ICTworks

Facebook’s recent release of highly accurate population maps for Africa (as well as its impending release of the entire world), and the rising prominence of AI in ICT4D make it a good time to consider the state and future of mobile data collection and the separate, but overlapping field of mobile network data analysis.

We as a community have been in the business of collecting and analyzing digital data for a long time.

• Ken Banks created FrontlineSMS in 2005, the same year the first YouTube video was uploaded.
• Ushahidi was launched a year after the first iPhone in 2008, aiming to organize the power of crowdsourcing.
• This September is the 10-year anniversary of the founding of UN Global Pulse, the United Nations’ entity focused on big data analytics.
• It has also been 10 years since Open Data Kit (ODK) was brought to the University of Washington for further development after a successful launch as a Google.org project.

Meanwhile, Google.org recently announced 20 winners of its AI Impact Challenge and if the agenda for this year’s ICT4D Conference is any indication, our sector is already knee-deep into applying AI to every genre of development.

But what can be said about our level of expertise with data analytics, our ethical standards and the efficiency of our tools?

We have raised and addressed these questions before: Almost every year for the past decade someone has published a guide to mobile data collection, comparing the pros and cons of the various platforms, highlighting best practices. ICTWorks itself published The Definitive Guide to Mobile Data Collection in International Development in November, with a helpful list of other guides. This was of course a big improvement on the 2012 article examining the exact same thing.

Even with these efforts, however, there are several factors that suggest we are entering a new era, one which will require fundamental changes in thinking.

Fast Changing Data Ecosystem with New Players

Obviously, there are more data out there, with more people accessing the internet and other forms of communication in more ways. There are also more niche companies selling data analysis services and more nefarious actors seeking to manipulate information to influence behaviors. There is a trend towards all organizations seeking to hire their own data scientist.

Working with Facebook’s Data for Good team is great, because they are responsive and eager to co-develop applications with you, including on improvements to the interface.

Are more companies going to mobilize their data to support social causes?

If any genre of company is likely to be looking to partner development organizations in the coming years, it is the satellite company, including those offering communications, earth observation and IoT services. “According to Northern Sky Research’s Shagun Sachdeva, the next decade will see the launch of around 8,500 new communications, IoT and earth observation satellites, a large portion of which will be constellations like SpaceX, OneWeb and Amazon.”

The development sector is also working hard to organize itself around more effective and responsible use of data.

• The Digital Square collective just announced the Global Goods
• NYU’s GovLab has set up the Data Collaboratives project, to support “a new form of public-private partnership, in which data held by private-sector entities is leveraged in a responsible and ethical manner to provide public value.”
• Others have been promoting the related idea of civic trusts which would “[own]the underlying code and data generated by a technology and licenses it to a for-profit company that commercializes it.”
• Considerations for Using Data Responsibly at USAID was just released last month.

But perhaps the most exciting aspect of an ecosystem growing in size and complexity is the increasingly common activity of pursuing meaningful correlations amongst diverse data types.

Rapid Change in Data Privacy Landscape

The privacy regulatory landscape is a quickly moving target. According to research by Sidley Austin, Angola, Benin, Burkina Faso, Cape Verde, Chad, Cote d’Ivoire, Equatorial Guinea, Gabon, Ghana, Lesotho, Madagascar, Mali, Mauritius, Morocco, Seychelles, South Africa and Uganda all have privacy and data protection laws. And approximately 50% of these were enacted within the past five years.

GDPR is serving as the inspiration for new data privacy laws from California to Nigeria, though the need for new data privacy laws has long been apparent. As the 2019 GSMA’s Mobile Economy: 2019 puts it, “Countless rules are still in place that were devised for a mobile industry that was fundamentally different to how it is today — different technologies, different competitive dynamics, different market realities. The world has changed, and regulation needs to advance with the times.”

