A FOSS Dilemma in Governments

Discussion
1

I was in a conversation with a UNFPA person who wants to build a Machine Learning model for demography predictions for Odisha. (Hi Census!)

There’s one research firm that developed one such model over the last 6-8 years. It’s not an open-source model. So they want to license their model to UNFPA in helping them build what they want to build.

But UNFPA prefers open source models because it helps in final transfer to the state government. Eventually the model has to be with the government and they’d use it for policy making. So they ideally want to build an ML model from scratch in 4 months :confused: and open-source it.

Questions raised during the conversation for which I couldn’t get a clear answer:

  1. The government doesn’t have the capacity to run and update the ML model even if they get access to a good open source model. Would it be another open-source ML model that would fall to disuse? Is it not better if people good at this do their work and government uses their work by paying a fee?
  2. How can incentives for the researchers involved be managed here within FOSS license regimes?
  3. What other licensing mechanisms can be explored to solve this dilemma?
  4. Any good case studies on similar issues?
2

What is the use case for using ML in census?

3

So the UNFPA wants to predict demographics of the state at a hyperlocal level and use it for policy planning.

Name of the project: Demographic and Data Intelligence’ by harnessing Geospatial Technology and AI for informed Policy and Planning

I couldn’t find the link to the RFP as it was closed recently.

Basically they want to use satellite data (buildings data, building heights data, land use data) to predict population dispersion. They also want to predict the age, gender profiles by using other datasets as available. (Civil registries etc)

It is not a usecase for Census. Their idea is loosely that governments need not depend on Census for demographics and we can build ML models for it.