00:02 Hi everyone,
00:03 we're gonna get started.
00:04 We're really excited to have Sieta Minahova,
00:06 uh,
00:06 give
00:08 an overview talk.
00:09 So,
00:09 you know,
00:09 kind of the state of play of mobile data
00:12 and what its role is in global development research,
00:16 um,
00:16 and,
00:17 uh,
00:17 yeah,
00:18 so super excited to have Svita.
00:24 All right.
00:25 Hi,
00:26 everyone.
00:26 Um,
00:27 hope everyone had a nice lunch.
00:29 Um,
00:29 I'm really excited to be here today and to have a chance to give the state of play on
00:34 mobile phone data in development research.
00:36 And
00:37 I guess we'll wait a minute to get the slides up.
00:40 They come a little closer oh
00:42 yeah.
00:45 Mhm.
00:46 Yeah,
00:48 yeah,
00:48 yeah.
00:50 Uh.
00:53 All right,
00:54 perfect.
00:57 All right,
00:57 so,
00:57 uh,
00:57 as we've already heard today,
00:59 um,
00:59 there's a huge amount of data that is available worldwide,
01:03 um,
01:03 especially generated in the last two years,
01:05 and much of it coming from the over 8 billion mobile phone subscriptions worldwide.
01:11 This provides a really important opportunity both
01:13 for research and for policy again,
01:15 as we've seen through the presentations today.
01:18 Now,
01:18 in terms of types of data,
01:20 we've seen a lot of examples.
01:21 Um,
01:22 we have data generated through call detailed records or metadata.
01:26 So those are the observations that every single
01:28 phone generates when you make a call,
01:30 receive a text message.
01:31 We also have data coming from smartphones,
01:33 from mobile app content,
01:35 as well as from mobile surveys.
01:37 Now focusing on the passively collected data,
01:40 so GPS and the mobile phone metadata,
01:43 the GPS data can provide us more accurate spatial granularity.
01:46 We saw some really great visualizations in the last session,
01:50 um,
01:50 and also it can be easier to access through
01:53 global companies like Facebook or through companies like Google.
01:57 We're able to get data across a number of countries,
02:01 but we also need to recognize the limitations and where we are today.
02:04 And the situation is that smartphones are still um.
02:07 Not widely used
02:09 in low income settings and so if we're using this data,
02:11 it's really only capturing the very high
02:14 income populations in some of these developing countries
02:17 and um while on the other hand,
02:19 when we use mobile phone metadata that's able
02:22 to capture information from any type of phone.
02:25 And that will also include smartphones
02:27 and so we're able to capture a much larger portion of the population.
02:31 So when we're thinking about setting up policies that affect the entire country,
02:35 it's really important to think about data that can allow us to
02:38 understand what is happening for the,
02:41 for as much of the population as possible in that country.
02:44 And for that reason,
02:45 I'll be focusing today on specifically mobile phone metadata.
02:48 I could go on for hours about all the different types of data.
02:51 I'm really excited about all those sources,
02:53 um,
02:54 but I'll keep it to,
02:55 to this one specific type,
02:57 and I'll start out giving a very brief overview of some of the different uses,
03:00 and I'll spend most of the time talking about
03:03 what are the challenges that we have,
03:05 uh,
03:05 in terms of being able to actually mainstream the
03:08 use of this data in policy and research.
03:12 So what we've already seen again a lot of examples today they're really exciting.
03:15 We're gonna continue to see more examples in the afternoon,
03:18 but to give you a brief overview,
03:20 we've seen this type of uh work and and data used in economic development,
03:24 developing national statistics like poverty
03:27 or social protection.
03:28 We've also seen this in financial services,
03:31 um,
03:32 through the development of CRIS,
03:34 uh,
03:34 of
03:35 uh credit risk models,
03:36 um.
03:37 Uh,
03:38 one of the first uses I would say is,
03:40 um,
03:40 in the use of health and looking at spread of disease such as malaria,
03:44 and also in disaster planning,
03:46 urban planning.
03:47 So,
03:47 in the next session,
03:48 we're gonna be talking more about transport,
03:50 and also a lot of other new uses like food security,
03:53 um,
03:54 looking at crime,
03:55 real estate,
03:56 so a lot of exciting ways and opportunities to apply this data.
04:00 Now I won't go into details because
04:02 again we've been seeing that throughout the sessions,
04:04 but what I want to really highlight
04:06 is this transition that has happened,
04:08 going from academia
04:10 into the public policy sphere.
