Poverty reduction and development gains in LMICs have often been painfully incremental or ephemeral. When they have succeeded, they are often “home grown” rather than the product of deductive exogenous planning by elites (Okereke and Agupusi 2019; Sachs 2020; Carr 2014). Analysts using AI to examine socio-economic issues almost universally rely on data that has been produced via deductive means (Usmanova, Aziz, Rakmonov, Osamy 2022; Hamdan, Abu Bakar, and Sani 2020). For example, the Oxford Poverty and Human Development Index (Alkire and Foster 2007, 2011 a, b) uses data from the Demographic Health Survey (DHS), the Multiple Indicators Cluster Survey (MICS) and the World Health Survey (WHS). Deductive data collection approaches such as these employ questions and indicators that are determined a priori—without the participation of the stakeholder community. In this project, we will use AI to examine poverty data that has been collected through inductive means. Inductive means that stakeholder communities are engaged in defining their conditions of poverty, rather than having these assumed by exogenous actors.

Why does this matter? If we are to solve poverty, especially as it exists in low- and middle-income countries, we cannot begin with our own definition of poverty. Rather, we need to understand the needs of people on the ground and the resources they have to produce and optimize their own free markets for local trade, investment, and self-sufficiency. Here, Hayek’s principle of Scientism, which is a critique of economic science in practice, is appropriate. Hayek argues that Scientism is when researchers impute information to obtain objective facts so they can explain phenomena as a general rule. Scientism runs against Hayek’s notion of knowledge in society, where he argues that it is impossible to understand how to secure optimal outcomes without understanding the relative importance only individuals in society know. According to Hayek’s notion of knowledge in society: better data, better allocation of resources, and better outcomes follow when the knowledge of the individual is preserved and not rendered oblique by the state or other centralized organizations. Data collected inductively preserves individual preferences and AI enables them to be described, clustered, and curated in ways that can identify more nuanced needs of people and the market conditions required to support them.

To carry out this study, the project team will use a branch of AI called machine learning (supervised e.g., random forest, and unsupervised learning, e.g., clustering) to identify market opportunities. Their dataset was collected inductively using the Poverty Stoplight Methodology (PSL). The PSL was developed and deployed by Paraguayan social entrepreneur and PhD economist Martín Burt. Since its beginning, the PSL has amassed data from over 200,000 families in 50 different countries. Using machine learning, we will examine the PSL data to determine economic opportunities from the perspective of those who would benefit from the coordination of resources. Then, using ethnographic approaches the project team will “ground truth” our findings and adjust the algorithm as necessary.