Harnessing AI for Anticipatory Action in Humanitarian Aid

Thomas Byrnes Humanitarian & Social Protection Consultant | Tech-Driven Innovation for Cash Transfers, Economic Analysis & Digital Transformation

Introduction

As climate change intensifies, the frequency and severity of natural disasters are escalating, disproportionately affecting vulnerable communities. In this context, anticipatory action—acting before disasters strike—is becoming a critical component of humanitarian aid. Recent developments in artificial intelligence (AI) are enhancing our ability to forecast disasters and implement early interventions.

Drawing insights from the recent Global Cash Forum host by CALP Network and the innovative collaboration between Google and GiveDirectly in Nigeria, this post explores how AI is revolutionizing anticipatory action and what this means for the future of humanitarian aid.

The Urgency of Anticipatory Action

Lessons from Recent Disasters

The devastating floods in Libya's port city of Derna, resulting in over 20,000 deaths, highlight the catastrophic consequences of inadequate preparedness. As Anita Auerbach (née Yeomans) the Cash and social protection advisor from the Anticipation Hub emphasized during the Global Cash Forum yesterday, many of these tragedies could be mitigated—or even prevented—with effective early warning systems and timely interventions.

The Increasing Predictability of Disasters

Advancements in AI and forecasting technologies are improving our ability to predict disasters with greater accuracy and longer lead times. For example, the ability to track tropical storms has significantly improved since the 1960s, with reduced errors in predicted paths and intensities.

However, uncertainty remains inherent in forecasting. Both humanitarian planners and at-risk communities face challenges in interpreting and acting upon forecasts. Therefore, developing strategies that accommodate this uncertainty is essential.

Understanding Anticipatory Action

Defining Anticipatory Action

Anticipatory action involves:

1. Timing: Implementing interventions before a forecasted disaster occurs.
2. Objective: Aiming to prevent or reduce potential impacts rather than responding afterward.
3. Decision Basis: Relying on credible forecasts and risk analysis to inform actions.

It's important to distinguish anticipatory action from preparedness activities. While preparedness focuses on getting ready to respond after a disaster, anticipatory action is proactive, seeking to mitigate potential harm before it happens.

Different Approaches to Anticipatory Action

• Large-Scale Mechanisms: Programs like the UN's Central Emergency Response Fund (CERF) and Start Network's Start Ready provide pre-arranged financing to enable rapid action.
• Community-Led Initiatives: Empowering local communities to use indigenous knowledge and local forecasts to take early action using their resources.
• Integration into Existing Programs: Embedding anticipatory elements into ongoing projects, such as incorporating early warning systems into cash transfer programs for maternal health.

AI's Role in Enhancing Anticipatory Action

Improving Forecast Accuracy and Lead Times

AI technologies are enhancing our ability to analyze vast amounts of data from satellite imagery, weather stations, and historical patterns. This leads to more accurate and timely forecasts, allowing for earlier interventions.

Case Study: Google and GiveDirectly in Nigeria

Google.org, in partnership with GiveDirectly and the International Rescue Committee (IRC), is leveraging AI to predict severe flooding in Nigeria—a country increasingly at risk due to climate change. By analyzing satellite data and local conditions, Google's Flood Hub can forecast floods up to seven days in advance.

How the Program Works

• Predicting Floods: AI models analyze data to identify communities at imminent risk.
• Geographic Targeting: Importantly, the AI supports geographic targeting rather than individual-level targeting, focusing on areas most likely to be affected by flooding. As noted by expert Valentina Barca , this approach is positive because individual-level targeting with AI can be significantly more problematic due to privacy concerns and the potential for bias.
• Sending Cash Transfers: GiveDirectly sends mobile cash payments to vulnerable households within these high-risk areas before the floods occur.
• Empowering Communities: Recipients use the funds to prepare—stockpiling food, reinforcing homes, or evacuating to safer areas.

Positive Outcomes

• Enhanced Preparedness: Early cash transfers enable families to take preventive measures, reducing the disaster's impact.
• Ethical Considerations: By utilizing geographic targeting, the program mitigates risks associated with individual-level data processing, aligning with ethical standards and preserving beneficiary privacy.
• Evidence of Effectiveness: In prior pilot programs, households receiving anticipatory cash showed improved food security and resilience.

Lessons Learned from Pilot Programs

GiveDirectly in their 2023 report Sending Cash to Flood Survivors – 4 Things We Got Right and Wrong. have also shared several critical lessons emerge from their pilots in Nigeria and Mozambique in 2022 and 2023.

Successes

1. Community-Centered Design: Engaging with communities to understand their needs and communication preferences improved program effectiveness. For instance, in Nigeria, using the local dialect Egbura Koto increased trust and accessibility. Field staff who spoke the local language made the program more credible and easier to access.
2. Timely Interventions: Providing cash before disasters allowed recipients to take meaningful actions, such as stockpiling food, harvesting crops early, or reinforcing homes. In Nigeria, recipients could access functioning markets even during floods, demonstrating the importance of timely cash assistance.
3. Ethical Use of AI: The focus on geographic targeting over individual-level targeting is a significant positive aspect. As Valentina Barca pointed out, individual-level targeting using AI can introduce substantial ethical and practical challenges, including privacy concerns and the risk of exacerbating inequalities. By targeting at the geographic level, the program avoids these issues while still effectively reaching those in need.