04:13 So when this work,
04:14 this type of work first started back in 2008,
04:17 1 of the first papers
04:19 was looking at human mobility in a very theoretical way.
04:23 As we move into 2012,
04:25 we now start to see this data applied to development pilots,
04:29 especially through the data for development challenges that
04:32 were done through Orange in 2012 in Cote d'Ivoire
04:35 and then 2014 in Senegal,
04:38 a much larger.
04:38 Larger um research community was exposed to this type of data
04:42 and could apply to different challenges
04:44 so that's actually where I first got involved was in 2014 when I had um the ability to
04:49 um work with the data from Senegal to look at spread of malaria
04:54 in the country and so it was a really
04:55 exciting opportunity to bring on a lot of researchers
04:59 and to get people exposed to working with this data.
05:01 So I'll definitely say those sorts of um opportunities are fantastic
05:05 to increase the community that is working with the data.
05:09 We then move um further into into kind of
05:12 a larger application that was happening in real time,
05:15 which was the Ebola crisis.
05:16 And when Ebola hit,
05:17 there was a commentary out saying we can use CDR data,
05:20 we can use this type of new data to understand mobility
05:24 and to help us actually decrease the spread of Ebola.
05:27 In West Africa
05:29 unfortunately we were not there yet so there
05:32 was some historical data used at the time
05:34 but really most of the analysis happened after the the crisis and um
05:39 afterwards in in 2018 and so it was really much more
05:42 retrospect uh uh retrospective.
05:46 And now,
05:46 um,
05:47 in 2020 with the COVID-19 pandemic
05:50 we've seen that the applications of this data in real time so
05:53 it's really exciting to see this transformation going from the very theoretical
05:57 to actually having these data in countries like Ghana,
06:00 The Gambia,
06:01 all across Europe
06:03 that allow us to see in real time
06:05 what is happening in terms of mobility,
06:07 how is the disease spreading,
06:09 and also how are the policies that are being implemented working.
06:14 But I,
06:15 the,
06:15 the next step,
06:16 so now we've gone from,
06:17 from academia to these really successful public policy examples.
06:20 The next step is to see how do we mainstream this.
06:24 So how do we take this from having
06:26 examples in one country and another country to
06:28 actually being able to work with this data
06:31 across the whole world.
06:33 And so that's where I'll talk about some of the challenges that I face and,
06:36 and the teams that I've worked with have faced.
06:38 Um,
06:38 I,
06:39 I'll say that these are not all of the challenges,
06:41 but hopefully they give you a bit of a sense of some of the different
06:44 areas we need to be thinking about if we really want to mainstream this data.
06:50 So the first is around data access and,
06:52 and this has actually come up a little bit today in some of the questions.
06:56 So,
06:56 during the COVID-19 pandemic,
06:58 a group of us here at the bank across a
06:59 number of departments came together to see if we can
07:03 access this data around the world in order to be
07:05 able to support our colleagues with policy um implications around
07:09 uh mobility.
07:10 And so we looked into accessing this data.
07:12 In over 40 countries
07:14 we use a lot of different models,
07:16 so looking to reach out to
07:18 mobile network operators,
07:19 looking,
07:20 reaching out to regulators,
07:21 reaching out to third party organizations,
07:23 academics that already had access
07:25 and trying to see how can we actually
07:27 put this data to action right away.
07:30 And one of the first things we realized is that just going to the supplier,
07:34 so that the providers of the data is not enough.
07:36 It's really a large ecosystem where you have the suppliers,
07:40 so that's the mobile network operator,
07:42 where you need to have the technical capacity,
07:44 you potentially need to have the funding.
07:46 And the legal agreements in place,
07:48 but you also have the demand side,
07:49 so you need the government to actually want to use this data to want to apply this data
07:54 um in order for them to provide you with the enabling environment
07:57 and for that you need a lot of technical awareness and capacity.
08:01 And finally then there's a data translator so that's where we
08:04 were coming in and a lot of other organizations around the world
08:07 were coming in to see how they can help go from the raw data that that's at the MNO
08:12 to the actual policy analysis and implications that the government can use.
08:18 Now,
08:18 in our case,
08:19 as I mentioned,
08:19 we started out with these 41
08:21 um country cases,
08:23 and we slowly see all of the different challenges that came along the way.
08:26 So we went from those 41 cases down to 37.
