On May 12th, 2022, Dr. Manuel García-Herranz conducted a compelling presentation on “Algorithms for Humanitarian and Development Work,” giving an internal perspective on the role of Artificial Intelligence, its limitations, and possible solutions. Upon joining UNICEF in 2014, Dr. García-Herranz saw the need and potential for AI and big data practitioner offices within the organization. At the same time, he observed the steady growth in scientific publications mentioning AI in humanitarian work and sustainable development goals, further highlighting the potential and promise of interdisciplinary research on these topics. The promise of AI in this field is its power to automate, optimize and forecast. These are all beneficial for issues tackled by UNICEF, such as epidemics, polarization, disinformation, future of jobs, rapid urbanization as well as climate change.
The presentation tackled the question of the AI talent gap and its development. Available data tends to be geographically clustered and in sectors. Furthermore, its results are fragmented and hide important discrepancies. Manuel García-Herranz used an example of a diagram representing mobile phone users, split into five quintiles by income. High-income countries represented more than half of the data, while lower-income countries constituted about 3 or 4%. This data also does not consider the parts of society who do not own a phone, for example, children. The latter is an essential and vulnerable demographic that should be considered and is especially relevant in humanitarian work.
This leads to his second point: data sets can contain bias given how they are collected, leading to wrong conclusions upon their analysis. Dr. García-Herranz explains how while bias is smaller for high-income countries, its approach of collection results in a high amount of bias within low-income countries.This being said, if collected correctly, specific data sets can be highly useful to organizations like UNICEF. Satellite images and quality in maps can provide real-time data on the location and state of an area in order to allocate resources efficiently and quickly. New scientific methods have emerged to measure poverty, economic development, and the diversity of social networks with machine learning, deep learning as well as combined datasets. This comes with limitations; these models lose accuracy over time as it is difficult to find good data to validate. Therefore, transferring models across countries is laborious, given the little information and transferability. Finally, the average metrics of performance hide other algorithmic inequalities.
Dr. García-Herranz concluded his presentation by explaining that although there are capacity gaps within the development of technology, there has been a simultaneous growth in the attention to Artificial Intelligence, Machine Learning, Sustainable Development Goals, and Humanitarian work. This gap can be bridged with the establishment of a common language. The issue is that there are not enough multidisciplinary ecosystems and focus on strengthening the existing systems, which ensure that AI is fit for humanitarian and development purposes.