In last year’s “On the privacy-conscientious use of mobile phone data”, essential reading for anyone interested in carefully moving beyond wet blanket warnings about potential dangers of data collection and use, 21 of the world’s leading experts in mobile data collection lament that, “The limits of the historical de-identification framework to adequately balance risks and benefits in the use of mobile phone data are a major hindrance to their use by researchers, development practitioners, humanitarian workers, and companies” and propose four models for accessing and using the data in responsible ways that avoid people being identified.

A recent article in Nature questioned the usefulness of Call Detail Records (CDR) analysis, though it didn’t identify any new issues these experts themselves haven’t already been raising for years.

CDR analytical capabilities are indeed more developed than systems allowing individuals to know how their data are being used, permission its use or to get paid for it. Depending on the country, the sale of call records might not violate an MNO’s terms of service or the law, but still privacy advocates raise concerns. (It would be interesting to know if these people were Gmail users.)

We Need to do Better Data Analysis

While everyone is jumping on the AI bandwagon, organizations focused on collecting and aggregating national health data are still avidly developing and promoting the use of the comparatively simply DHIS2.

The increase in data availability and the ability to correlate data pose the question of what counts as an acceptable minimum of complexity. A recent “demonstration” project funded by DIAL and implemented by Cooper/Smith in Malawi sought to use mobile network data to build a picture of population density, enabling planning of where future health facilities should be located.

In 2019 are two different colored dots on maps enough? Should the study have also analyzed mobile money transfers to identify potential variations in clinic size and services offered? Used remote sensing to determine which clinics would be inaccessible during the rainy season?

To generate data for its DFS Use Among Kenyans report, Caribou Digital paid monthly stipends to people who agreed to have a data monitoring app installed on their phones. The “aggregate data [that] show[ed] patterns in how people use their devices for communications, entertainment, and financial transactions.” The high quality of the report’s insights make you wonder if you should go the same route.

We Should Get Out of Platform Business

What is the most valuable skill an ICT4D professional brings to their profession? The ability to code? Write algorithms? Many ICT4D professionals believe their preeminent contribution is that of interpretation. You understand at a profound level how a technology will affect a situation and work to ensure the best outcome.

You might argue that for years many organizations have confused the ends with the means, such as those who believed they should spend tremendous time and resources developing a new data collection system, because none of the available platforms had the needed functionalities, or those who believed that building a new data platform would lead to comprehensive contributions and profound coordination.

After all this time and experience, it seems we should recognize that the skills gap widens with the introduction of each new big technology. It also seems worth acknowledging that organizations often expect their data people to manage the technological aspects of data collection, implement data quality reviews, do the actual analysis and to make sure that the organization is following data responsibility guidance. That’s a lot on one person’s shoulders.

Imagine the question of the best mobile data collection platform were settled ten years ago, and that we had spent our time figuring our thornier data responsibility questions. Given that the application of AI in international development is still relatively new, we still have time to use what we have learned from our experiences with mobile data collection and mobile network data analysis to focus on activities where we bring unique value.

Need for Self-Education and Coordination

If you are struggling to follow developments in data privacy and wrap your head around the increasingly complex implications of evolving tools and functionalities, you are not alone. Though, don’t let this console you: Pursuing a deeper understanding of these tools might not be optional if you want a long career in ICT4D. As the technologies get more complex, so must your skills of interpretation.

You can start by reading the research of Data and Society, Yves-Alexandre de Montjoye, Oxford Internet Institute, The Engine Room, The Mozilla Foundation, The New York Times Privacy Project and others. As you read more, pay particular attention to events and groups of people seeking to collaborate, because your understanding will deepen more quickly through conversation and engagement.

Ideally this will lead to tighter, focused collaboration across groups. And, in the name of collaboration around responsible and effective data use, check out this new draft Wikipedia page on international development digital data types, which is being inaugurated with this blog. (It is pending review by Wikipedia editors, but visible to the public.) Remember that Wikipedia pages can be edited by anyone.

This list should grow.

The point of sharing every conceivable data type of potential relevance to development is that it might help quicken our effective use of AI and give us a clarifying, granular picture of the data we should care about.