AI in the Humanitarian Sector

 Benefits, Challenges, and the Path Forward

By Leila Toplic, Lead for Emerging Technologies Initiative, NetHope

What are the benefits of using AI in humanitarian contexts? What are the challenges? And what do we need to consider if we seek to incorporate more AI into the NGO sector?

Today, close to 80 million people are displaced due to conflicts and persecution. As many as 143 million people could soon be displaced due to climate change, which is disproportionately impacting already resource-constrained regions such as sub-Saharan Africa. At the same time, inequality is growing sharply across multiple dimensions including education, gender, and economic development. 258 million children were out of school before Covid-19 and an additional 463 million were cut off from education during the pandemic. These are just some of the humanitarian challenges that we’re facing.

Our world is facing ever more intense and protracted humanitarian crises, and as a result the global community is pressed to find new ways to help people and communities in need. Artificial intelligence (AI) is one of the tools that has potential to help us tackle some of the toughest humanitarian challenges.

AI systems – with their capacity to learn, to predict, and in some instances, to make decisions based on those predictions and take specific actions – are fundamentally changing the world around us and improving how we live and work. So, it’s no surprise that AI has become a hot topic in the humanitarian sector, with many discussions about its benefits, risks, and appropriate use cases as humanitarian organizations look to incorporate AI in humanitarian programming.

As a technology consortium of the world’s leading NGOs, we at NetHope see the potential in AI to support every aspect of our work. We believe that the humanitarian sector has a responsibility to the people and communities we support to explore and apply AI in a responsible, impactful, and sustainable way.

Since 2017, we’ve been bringing together global NGOs and technology experts from the private sector and academia to share, learn, and collaborate on all aspects of AI application in the humanitarian sector. Recently, we convened a group of leading humanitarian NGOs – members of our AI Working Group – for a round of consultations focused on exploring the benefits, challenges, and the path forward for AI in the humanitarian sector. I had the opportunity to speak with organizations that are focused on disaster response, the needs of children and women, protecting refugees, and promoting respect for human rights, including Danish Refugee Council, International Rescue Committee, Mercy Corps, Norwegian Refugee Council, Catholic Relief Services, SOS Children’s Villages, and Humanitarian OpenStreetMaps Team.

This research was supported by Microsoft’s AI for Humanitarian Action Initiative and will be used to inform the new call-for-proposals that Microsoft issued on September 24, 2020, with project proposals due by October 31, 2020.

As you consider this and other opportunities for AI in the humanitarian sector, I wanted to share what we know about the benefits and challenges, and several considerations for the way forward.

Benefits

We know from early practical implementations of AI/machine learning (ML) in the humanitarian sector that there is a whole set of problems that AI/ML, along with other tools, could help us solve.

AI systems can help NGOs:

• Reach more people with services and information they need – such as education, legal and health information.
• Free up limited human resources to focus on high-priority work.
• Make decisions and act faster in emergencies through real-time awareness of the situation.
• Predict emergencies before they spread and escalate through early detection and warning.
• Improve outcomes through real-time feedback on the effectiveness of programs and provide recommendations for improvements.

The sector is taking the first step with AI by exploring a whole range of programs and experimenting with different AI/ML capabilities. Some examples of early practical implementations include:

Danish Refugee Council is using AI/ML to forecast forced displacement in places like Burkina Faso, Mali, Niger and Nigeria in west Africa. The Foresight tool uses open data from sources including UNHCR, the World Bank and the NGO agencies to predict forced displacement in a given country.

International Rescue Committee is using AI/ML in a number of projects including for optimizing service delivery to refugees, for predictive modeling of conflicts and crises, to facilitate jobs-matching for refugees, and for individualized learning experiences for children affected by crisis.

Norwegian Refugee Council’s chatbot assists Venezuelan migrants in Colombia with learning their rights according to current immigration policies and laws.

The Carter Center is using AI to get more accurate and timely analysis on the Syrian conflict.

It’s important to note that while the opportunity for AI to help in the humanitarian sector is vast, many of the AI/ML initiatives are still in the exploration or piloting stage and they are not yet delivering significant impact on a sustained basis. This is due to some of the challenges outlined below.

Challenges

AI for humanitarian response shares some of the same challenges as other technologies that are being used in the humanitarian sector. They include barriers related to data, sustainability, inclusion, funding, and oversight. However, the barriers to adopting AI/ML remain higher in comparison to other technologies due the fact that many of the NGOs are just not far along on their digital transformation journey.

Barriers to overcome include:

• Awareness – awareness of AI in the humanitarian sector is still nascent. While some NGOs are beginning to hire technical experts and build capacity internally, the awareness of ‘what AI is good for’ across their organizations – especially, among the program staff – remains low. This is why NetHope has been focusing on building capacity in the humanitarian sector to evaluate the suitability of AI/ML for their organizations and to design and deploy AI/ML solutions in a responsible, sustainable, and impactful way.
• Expertise –  most nonprofits don’t have the expertise to develop and maintain AI solutions and they have to bring in outside expertise. Challenges related to bringing in outside experts include: (1) Nonprofits are competing for technical experts with the commercial sector. (2) Onboarding in-kind support (eg volunteers from tech companies) can be challenging because of cultural differences and timelines. (3) Nonprofit managers don’t always know how to manage new technical experts and provide technical oversight.
• Data – data issues are one of the most common barriers to AI adoption in the humanitarian sector. Issues range from the lack of required, representative data, problem statement, and context, to data collection, preparation, and analysis being done manually. NGOs highlighted that data wrangling and preparation is time and resource-intensive, often cost-prohibitive and impacting their ability to scale projects to achieve a greater impact.
• Funding – finding donors that want to invest in new, potentially risky and unproven solutions is a challenge. Adoption of new technologies like AI requires a new approach to funding, including: (1) Funding that not only funds development of new AI solutions but also funds data, infrastructure, and technical experts. (2) Allowing time for exploration and iteration rather than asking for clearly defined solutions from the outset. (3) Not expecting immediate and sizable impact in the short term.
• AI Ethics/responsible innovation – there’s a whole set of ethical issues related to AI that have emerged over the past few years, and nonprofits are eager to develop capacity to (1) evaluate AI/ML for ethical risks (ie scan for consequences, determine if AI is the right solution considering the risks); and (2) know how to mitigate those risks by operationalizing a set of ethical values and principles (such as fairness) across all touchpoints – people, process, partnerships, and technology.
• Tools and Services – considering the challenges related to technical expertise, there is a need for more AI/ML tools and services that don’t require specialized expertise – no-code or low-code tools such as Power Virtual Agents tool from Microsoft.
• Connecting insight to action, scaling, and sustaining – NGOs struggle with connecting insights to action due to the lack of awareness and understanding of AI across their organizations, some of which operate in a federated model. Additionally, sustaining and scaling of the solutions is a challenge because most of the early practical implementations have come out of cross-sector collaborations: NGOs working together with tech companies and/or academic institutions who have AI/ML technical expertise and can help with the initial model development and data preparation. Transitioning from the pilot phase (where the model is developed and data is prepared by a partner) to scaling and sustaining (where NGOs own re-training of the model and refreshing the data) is challenging as it requires new resources (eg technical experts, funding for tools and services) and internal processes.

Path forward

So, what's the path forward for AI in the humanitarian sector? Humanitarian organizations have an obligation to bring the best, most appropriate tools to support their work in disaster response, protecting and supporting refugees, and promoting respect for human rights. As advancements in AI accelerate and AI gets embedded all around us, humanitarian organizations need to understand the best uses for AI/ML as well as the potential risks, and know how to determine when it’s appropriate to use this powerful technology.

To get started with AI in humanitarian contexts, we recommend you:

1. Start small and focus on the problems that would benefit the most from AI.
2. Resource appropriately for all stages (concepting, MVP, scaling, sustainability) and implement change-management to integrate the AI solutions into existing organizational processes.

PROBLEM

It’s no surprise that the problem areas in the humanitarian community that would benefit the most from AI are mostly focused on improving existing programs and processes rather than creating solutions that would not be possible without AI/ML. Specifically, NGOs are looking to:

1. Free up staff time to focus on other high-priority work. AI can help augment staff capacity through automation of repetitive tasks.
2. Extend reach of humanitarian services and support. AI (eg chatbots) can help provide self-help options for people and communities in need.
3. Improve effectiveness of humanitarian interventions. AI can help us optimize and enhance humanitarian interventions through prediction based on the analysis of large amounts of data.

There are several reasons why Augmentation, Self-Help and Prediction are some of the most promising use cases for AI at this time, including:

• Growing number of people-in-need exceeds human capacity and traditional resources.
• Limited resources and capacity to respond to a growing number of emergencies and protracted crises are driving NGOs to get better at anticipating/predicting.
• Starting ‘small’ – with an existing program – enables NGOs to learn how to develop and implement AI solutions effectively and prove the value of AI/ML internally and to donors.
• NGOs are eager to put to use the data that has already been collected and start learning from the existing data.
• Most NGOs lack expertise, resources, and data to do anything more complex.

When evaluating suitability of new AI project concepts in the humanitarian sector, we recommend asking the following questions:

• Is the problem the right fit for AI/ML?
• Are there any ethical issues, especially those where the risks of using AI outweigh the benefits? Are humanitarian responders and affected individuals and communities ‘in-the-loop’ (included in the decisions related to the design and implementation of AI systems)?
• Does the organization have, or can they access, required data, preferably in the digital format?
• Is there a historical baseline to compare the effectiveness of the new solution?
• Can the organization connect insights to action?
• Does the organization have resources and management buy-in for maintenance and scaling of new AI solutions?

For additional questions to ask when determining the suitability of AI for your programs and organizations, please review NetHope’s AI Suitability Framework.

RESOURCES AND SUPPORT

In order for NGOs to overcome the challenges outlined above and turn their ideas into sustainable, responsible AI solutions – they need internal (leadership) commitment to digital transformation and external support in the form of:

1. Capacity building – programs that help NGOs (both IT and program staff) learn about what AI is good for and how it can apply to their work.
2. Cash – for staff time, acquisition of data, acquisition of additional infrastructure that might be needed for the development or implementation of an AI project.
3. Technical expertise – for development, testing, iteration, and initial implementation of an AI project.
4. Product – in the form of cash or credit. Product pricing (eg. cost per chatbot conversation) and business model that are adapted for the nonprofit funding model.

If you are interested in learning more about AI in the humanitarian sector, we invite you to join us at the NetHope Virtual Global Summit later this month where we’ll have a number of sessions focused on lessons learned from practical implementations of AI in the sector, demos of no-code and low-code AI tools, training on how to get started with developing chatbots, and an AI ethics workshop.

Special thanks to Alix Cabrall (IRC), Giulio Coppy (NRC), Grant Gordon (IRC), Priyanka Jagtap (CRS), Alexander Kjaerum (DRC), Brett Koblinger (SOS Children’s Villages), Neal Moffitt (IRC), Bo Percival (HOT), Ognen Plavevski (CRS), Mary Rochelleau (Mercy Corps), Ahmed Shoukry (CRS), and Ric Shreves (Mercy Corps) for their participation in the consultations.