Using Data to Fight COVID-19 - And Build Back Better Emmanuel Letouzé Maria Antonia Bravo - Vodafone Institute
←
→
Page content transcription
If your browser does not render page correctly, please read the page content below
Emmanuel Letouzé Maria Antonia Bravo Nuria Oliver Natalie Shoup Using Data to Fight COVID-19 And Build Back Better October 2020 Paper Series No. 2
Content
06 18
03
2. Four sets of
considerations
Introduction and concerns 2
raised by the use
of digital data
07
Executive and technologies
Summary for COVID-19
response efforts
About this paper
1. COVID-19
digital initiatives
T 27
his paper is the second in a Paper
Series published jointly by Data-Pop
Alliance and the Vodafone Institute
for Society and Communications, following
“Sharing Is Caring: Four Requirements for
Four Key Requirements for Sustainable 3. Key recommen-
Private Data Sharing and Use for Public dations for a fairer
Good, published in November 2019.11 post- COVID-19
world
It was written by Emmanuel Letouzé,
Nuria Oliver, María Antonia Bravo and
Natalie Shoup. It benefited from comments
and inputs from Inger Paus (Managing
Director, Vodafone Institute for Society and
Communications), Pedro Rente Lourenco
32
(Lead Researcher and Data Scientist,
Vodafone Group Big Data & AI), Matthew
Allison (Senior Public Policy Manager,
Data, Platforms & AI, Vodafone Group),
Julia Ebert (Senior Research Manager,
Vodafone Institute for Society and Com-
munications) and Juan Camilo Mejía (Pro- Concluding
gram and Research Manager, Data-Pop remarks
Alliance).
The report was copy-edited by Carola
Miras and Juan Camilo Mejía.
1 https://www.vodafone-institut.de/wp-content/uploads/2019/11/
DPA_VFI-Sharing-1.pdfExecutive
summary 3
S
ince the start of the COVID-19 pan- that the pandemic is more of a syndemic,
demic in the first quarter of 2020, which refers to a health issue that clusters
numerous governments and public along social lines.3 With COVID-19, it is as
institutions around the globe have devel- if the veil of feigned ignorance about the
oped or promoted initiatives leveraging features, drivers and effects of injustices,
digital data and technology in support of such as the indecent growth in income
response efforts. Some have sought to concentration and inequality, the differen-
identify and predict hotspots, others to tial impacts of environmental degradation
evaluate the effectiveness of containment and pollution, systemic racism and sexism,
policies or to detect and trace the close and even the risks posed to democracy
contacts of infected individuals. As many around the globe, including those fueled
countries are still struggling with the first by digital data and technology, had been
wave of the disease and several are fearing shredded in a few months.
or already grappling with a second—often
in tumultuous socio-political contexts—it is In this context, digital data and tech-
essential to take stock of the key features nology serve as lenses on the world and
and expected benefits of major initiatives as levers of change, for good or bad.
and summarize the main debates and A decade into the “data revolution” and
questions they have raised—about their with a decade left to make progress
usefulness, implications, limitations, risks towards the Sustainable Development
and requirements in the fight against the Goals (SDGs), the current crisis provides
pandemic, and beyond. a unique opportunity to ask how digital
data and technologies can truly and struc-
The pandemic has also laid bare long- turally improve our world by both fighting
standing and deeply rooted structural fault the pandemic and “building back better”.
lines in our world. Far from hurting everyone It is evident that reliable and timely data
indiscriminately, the virus and its socioeco- are of paramount importance to fight the
nomic effects have affected disproportion- pandemic. Yet, they are of no use if they
ately poor people, people of color, women, are concealed or manipulated for and
people with disabilities, migrants, and peo- by officials interested in scoring political
ples governed by populists. points, drowned in an ocean of dubious
claims and rumors, or not effectively com-
Some, such as the outgoing UN Special municated and understood.
Rapporteur on Extreme Poverty and
2 https://amp-theguardian-com.cdn.ampproject.org/c/s/amp.
Human Rights, even consider that COVID- theguardian.com/global-development/2020/jul/11/covid-
19 has “revealed a pandemic of poverty 19-has-revealed-a-pre-existing-pandemic-of-poverty-that-be-
nefits-the-rich
that benefits the rich”.2 Others have argued 3 https://doi.org/10.1016/S0140-6736(20)32000-6Similarly, it is clear that more advanced Social media companies and social plat-
initiatives leveraging digital data and forms have a duty to the public to provide
technology that are at the core of this safeguards from theories that weaken
paper—such as contact tracing applica- trust in their governments and in science.
tions or hotspot detection algorithms— Beyond citizens, the COVID-19 pandemic
can and must play a role in fighting the has brought to light the evident lack of
pandemic. But these digital “solutions” data and digital literacy among many
are not, as the saying goes, “silver bullets” public officials and decision makers, with
that will solve our human-made problems potentially devastating consequences.
by themselves. We are once again expe- Education and the long-term collabora-
riencing the very real risk of jumping to tion of a diverse set of experts in relevant
“technological solutionism” without under- areas—such as data science, epidemi-
standing and addressing the key impli- ology, anthropology, computer science,
cations—technological and scientific, immunology, public health, economy
political, economic, ethical—of new data and sociology—with public administra-
and technology. tors must be ensured to assist in more
evidence and knowledge-driven decision
Fundamentally, this crisis ought to be making. These collaboration of a diverse
a moment in our lifetimes when we reas- set of experts need to analyze the incen-
sess our ways of life, our incentives, our tives and constraints of participants and
priorities, and push for real change with work together to accomplish beneficial
some of the most powerful tools avail- outcomes for all parties.
