Inclusive Artificial Intelligence: Putting people at the center of data-driven decisions - 961 KB
Government and social sector leaders often make high-stakes decisions that require prediction, for example, where to target programs to reach those most in need, or which entitlement programs a given household is eligible for. Machine learning (ML) models, a subset of Artificial Intelligence, can make these decisions more efficient. However, for decisions to be accurate, equitable, and unbiased, the models that underpin them must be based on high-quality data that truly represent the people the programs aim to serve.
Over the last five years, IDinsight has been evolving how it uses data science to support partners’ decision-making. For example, IDinsight has worked with a non-governmental partner in India to build, test, and refine models to predict where out-of-school girls are to serve them more cost- effectively. We have provided training data for satellite imagery models predicting crop types and boundaries in India. We have worked with a government partner in Africa as they developed an ML model to predict households’ ability to pay for contributions to a national insurance scheme. In each case, we strive to improve the reliability and equity of partners’ models to inform decisions that directly affect people’s lives and well-being.
For partners interested in using machine learning, IDinsight can support the entire life cycle of model development and deployment – from identifying the problem, developing, testing, and refining the model to evaluating the impact of its implementation (see Figure 1). This note focuses on how IDinsight’s DataDelta and Data Science teams can help collect and use data to create and maintain equitable machine learning models (see yellow triangles in Figure 1).
IDinsight’s DataDelta delivers large-scale, high-quality, representative survey data and insights to government and social sector leaders when they need it, enabling more informed decisions to improve people’s lives. DataDelta uses innovations in statistical methodology, data collection operations, and data science to make it easier and faster to collect high-quality data directly from people, including underrepresented subgroups in the population.
IDinsight’s Data Science team supports social sector leaders to better predict, optimize, or model important program outcomes to improve decisions and amplify social impact. All solutions built for our partners are tested for bias and performance with careful documentation of code such that solutions are maintainable by IDinsight, our partners, and any future collaborators.
Read the full statement to know how DataDelta is putting people at the center of data-driven decisions.
5 December 2024
4 December 2024
3 December 2024
3 May 2024
13 October 2021
14 May 2019
14 July 2022