematic applications and then advocating for policy changes, are insufficient due to a lack of public awareness and political will.
Historical analysis of technologies that influenced policy shows that public awareness and civic engagement play a more significant role than initially assumed, preceding and driving policy changes.
Simple, technologically-based measurement systems are crucial for making elusive societal harms quantifiable and actionable, laying the groundwork for future policy interventions.
Developing AI that can detect societal harms, such as those related to mental health or misinformation, requires AI itself, creating a dependency on funding for public monitoring programs.
Entrepreneurial efforts in 'startups for humanity' demonstrate a 'seed-planting' design philosophy, where new business and policy models are intentionally cultivated to support socially beneficial technologies.
The financial aspect is a critical, often overlooked, enabler for societal good technologies, requiring significant investment for public monitoring and influencing public opinion and policy.
The limitations of designing 'harm-free' AI applications
The prevailing approach to addressing the societal impacts of generative AI focuses on creating applications that are less harmful or more beneficial. For instance, if an AI chatbot is suspected of reducing creativity, the proposed solution is to build an alternative AI that actively fosters creativity. However, this strategy faces a significant hurdle: these improved designs are unlikely to be adopted by major AI providers like Claude or ChatGPT due to their focus on user engagement and revenue. Furthermore, users themselves may be hesitant to switch from readily available commercial chatbots. This leads to a 'catch-22' situation where even well-intentioned, well-designed systems struggle to gain traction and make a broad societal impact. Researchers and designers often find themselves calling for policy changes to mandate the adoption of their beneficial designs, but these calls are rarely heeded by policymakers, resulting in a stalemate. This lack of uptake leads to feelings of despair and an identity crisis among designers who want to contribute positively but feel their efforts are futile in the face of market forces and user inertia.
Rethinking 'designing AI for societal good' as planting seeds
The lecture proposes a fundamental shift in how we approach designing AI for societal good. Instead of solely focusing on creating 'harm-free' applications, the emphasis should be on developing 'systems for harm measurement and mitigation.' This approach acknowledges that understanding and addressing complex societal harms, such as those affecting mental health or individual creativity, is an ongoing process. The key insight is to move beyond simply building a better product and to actively cultivate the broader ecosystem needed for that product to have a positive impact. This includes intentionally designing for changes in public opinion, policy, and business models. These changes are not immediate; they are akin to planting seeds that will eventually bear fruit. The presentation argues that neglecting these broader factors early in the design process, and treating them as secondary concerns to be addressed after deployment, severely limits the potential for technology to effect meaningful societal change. Therefore, influencing public opinion, policy, and business models should be considered core design goals and metrics of success, not afterthoughts.
Learning from history: technologies that shaped policy
To understand how to effectively influence policy and achieve broad societal impact with technology, the research team conducted an extensive review of historical precedents. This involved delving into Science and Technology Studies (STS) and examining how various technologies have impacted society and policy over time. They found that Human-Computer Interaction (HCI) researchers have historically influenced policies related to privacy, accessibility, social media, and content moderation. However, the specific mechanisms and stories behind these policy changes are often not well-documented or shared within the HCI community. To address this, a workshop was held in 2023, gathering over 80 researchers from different continents to share their experiences in engaging with the policy world, including successes and failures. This historical exploration revealed that the path to policy influence is often long and complex, involving more than just designing a better system and asking policymakers to adopt it. It highlighted the crucial role of public awareness, sustained evidence generation, and navigating intricate political landscapes.
The emergent roadmap: measuring harm, influencing public opinion, and policy
The research findings suggest a revised roadmap for designing AI for societal good, moving away from the two-step model of creating less problematic apps and then calling for policy changes. Historically, while these two steps have played a role, they are part of a much longer and more intricate process. Firstly, the mere design of a system, whether harmful or not, is insufficient. For example, simply creating an 'less addictive' social media platform doesn't address the root issue if individuals cannot thrive without it. Similarly, the impact of AI companions on loneliness in countries where they were banned highlights the need for individuals to first develop the capacity to live without potentially harmful technologies. Secondly, policy change, while important, is heavily dependent on public opinion. Policies in democracies tend to reflect public sentiment; therefore, without a public aware of the need to reduce AI use for personal well-being, radical policy shifts are unlikely. This underscores the critical importance of public awareness and civic engagement. The research also emphasizes the significant role of simple, technology-based measurement systems in quantifying elusive societal harms. These measurements, even if approximate, can empower individuals to take action, gather evidence for policymakers, and gradually shift public opinion over time. This evidence-gathering process can be slow, taking years, but initiating it early is crucial to be ready when political will emerges.