08:29 It's 4 of those we decided were politically too
08:32 sensitive to actually try to pursue this data.
08:34 It was,
08:34 it was just too dangerous to
08:36 um try to work with this data in certain settings.
08:39 From those 37,
08:40 we then found that in 21 cases,
08:42 we were not able to get permission from the MNO to work with their data.
08:46 But then even in the 16 cases where the MNO said yes go ahead here's our data take it
08:51 we actually were not able to in 6 of
08:54 those cases because we still needed the government permissions.
08:57 So just because the company
08:59 is potentially willing to share the data,
09:01 um,
09:01 that doesn't necessarily mean that we can access it
09:03 because in many countries there might be regulation in
09:06 place or even the MNO might want to make sure that the government is on board before.
09:10 They actually will share the data
09:12 and so in the end there were 9 successful cases where we were able to access the data.
09:17 Now that's still quite a success given that usually even just
09:21 doing one of these types of cases can take years.
09:23 Um,
09:24 so I would still say that
09:25 that we've made quite a lot of progress in the space,
09:28 but I think this just illustrates
09:29 the amount of time and effort and all of
09:31 the various challenges that come along the way.
09:34 And one final item uh I'll mention is with the um cost as well.
09:37 So there was one case where cost was specifically prohibitive,
09:40 but even the 21 cases um where the MNO was not able to provide us the data,
09:46 potentially cost may be an underlying factor where maybe if we had
09:50 funding or there was a business model,
09:51 they may have been more willing to actually share the data.
09:55 And so there were a number of broad obstacles
09:57 that we were able to highlight through this example.
09:59 So I think there's a lot of learning that also came
10:01 out of this that I hope can be applied more broadly,
10:04 and I'll just highlight a couple of areas
10:06 which are the need for investment in skills,
10:09 technology,
10:09 legal frameworks,
10:10 and trust.
10:11 So you really need trust between
10:13 all of those stakeholders that are part of the ecosystem.
10:16 And moving into the trust piece,
10:18 one of the key pieces is on the data privacy.
10:21 In every single conversation that we had,
10:23 whether with an MNO,
10:24 with the government,
10:26 or anyone,
10:26 data privacy was one of the first things that always came up.
10:30 And we've seen this more recently in the media that
10:33 um data privacy has been um has been brought up in a number of settings so
10:38 John Oliver,
10:39 a comedian,
10:39 and a few weeks ago was in the press as
10:42 um he wound up uh being able to access data that
10:47 is technically anonymous but could very much be
10:49 linked potentially to unique special individuals in Congress,
10:53 um,
10:53 and so there was a risk that that.
10:55 That could reveal sensitive information about these people
10:58 and this is just one example,
11:00 but I think it's um one of the big
11:02 risks when working with this type of mobile phone data
11:05 and with call detail records,
11:07 especially where
11:08 even if you anonymize um the data in terms of removing
11:12 identifiers,
11:13 removing the phone number,
11:14 there's still a very high risk that you
11:16 can identify unique people from their mobility patterns.
11:20 And so there's a number of models that have been developed to try to address this,
11:24 um,
11:24 for example,
11:25 limited release where you're only releasing parts of the data and you potentially
11:29 manipulate it to make it less likely that you can identify individuals.
11:33 You can work with pre-computed indicators,
11:36 um,
11:36 and so then only aggregated indicators,
11:39 for example,
11:39 the total number of calls an individual makes
11:41 or the total number of trips between locations
11:45 is shared.
11:45 You could have remote access,
11:47 so where you don't actually remove the data from the telecom company.
11:52 Instead you can VPN in
11:53 and work with the data there and that's important because
11:56 just transferring the data can lead to data breaches,
11:59 so we need to be careful of that.
12:01 And then also a question answer model
12:02 which is similar to the pre-computed indicators,
12:04 basically
12:05 you have a query say,
12:06 I would like to know mobility from A to B,
12:08 and then that query comes out.
12:11 Additionally,
12:11 especially,
12:12 um,
12:13 in,
12:13 um,
12:14 Josh Blumenstock's lab and,
12:15 and,
12:15 um,
12:16 in that team they're working quite a lot on looking
12:18 at differential privacy which I think is an exciting area,
12:21 um,
12:22 in terms of being able to work with this data in a way that is
12:26 safe and that allows more researchers to have access
12:29 and of course best practices like limiting the access to.
