What does the term ‘machine learning’ bring to your mind? You may think about product recommendations, stock market predictions, automated text replies or facial recognition. But its connection with social good and international development may not be as pronounced. That said, machine learning has tremendous potential to solve a multitude of developmental challenges faced by governments and nonprofits in the global south. We see four key dimensions where machine learning can provide timely and actionable data and insights to help solve pressing social problems.
Any kind of natural, economic, or political disaster accrues huge costs to human civilization, more so in developing nations where social structures are more fragile. If such events could be predicted in advance, these costs, both human and economic, can be significantly reduced. Forecasting can help us anticipate floods, violent conflict, the spread of an epidemic, and crop failure. This knowledge can be incredibly useful for social institutions to prepare for a timely, efficient, and effective response.
Effective targeting of policy interventions – such as vaccination drives, cash transfers, food distribution – requires accurate, up-to-date and granular data on the socio-economic status of communities. Such data is rarely available, especially for the most vulnerable populations. Furthermore, collecting this data through surveys can be prohibitively expensive for both governments and other social sector entities. Modern machine learning tools can help extract this data – for instance, machine learning models can be used on satellite imagery to predict poverty levels in communities, and to target villages for unconditional cash transfers or for siting solar-powered microgrids.
Another common policy problem faced by developing nations is tracking / monitoring of populations, interventions, or resources, primarily because of the high costs associated with frequent and extensive data collection efforts. In recent years, we have seen machine learning being used to monitor land cover changes, to keep track of higher education quality in colleges and universities, and to observe democracy and economic growth over decades.
Every professional working in the development space knows the importance of identifying causal relationships between policies and outcomes. Over the last couple of decades, randomized control trials (RCTs) have become widely popular in the developing world and are considered to be the gold standard for identifying causal relationships. However, RCTs have their own disadvantages – they lack external validity, are costly, time consuming, and challenging to implement. Causal inference and machine learning have largely stayed disconnected, but in recent years, they are beginning to come together. Researchers are beginning to harness the power of machine learning to make causal claims about important policy problems, such as the impact of air pollution control policies on PM2.5 levels and the social and economic impacts of climate change.
Machine learning tools are most effective when combined with a thorough knowledge of the target context and the end user’s needs – whether it is a farmer aiming for a plentiful crop yield, or an NGO promoting solar power use in a remote region. Governments and non-profit organizations are best placed to leverage these advancements in machine learning for tackling challenges in global development.