The financial bottleneck and the 'startups for humanity' model
A significant challenge identified is the financial aspect of developing and deploying technologies for societal good. Detecting and measuring many AI-induced societal harms, especially those occurring in natural language and multimedia, requires AI itself, incurring substantial costs. These costs are unlikely to be borne by commercial AI providers or third parties. Therefore, public funding is often necessary, which, in a democracy, requires political will and public support. This creates a cycle: public funding for monitoring is needed, which requires evidence of harm to shift public opinion, which in turn requires money for measurements. This financial bottleneck may explain the historical slowness in addressing technology's societal impacts. The research explores a nascent movement of 'startups for humanity' and 'entrepreneurs for humanity' who are actively trying to solve these problems. These entrepreneurs view the design process as 'planting seeds,' intentionally cultivating new business and policy models to support socially beneficial technologies, especially in areas like digital health where long-term well-being isn't always financially rewarded. They understand the regulatory and business environments deeply, designing their technologies to be financially viable within existing systems or by creating new ones. While many of these ventures are experimental and may fail, the hope is that a single success can pave the way for future ventures to follow.
Empowering societal well-being through early evidence and multi-stakeholder negotiation
The research concludes that while designers are adept at creating impactful technologies, making them effective for societal good requires a broader approach. The conventional wisdom of building harm-free tech and waiting for policy adoption is insufficient. Instead, the focus must shift to actively measuring societal harms, influencing public opinion, and driving policy changes, recognizing these as integral design objectives. The HCI community, with its skills in identifying potential harms early and mediating between diverse stakeholder needs, is well-positioned to tackle this complex challenge. However, it requires a departure from simply publishing research on, for example, creativity-enhancing AI. The real challenges lie outside the technology itself, in areas like policy, public opinion, and business models. The lecture emphasizes that not all evidence translates directly into policy change; often, evidence is framed as individual responsibility rather than collective harm. Therefore, developing skills in policy design, public engagement, and understanding business models is crucial for designers aiming to create technologies that genuinely benefit society. The ongoing projects presented, such as measuring the impact of chatbots on well-being and developing tools to enhance interpersonal communication, exemplify this new direction, focusing on harm measurement and fostering capabilities that reduce reliance on potentially harmful technologies.
Mentioned in This Episode
●Software & Apps
●Companies
●Organizations
Common Questions
Generative AI raises concerns about job displacement, increased reliance on chatbots over human interaction, and potential impacts on self-esteem and mental well-being due to interactions with overly agreeable AI.
Topics
Mentioned in this video
Mentioned as a large organization that develops AI technologies like ChatGPT, which are unlikely to proactively adopt user-centric designs due to business needs.
Mentioned as a platform where code repositories are stored, contrasted with the need for AI models to analyze natural language and multimedia for detecting societal harm.
A US government health insurance program mentioned in the context of how digital health startups navigate reimbursement to make their socially beneficial technologies financially viable.
A US government health insurance program discussed in relation to startups seeking reimbursement for socially beneficial health technologies.
More from Stanford Online
View all 162 summaries
75 minStanford AA274A Principles of Robotic Autonomy | Autumn 2019 | The Robot Operating System (ROS)
61 minStanford CS547 HCI Seminar | Autumn 2026 | Sensing Uncertainty in Sustainable HCI
79 minStanford AA274A Principles of Robotic Autonomy | Autumn 2019 | EKF Localization
57 minWebinar: Regulating at the Speed of AI with Francis Fukuyama and Nate Persily
Ask anything from this episode.
Save it, chat with it, and connect it to Claude or ChatGPT. Get cited answers from the actual content — and build your own knowledge base of every podcast and video you care about.
Get Started Free