12:33 A small number of individuals,
12:35 um,
12:35 also limiting the amount of data that we use because
12:38 as researchers we always want more data,
12:40 at least I always am,
12:41 I'm looking for more data,
12:42 but I think when working with sensitive data,
12:44 it is important to be much more careful and,
12:46 and recognize
12:47 what are the variables we actually need for the analysis we want
12:51 and try to limit the,
12:52 the data to,
12:53 to those necessary variables.
12:55 Data storage processing,
12:57 everything should be happening
12:58 in secure environments,
13:00 and we could think about,
13:01 especially if this is something that's kind of done broadly at a government level,
13:05 are there ways in which we could actually obtain consent?
13:08 So the way that a lot of GPS,
13:09 um,
13:10 data
13:11 will ask for consent in order for the GPS data to be shared or,
13:13 or companies that use GPS data will ask for consent,
13:16 we could think about whether such a model may be possible even with CDR data.
13:22 And so again,
13:23 uh,
13:23 some things I'd like to highlight with privacy is that it's
13:26 really important to invest in the infrastructure and the capacity,
13:30 especially for things like remote access.
13:32 It's really important to be able to have that technology
13:35 at the MNO and to have the capacity within the
13:37 MNO to create indicators that then can be shared,
13:40 as well as also thinking about regulation more broadly
13:43 that can be standardized across countries.
13:48 A third area that I think is quite important is around data quality.
13:51 So when we work with these types of data,
13:53 we need to remember
13:55 they're not made for researchers,
13:57 they're for business purposes.
13:58 They're made for billing,
14:00 and so oftentimes there can be issues with the data and
14:02 we can't actually necessarily just take it at face value that.
14:05 Um,
14:06 we can interpret everything as being behavior.
14:09 Um,
14:09 I particularly want to highlight three areas that I've
14:12 seen basically in every data set I've worked with.
14:15 One is issues with completeness,
14:17 so sometimes,
14:18 especially if data is being transferred,
14:19 literally entire days of data might just go missing.
14:22 Or you might have a location where for some reason there's no data.
14:26 And if you don't notice that,
14:27 if you don't flag that,
14:28 if you don't account for that,
14:30 then you may mistakenly wind up aggregating the data
14:33 and producing analytics that you share with the government
14:36 that then actually are not what is happening on the ground.
14:39 Very similarly to that,
14:40 you also need to be watching out for anomalies.
14:43 So,
14:43 the completeness is a little bit easier to see,
14:45 so you can see right away in the figure that there's an entire day of data missing.
14:49 But you can also oftentimes have situations where a
14:51 tower goes down for a couple of hours,
14:53 so there's still some data,
14:54 but it's going to be much lower.
14:56 And what's that,
14:57 what that's going to mean is that
14:58 in that in that specific tower there's going to be
15:01 fewer observations and then maybe in the neighboring
15:03 tower there's going to be more observations.
15:05 So again,
15:05 if you don't notice that this is actually a data.
15:08 Anomaly in terms of how the data is generated,
15:10 then you might assume that means there's fewer people
15:13 in one area and maybe more movement into another area
15:16 and if that interpretation is then communicated with the government,
15:19 then they might wind up taking policy action based on something that's coming from
15:23 the data and how it's generated and not actually what is happening on the ground.
15:27 And finally,
15:29 something that that's really important is using other
15:31 data sources as well to check for consistency.
15:34 So we have
15:35 so much survey data and other types of data that that is
15:38 available for us to be able to cross-check and to make sure that
15:41 what we're seeing makes sense.
15:43 Um,
15:43 we can also crosscheck within the data.
15:46 So are the number of users making sense
15:47 compared to the number of observations per day.
15:50 All of these types of things need to be integrated,
15:53 and we need to invest in the capacity and standard
15:56 code to be able to do these sorts of real-time checks
15:59 so that the policymakers can trust the
16:01 data and the analytics that they're receiving
16:04 and can take action knowing that
16:06 what is coming in is something that that is actually indicative
16:09 of the behaviors and what is happening on the ground.
16:13 And then finally,
16:14 a 4th area I want to highlight is around bias in the representativeness of the data.
16:18 Again,
16:19 this probably comes up just as often as privacy does in these conversations.
16:24 And as,
16:24 as we know,
16:25 and,
16:25 and we've touched upon their important limitations in terms of who is covered.
16:29 So,
16:29 obviously,
16:30 anyone without a mobile phone is not going to be in this data.
