Generative AI (GenAI) is powerful and useful. It can also hallucinate, be overly general, or provide misleading information. On this panel, practitioners with direct experience serving vulnerable populations discuss how to responsibly implement GenAI including how to build trust among the user population and how to mitigate and implement guardrails against misleading and unsafe AI generated responses. Panelists will speak from their first-hand experience deploying GenAI to provide sensitive information to vulnerable populations. They will describe the important nuances to be addressed in safely and effectively deploying GenAI technology, sharing lessons learned, necessary pivots, and program results. Particular focus will be given to the processes of evaluating inaccurate, misleading, or non-permissible responses in generative AI. Discussions will cover known challenges like Offensive or Harmful Content, Misinformation, Inherent Biases, Fair Representation, Data Sensitivity, and Consent, along with lesser-explored issues like Unexpected User Interactions and Jailbreaks (like the famous ‘Grandmother’ exploit).

In an era of rapid technological evolution, the integration of Artificial Intelligence (AI) into the fabric of aid and development organizations requires a robust data foundation. This panel session will delve into the critical aspects of establishing strong data management, governance, and infrastructural support to harness the transformative potential of AI for social good. The discussion will explore impactful approaches to navigating the intricate intersection of foundational data practices and AI implementation. Experts from diverse backgrounds will share insights on implementing strategies that empower organizations to responsibly collect, handle, and leverage their data. Join us in this panel session to explore the dynamic landscape of data management and governance, and discover how organizations can position themselves for success in utilizing AI to address complex social challenges responsibly and effectively.