Born from the widest ever public consultation by the UN, the 17 sustainable development goals (SDGs) are a chance to ‘press the reset button on development’. Critically, the SDGs encourage countries to improve data analysis and to track progress with common indicators. Many of the goals are inter-related and influence each other.
But with such a wide diversity of categories, the SDGs risk reinforcing the fragmentation of the development field. The body of data about the relationships between goals is too overwhelming for professionals to realistically sort through on their own.
To understand how factors in one area affect those in others will require drawing on sophisticated data analysis that can only be done with the help of new tools.
This January, participants in the “How Can We Use Machine Learning to Achieve the SDGs?” Bangkok Technology Salon shared insights and discussed issues on how machine learning could help facilitate progress towards achieving the SDGs. Dr. Tim France, Managing Director at Inis Communication and one of the creators of SGD Insights, led the conversation.
To begin, Dr. France shared his view on four barriers to achieving the SDGs.
- First is fragmentation. There are 260 distinct international agencies working on some aspect of the SDGs. Dr. France compared this to a car that’s been taken apart with all its pieces lying on the garage floor. Everything is there to make a working, moving automobile, but separated, the pieces are useless. The pieces of the development machine must be put together correctly – and the information silos broken down – if we hope to make progress.
- Next is a failure to think together. The development field is good at debate, but bad at generating ideas together. Professionals concentrate on their jobs, but collaboration to achieve an end is less common.
- Literacy about others’ work is lacking. For example, the WHO doesn’t understand climate data, even if a changing climate is affecting people’s health. Interdisciplinary expertise is rare.
- The filter bubble obstructs cooperation. People are hindered by their capacity to screen information, and a personal strategy is required to break free of the bubble.
One example of the way these barriers affect development efforts concerns the obesity epidemic. Practitioners were witnessing a simultaneous increase in tuberculosis cases, but it took time to make the connection that obesity was leading to more common diabetes, and diabetes was in turn increasing the risk of TB.
But the volume of information required to make these connections is daunting. Only with machine learning can such links become apparent to those who otherwise wouldn’t think to look.
The discussion then moved on to exploring whether machine learning could be helpful at tackling these silos and foster the connecting of dots. Information systems can now be designed to employ algorithms that make connections and present useful, unique information to users.
Participants agreed that the ‘portal solution’, wherein organizations would seek to concentrate all information about certain topic on one website, was outdated and a failure. An example for comparison is how the fashion industry has used machine learning to interpret and set trends. Why can’t a similar approach be applied to development?
However, algorithms have their own pitfalls. ‘Algorithms are only as valuable as their weighting factors’, one discussant noted. Everyone uses a different set of weights – for example, the US prioritizes negative rights, whereas the UN prioritizes positive rights.
Algorithms designed around these value systems would turn out different results. Some participants called for ‘algorithm dials’ which would allow users to give different weights to different systems.
Others focused on the politics of fragmentation and noted the difficulties in changing policies at big development agencies. The structure of the development field is based on intentional choices that governments and other big donors have made in their own interests.
No amount of information availability or silo-destroying will discount this. Some participants expressed frustration that the focus of M&E in the field tends to overemphasize methodology at the expense of consideration of how to use the results.
One interim suggestion was simply to emphasize better digital literacy. ‘Teach people how to search properly.’ While useful information is out there, much of the effort in searching is wasted by those who aren’t aware of how to use free tools.