16:33 And so if we're thinking about national policies,
16:36 we need to be careful in terms of who is potentially excluded.
16:39 From those policies if they're only based on
16:42 using mobile phone data,
16:43 even within those users that are in the data,
16:45 there might be differences between
16:48 different socioeconomic groups or men and women
16:50 in terms of how they use mobile phones
16:52 and so we need to be cognizant of that.
16:55 There could be differences in terms of geography,
16:57 so
16:58 some uh areas might have more mobile phone towers than others,
17:01 and that's going to affect them,
17:03 um,
17:03 the measures that we're able to get in those different areas.
17:06 And then of course,
17:07 if we're only working with one provider,
17:09 then that's going to limit
17:11 what we're able to say.
17:12 And if that provider,
17:13 for example,
17:13 tends to be used by more higher income individuals
17:16 or they have coverage only in certain geographic areas,
17:20 then that links back to some of these other issues as well.
17:23 In addition to
17:24 the sort of representativeness,
17:25 which I think is kind of the first thing that people think about,
17:28 another really important area that I also think is,
17:30 is really interesting to study is around
17:33 the changes in the user base.
17:35 So
17:35 oftentimes the policies and events that we want to study
17:39 will not only affect the behaviors we're interested in.
17:42 So for example,
17:43 we're interested in seeing how mobility changes during COVID-19.
17:46 But it can also affect the data generating process.
17:49 What I mean by that is that how people use
17:52 their phones might be a function of what is happening.
17:57 So one example,
17:59 one example of that is
18:01 looking at the number of trips taken early on the pandemic in one country,
18:06 we see that there is an increase over time from January to March.
18:11 But then when we plot the number of observations per subscriber,
18:15 we see a very,
18:16 very similar pattern.
18:17 We see how correlated movement is with the number of observations per subscriber.
18:22 So what that means is that
18:24 if an individual is now making more phone calls,
18:27 then we're more likely to see them in more locations.
18:29 And so that,
18:30 so as we see more movement happening,
18:32 the question is,
18:33 is that actually more movement indicative of people taking more trips,
18:37 or are we just seeing them in more locations because now they're making.
18:40 Making more phone calls
18:42 and if we can't distinguish between the two,
18:44 then it's very hard to know
18:45 what is actually happening and what that should mean for policymakers
18:50 and this is actually quite common because we
18:52 know that the way people use their phones
18:54 is affected by what is happening.
18:55 Um,
18:56 there was a presentation earlier that was
18:58 actually about that in terms of the number of contacts that people have
19:01 might change because of what is happening on the ground
19:04 and so we need to be thinking about that and thinking about
19:06 how we can factor that in when working with these data.
19:11 And,
19:11 and finally,
19:12 uh,
19:12 another point related to that is that as we start to use the data for policy,
19:17 it can actually lead to incentives for people to change their behavior.
19:21 So if we're in a situation where someone,
19:24 uh,
19:24 where we're developing an algorithm
19:26 that will then decide if someone receives a cash transfer that's based on.
19:30 Calls and networks and
19:32 we say that someone with fewer calls and fewer network contacts means
19:35 that they're lower income and so then they're going to receive a transfer
19:38 well then that person is going to make less calls
19:42 and is going to call less people if they figure
19:44 out that those are the types of indicators that are
19:46 being used to decide if they get a cash transfer.
19:50 There's a lot of really interesting work in the space,
19:52 especially
19:53 new research by,
19:54 uh,
19:54 Bjork,
19:54 Gern Bloom and Stock and Knight
19:56 that looks at this and actually develops,
19:58 um,
19:58 algorithms that can be,
20:00 um,
20:01 that can be robust to this type of manipulation.
20:03 And so
20:04 it's really important again as we start to use this data more for policy
20:08 to be thinking about
20:10 these sorts of changes in behavior that might happen because of the
20:13 actions that we're taking in terms of how we're using the data.
20:17 And so in terms of data limitations,
20:19 I think the main areas we need to invest is first in phone
20:23 access and so this has been brought up a few times today.
20:25 I think that's where there's an opportunity for
20:28 um for projects for governments to invest in
20:31 having more people actually be part of the
20:33 network and providing them with mobile phones.
20:35 And we also need additional research into potential biases
20:38 and it it's um likely that those biases will actually
20:43 potentially evolve as we start to use the data more for
20:45 policy so we need continuous research that is trying to understand
20:49 what is happening on the ground and and how things are evolving.