Challenges

Forecast Uncertainty

• GiveDirectly detail that in Mozambique, GiveDirectly attempted to send anticipatory cash transfers based on AI flood forecasts from Google's Flood Hub during Cyclone Freddy's second landfall.
• Issue: Severe floods did not materialize in the targeted villages, although some nearby areas were affected.
• Implication: This underscores the need to refine prediction models and consider multiple triggers, such as incorporating local data and community insights, to improve the accuracy of forecasts and ensure aid reaches those who need it most

Limited Flexibility

• The same report highlights that limited pre-enrollment hindered GiveDirectly's ability to pivot to areas where severe flooding actually occurred.
• Issue: Due to budget constraints and pre-selected target areas, the program lacked the flexibility to respond to unanticipated disaster locations.
• Implication: Expanding the potential payment area and developing mechanisms for rapid enrollment can enhance responsiveness. This may involve trade-offs between scale and inclusivity, requiring strategic planning and resource allocation.

Data Limitations

• Finally GiveDirectly emphasized the challenges posed by relying solely on available data sources like historical flood data and government records.
• Issue: Such data may not capture real-time changes or local nuances affecting flood risks.
• Implication: Incorporating local knowledge, community input, and additional data sources can improve targeting accuracy. Engaging with local authorities and residents can provide valuable insights that augment AI predictions.

Operationalizing AI-Driven Anticipatory Action

Practical Steps for Humanitarian Organizations

1. Collaborate with Tech Innovators: Partner with tech companies to access advanced AI tools tailored for humanitarian contexts.
2. Invest in Data Infrastructure: Enhance data collection and analysis capabilities to support accurate forecasting.
3. Develop Flexible Funding Mechanisms: Advocate for funding that allows rapid disbursement of aid based on forecast triggers.
4. Build Local Capacity: Train local staff and communities in utilizing early warning systems and implementing anticipatory actions.
5. Incorporate Community Insights: Engage with communities to validate forecasts and ensure interventions are culturally and contextually appropriate.

Addressing Ethical and Practical Considerations

• Data Privacy: Ensure that data used in AI models is collected and stored responsibly, with informed consent from affected populations.
• Equity and Inclusion: Design programs that reach the most vulnerable, including those without access to technology like mobile phones. Consider methods like in-person enrollment and alternative communication channels.
• Transparency: Communicate clearly with communities about how forecasts are made and how decisions are reached. Maintain openness about the limitations and uncertainties inherent in forecasting

The Way Forward

Embracing a Proactive Mindset

Shifting from a reactive to a proactive approach in humanitarian aid can significantly reduce the loss of life and property. Anticipatory action, powered by AI, offers a pathway to more effective and efficient interventions.

Scaling Successes

To maximize impact:

• Expand Collaborative Efforts: Foster partnerships between humanitarian organizations, governments, and tech companies. Share data, resources, and best practices to enhance collective capabilities.
• Advocate for Policy Support: Encourage policies that facilitate anticipatory action, such as flexible funding and supportive regulatory environments.
• Invest in Research and Development: Continue refining AI models and sharing lessons learned to improve future interventions.

Conclusion

The integration of AI into anticipatory action represents a transformative advancement in humanitarian aid. By predicting disasters and enabling early interventions, we can empower communities to prepare and protect themselves, reducing the devastating impacts of climate-related disasters.

While challenges remain—such as forecast uncertainty and operational constraints—the successes to date demonstrate the immense potential of this approach. By working collaboratively and thoughtfully, we can enhance the resilience of vulnerable populations in the face of an increasingly unpredictable climate.

Let's commit to leveraging technology ethically and effectively to make anticipatory action a standard practice in humanitarian aid.

Join the Conversation

I encourage you to share your thoughts and experiences:

• Have you been involved in anticipatory action initiatives using AI?
• What strategies have been effective in your context?
• How can we address the challenges of forecast uncertainty and operational limitations?

Your insights are invaluable as we strive to improve and scale these innovative approaches.

Stay Connected

For those interested in delving deeper into this topic, please join our upcoming AI in Humanitarian Programming discussion group. Together, we can explore solutions and collaborate on advancing anticipatory action in our field.

References

• Peters, A. (2024). Google is Using AI to Predict Floods—and Sending Cash to People Before Disaster Hits. Fast Company.
• GiveDirectly. (2023). Sending Cash to Flood Survivors – 4 Things We Got Right and Wrong.
• Anticipation Hub. (2024). Resources on Anticipatory Action and Cash.

Edited for clarity with the assistance of AI, but the content, thoughts, and arguments are solely from the author.