Thinking Machines to build Southeast Asia AI Research Bank for Development (AI4D)
The AI4D Research Bank will support organizations in strategically using open data and machine learning to address the 17 UN Sustainable Development Goals
Thinking Machines will be developing the AI4D (Artificial Intelligence for Development) Research Bank to accelerate the development and adoption of effective machine learning (ML) models for development across Southeast Asia. With support from the UNICEF Venture Fund, we are building technology to support scientists and social sector organizations to use AI to enable evidence-based programs toward attaining the 17 UN Sustainable Development Goals.
Development sector organizations experience difficulty in accessing timely and robust data insights on vulnerability, such as areas with air quality issues, places with no access to reliable internet connectivity, and communities most at risk in climate disasters.
Machine learning researchers have been deploying AI estimation models to augment hard-to-acquire ground truth data, given these common data limitations:
- Data of interest not existing at all or is often outdated
- Data being available, but not in useful formats, difficult to analyze with other datasets, or highly aggregated and not geographically specific
- Data latency extremely high due to the manual collection process, which delays data analysis and reporting
In our machine learning work, we’ve discovered that access to unconventional useful datasets, having to continually reconstruct models from descriptions in academic papers, and the difficulty of processing large scale geospatial data are critical blockers in the development of effective and accurate machine learning models.
By increasing access to data and technical resources for ML training, we can reduce critical blockers in using AI for development.
The AI4D Research Bank will promote open source, open data, and open science by providing a platform for data transformation pipelines on common useful datasets like the DHS, code repositories for geospatial feature engineering, and technical documentation, so data scientists can build their own poverty estimators. For programme staff, the AI4D Research bank will have an intuitive website for exploring ML-generated development measures for them to develop an understanding of the use and limitations of these ML-generated datasets in their day-to-day work.
With UNICEF’s support, we are developing the following:
- Open source poverty estimation models for nine countries in Southeast Asia
- Open source haze detection model trained on Mapillary street images, satellite images, and meteorological data
- Open source geospatial feature processing pipelines for Ookla, OpenStreetMap, and Google Earth Engine satellite images
- Expanding the existing Geomancer open source geospatial feature engineering library
- Bringing in the open source and scientific communities via events
By open sourcing our solutions, we enable more organizations to build easily replicable and cost efficient machine learning models, while encouraging the uptake of free and unconventional data sources in the social sector to augment ground truth data.
About Thinking Machines
Thinking Machines Data Science is a data consultancy building AI and data platforms to solve high-impact problems. They are based in Manila, Philippines, with offices in Thailand and Singapore. As part of the Innovation Fund portfolio, Thinking Machines originally developed a poverty estimation machine learning model for the Philippines. They’ve continued to build on this work, developing geospatial AI models for measuring infrastructure development and economic growth across Southeast Asia where they work with governments, civil society, telecommunication companies, and the private sector to turn those insights into action.