20:54 So where are we now?
20:55 This is just um uh kind of a quick map looking at the countries that
21:00 um the GSMA has highlighted had uh wound up using
21:04 this type of data during the COVID-19 uh pandemic.
21:08 This isn't inclusive of all countries,
21:10 but it gives you a sense,
21:11 and we see that
21:12 there's a lot of usage in Europe,
21:14 and it really did get kind of.
21:16 Mainstreamed um
21:17 in Europe but very much not so in Africa where I would say
21:21 there's actually a lot more potential because there's so much less data available
21:26 to understand what is happening on the ground and so
21:29 while there's been
21:30 a lot of success during the COVID-19 pandemic,
21:33 I would say we're still far from actually
21:35 mainstreaming the use of this data more broadly.
21:38 And some of the things that have helped again during
21:40 the pandemic is especially focusing on developing open source tools,
21:45 um,
21:46 things like the mobility analytics task force task force work here at the bank,
21:51 um,
21:52 where we've been working to develop open algorithms
21:54 and trainings that can be available for.
21:56 For
21:57 anyone to,
21:58 to use in their own context,
22:00 um,
22:00 a lot of work by Flowminder for,
22:02 for years,
22:03 especially the development of Flow Kit which can allow for
22:06 MNOs to process this data on their own servers.
22:10 The use of open software like Docker and Spark
22:12 can make it possible for
22:15 individuals around the world to be able to use these tools
22:17 and to be able to work with this type of data.
22:20 And all this can help to facilitate access by providing
22:23 the uh the needed algorithms to to do this work.
22:26 It can help to address privacy,
22:28 especially
22:29 if those algorithms and those tool kits are made so that
22:32 the data can be analyzed at the mobile network operator.
22:35 It can help standardize the production of the indicators
22:39 and also ensure quality.
22:40 If we embed
22:42 quality checks in all these algorithms and these tool kits
22:45 and there's a large number of organizations that are working in this space,
22:49 um,
22:50 I've just listed a few,
22:52 um,
22:52 but there's there's a community that's been pushing this
22:54 work and and trying to work together and I think
22:57 that we're in a really great spot to kind of keep pushing that forward.
23:02 So then where do we go?
23:04 Well,
23:04 I think there's definitely an opportunity for us.
23:08 And again,
23:08 we've seen this really important transition
23:11 happen where we've gone from the theoretical
23:14 to actual applications on the right here,
23:16 this is actually a farmer that uh was a beneficiary of the program in Togo that
23:20 Emily talked about this morning
23:22 and,
23:22 and so it's really exciting to see that we're now
23:24 actually putting this data to practice in nationwide programs.
23:29 Nevertheless,
23:30 more still needs to be done to mainstream mobile data,
23:33 and I'm not going to do another call to action.
23:36 Uh,
23:36 we've had many of those,
23:38 um,
23:38 and we've actually seen a lot of action,
23:40 so we've seen a lot of work happening during the pandemic,
23:43 and I think what is needed now
23:45 is that we actually have investment
23:48 in sustainable,
23:49 uh,
23:50 work with this data and,
23:51 um,
23:52 in sustainable work to enable the environment.
23:54 To make this happen in the types of um data access policies legislation regulation
24:00 that is needed to do this on a wide scale
24:03 as well as a lot of investment in the
24:05 capacity and training of government officials of um MNO operators
24:10 of others so everyone that's part of this ecosystem
24:14 is needed to uh in order for this to move forward as as we saw earlier.
24:19 And in addition,
24:20 I think we need to continue with the high quality research we're seeing here today.
24:24 As we've seen this work
24:26 does require quite a bit of technical expertise.
24:28 We need to be careful also in terms of
24:31 the analysis that we're generating and,
24:32 and how those are being applied to policy.
24:35 And so as we hopefully do move into mainstreaming this data,
24:39 I think it's.
24:39 It's also important to remember that policymakers will need to continue working
24:43 with researchers
24:44 so that we can trust the data that we're producing.
24:47 We can then also evaluate the results of the applications of those data and we can
24:52 see how we can develop more effective systems that make use of this type of data.
24:57 So I'll pause there
24:58 if you have any questions.
25:05 I don't know if there's time for questions.
25:09 All right.
25:10 True.
25:17 There's any questions in
25:22 Yeah.
25:25 Think people can just go to the mic.
