Are machine learning technologies ready to be used for humanitarian work and development?

TL;DR

It is argued that, without organized and purposeful efforts, these new technologies risk at best falling short of promised goals, at worst they can increase inequality, amplify discrimination, and infringe upon human rights.

Abstract

Novel digital data sources and tools like machine learning (ML) and artificial intelligence (AI) have the potential to revolutionize data about development and can contribute to monitoring and mitigating humanitarian problems. The potential of applying novel technologies to solving some of humanity's most pressing issues has garnered interest outside the traditional disciplines studying and working on international development. Today, scientific communities in fields like Computational Social Science, Network Science, Complex Systems, Human Computer Interaction, Machine Learning, and the broader AI field are increasingly starting to pay attention to these pressing issues. However, are sophisticated data driven tools ready to be used for solving real-world problems with imperfect data and of staggering complexity? We outline the current state-of-the-art and identify barriers, which need to be surmounted in order for data-driven technologies to become useful in humanitarian and development contexts. We argue that, without organized and purposeful efforts, these new technologies risk at best falling short of promised goals, at worst they can increase inequality, amplify discrimination, and infringe upon human rights.

Figures (2)

• Figure 1: New technologies and tools.A, The percentage of people owning a phone is rapidly rising worldwideinternational2018measuring. B, Satellites are increasingly able to capture our world in greater detail. Data shows the maximal aperture resolution per year for all civilian and commercially launched satellitesbelward2015launched. C, The explosive growth of interest in the humanitarian field as reflected in number of papers per year; data from Google Scholar retrieved using the query ("artificial intelligence" OR "machine learning") AND ("sustainable development" OR "humanitarian action" OR "sdgs" OR "millennium development")
• Figure 2: Common pitfalls of data-driven tools.A, Models are not easily transferred between contexts, illustrated here by the correlation between the human development index (HDI) and observed mobility from social media. The correlation is calculated between HDI and the entropy $S$ of mobility data, where $S_i=-\sum_{j} p_{ij} \log{p_{ij}}$ and $p_{ij}$ is the probability of observing trips between regions $i$ and $j$. B, When models are validated on digital data from later periods, in this case mobile phone data from Iraq, we observe that model performance drastically decays over time. In this case we train a ML model to estimate poverty from mobile phone data using one month to train the model and evaluate how the model performs on data collected 1-5 months later. Model performance is measured using a cross-validated correlation coefficient. Here we show the relative model performance, which is rescaled with respect to the initial accuracy, to quantify how well the model performs for later months. Light blue area denotes the variation across multiple models trained using different cross-validation sets.