Methodologies for forecasting social impacts

AI systems can create effects far beyond their immediate use. Forecasting helps designers anticipate these broader impacts.

Here are the key approaches to predicting AI social impacts:

• Use scenario planning to develop multiple possible futures. For a facial recognition AI, create scenarios for beneficial use (finding missing persons), harmful use (unauthorized surveillance), and unexpected and undesirable consequences (changes in public behavior).
• Hold consequence scanning workshops that bring together diverse viewpoints to identify potential outcomes across different timeframes. Include technologists, ethicists, potential users, and representatives from communities who might be affected.
• Create “what-if” design exercises that explore how your AI might evolve or be used in unexpected ways. Think about edge cases and unintended audiences. For example, imagine a shopping recommendation algorithm being used by kids, leading them to discover and make expensive purchases without adult supervision.
• Develop impact assessment frameworks specific to your AI domain. For healthcare AI, assess impacts on patient autonomy, access to care, and clinical workflows. For content moderation AI, consider effects on free expression, community safety, and creator livelihoods.
• Map both direct impacts on users and indirect effects on non-users, communities, and institutions. An AI hiring tool affects not just applicants and employers but potentially entire job markets and communities.