25:30 Thank you so much for this um wonderful presentation
25:34 um so I think you touched upon really important issues of data privacy
25:40 and
25:41 you know you are making an agreement or
25:44 you know creating consensus between government and.
25:47 And the organizations like World Bank,
25:50 uh,
25:50 for
25:51 broader public use
25:53 of,
25:54 um,
25:54 phone data,
25:55 which is difficult to access.
25:57 But
25:57 even,
25:59 you know,
25:59 getting
26:00 permission from government
26:01 and,
26:02 you know,
26:02 the
26:04 network carriers
26:06 does that represent
26:07 citizens' agreement
26:09 to use their data.
26:11 Just like in the past,
26:13 there was a Kenyan government,
26:15 you know,
26:15 spying on.
26:16 Uh,
26:18 So there's are issues about,
26:19 you know,
26:19 whether foreign government or research institute
26:23 like Harvard Public Health,
26:25 um,
26:25 spying on
26:26 malaria data
26:28 on Kenyan citizens
26:29 and those like
26:31 data sharing agreement
26:33 maybe was only between government and
26:36 You know,
26:37 and the
26:38 university,
26:38 top university in the US.
26:40 So how
26:41 can we,
26:42 you know,
26:42 address this kind of issue?
26:46 Maybe I'll take a couple of questions and then I'll
26:48 address them.
26:48 Yeah.
26:49 Thank you,
26:49 Ms.
26:50 uh,
26:50 Millicheva,
26:50 for your talk.
26:51 Uh,
26:51 I'm wondering if you can talk about the,
26:53 the point where you left off in terms of mainstreaming opportunities.
26:57 How has the bank started,
26:58 uh,
26:59 considering the use of mobile data and operations
27:01 in terms of assisting governments through very pragmatic
27:04 applications with these data sets?
27:08 Great questions.
27:09 Um,
27:09 so starting with the first one on
27:11 privacy,
27:12 um,
27:13 I,
27:13 I mean I,
27:13 I do think that that's an important area where again we need to be.
27:18 Thinking a lot about how do we define
27:21 individual privacy,
27:22 how do we think about
27:23 group privacy,
27:25 um,
27:25 so again in,
27:26 in some of these cases
27:28 if the data can be applied in ways where,
27:31 um,
27:31 we're not actually looking at any individuals,
27:33 we're not spying on any individuals,
27:35 we're looking at group behavior
27:37 similar to,
27:38 um,
27:39 being able to look at,
27:40 um,
27:41 patterns in migration and in census data.
27:44 But I,
27:45 there is a point about,
27:46 you know,
27:46 whether the individual gives their consent for their
27:49 data to even be included in that,
27:51 um,
27:51 in that aggregate work and that's why I
27:52 also suggested that we could think about models
27:55 where people could opt into having their data used,
27:58 um,
27:59 how feasible that is,
28:01 uh,
28:01 is a question,
28:02 um,
28:03 but I think it's,
28:05 it's something where
28:06 there's a lot of different opinions,
28:08 um,
28:09 I've,
28:09 I've talked to him.
28:11 People sort of on on all
28:12 sides of the spectrum in terms of what is considered private data,
28:16 how you even define
28:18 um whether the data is um
28:22 I guess whether it's an individual or not or
28:24 or whether it's anonymized enough that that it's OK to use um so
28:29 I think the community has not reached a consensus and I
28:32 think a lot more discussion will be needed on that.
28:35 In terms of how the bank is moving forward with using this data,
28:38 I think one of the um big areas right now is the social protection space,
28:43 um,
28:44 so work similar to what Emily has,
28:46 uh,
28:46 Emily showed us today in Togo,
28:48 I think it's being um applied in in other settings in other countries,
28:52 and I think there's a lot of exploration in terms of how we can.
28:55 Use this,
28:56 um,
28:56 use this type of data to think about targeting to identify
29:00 poor uh economic areas in a country,
29:02 for example,
29:03 or individuals,
29:05 um,
29:05 and so I think that's one of the areas where there's been a big push within the bank,
29:09 but hopefully there will be a lot more,
29:11 so that's,
29:12 I think,
29:12 you know,
29:12 the.
29:13 The area that that I'm also trying to push and
29:15 it's an area where we need to work across departments,
29:18 so working with the digital development uh global practice in the bank
29:21 with the research um teams working with all the
29:24 different global practices where this data can be applied
29:27 and so it will take quite a lot of effort to move that forward.
29:34 All right.
- add-style
- lp-body-content