able: data and technology. We should
use this crisis as a testbed and catalizer A third one is evidently high-quality
for how data and technology could help data: to fuel better human systems to both
us set and achieve humanistic societal fight the pandemic and build back better,
objectives, as underpinned by the SDGs data are one of the most powerful tools at
and other frameworks—and not just serve our disposal. Data must be allowed to be
the interests of surveillance agencies and shared and analyzed in privacy-preserving,
large corporations. This paper there- interoperable manners. Decision makers 4
fore explores how data can help fight and citizens should be both informed and
COVID-19 and how COVID-19 also pro- involved in what data are being collected
vides an opportunity to better use data and how; what they represent; how and
to build back better. why they are stored and potentially shared
in raw or transformed forms. Data regu-
lators and controllers have a key role to
To realise this play in ensuring appropriate safeguards
with regards to privacy,
vision, four consent and inclusion
of data subjects, and to Social media companies
elements appear help navigate the trade-
offs between emer- and social platforms have
to be key: gency situations and
long-term conditions. a duty to the public to
One is context: we need to have a
thorough understanding of the goals, A fourth one is com- provide safeguards from
implications and the impact on citizens munication and trust:
and society of decisions in the longer a privacy-sensitive soci- theories that weaken trust
term (from a science/technology, eco- ety requires transpar-
nomic/commercial, social, political, legal ency and confidence in their governments and
and ethical point of view). It is also import- in the use of the data
ant to understand the different technol- collected. Honesty and in science.
ogies being designed and used for real transparency are key to
response, as well as the parameters and building trust, in addition
risks, benefits, limitations and impact to competence (i.e. efficiently carrying out
of each. Furthermore, it is crucial to be the task at hand) and reliability (i.e. compe-
mindful of the fact that not all responses tence sustained over time). The current sit-
can or must be digital, and that not all uation has been enlightening for different
people will be able to access digital solu- stakeholders, showing that even though
tions. This means that solutions have to be data could be the solution to some reali-
thought in a holistic way so that everyone ties, there are many different groups that
is included. are inevitably less connected and there-
fore not accounted for. This reality means
Another is education: citizens should that data and technology may have con-
be provided with clear, precise, under- tributed to spreading—just as much as
standable information. Huge amounts of to curbing—the pandemic, and this fact
dis- and mis- information are being pro- must be acknowledged, communicated
duced about and around the pandemic, and addressed. What can and cannot be
which makes it difficult for the non-expert achieved by these technologies must be
to discern the differences between facts, communicated transparently so that citi-
hoaxes and everything in between, which zens and societies can effectively use and
feed on and fuel political polarisation. accept them when they are deployed.4
Once these various digital technologies Develop “data literate” human and
are fully understood, it is important to crit- data systems. A major challenge
ically interrogate the wider implications and objective over the coming years
beyond immediate pandemic response. will be to actively strengthen “data liter-
Such implications should lead to a set of acy” among both governmental agencies
guiding principles impacting how each and citizens—defined as “the desire and
of them is designed, developed and ability to constructively engage in society
deployed. through and about data”.5 This will mean
building data skills and culture through
capacity building support in order to
With this in mind, base discussions and decisions on facts.
Building a data culture and systems of
we put forth interoperability is also key yet it is missing:
it should work across distributed networks
six main and systems thereby ensuring usability
between different apps within or across
recommendations: different countries.
1 5
Think and act boldly and decisively Test and scale sustainable busi-
—now. This may be a once-in-our- ness models. Now is also a good
lifetime opportunity for deep, ambitious time to think broadly and boldly
and long-term thinking, especially to fight about sustainable business models for
deep-rooted inequalities and excesses private-public data sharing and use.
fueled by complacency and greed that Today’s data boom and raised visibility of
have been exposed and exacerbated by digital solutions are great incentives for the
the pandemic. Now is the time to design, private sector to allocate more resources
deploy, test and scale digital data and into data sharing for the public interest, to
technology approaches to enable long- formalize public-private-people partner-
term positive social transformation. ships (PPPP) and develop and test “free-
5
2
mium models” that would ensure financial
Only deploy data and technology sustainability. At a European level, research
that are fit for purpose. Despite funding should be devoted to foster PPPP
its promise, technology is no silver Data4Good research consortia within the
bullet. Its strengths and limitations should next EU Horizon 2027 program.
6
be acknowledged. How to balance dig-
ital and non-digital technology solutions Consider and use regulation as
is of paramount importance. Furthermore, an enabler. Regulations must sup-
technological solutions should be thought port enabling principles such as
of as enablers, integrated with existing (1) encouraging data sharing through
structures when they perform well, such voluntary, market-driven mechanisms; (2)
as public health systems. They should sharing only under legally compliant, eth-
also have clearly stated rationale and ical and socially acceptable scenarios, in
purpose and be systematically evalu- line with the principles of trustworthiness
ated. Given the already staggering digital and privacy-by-design; (3) data for good
divide, omnipresent structural inequities initiatives should be subject to fair remu-
and biases, we need inclusive solutions neration, thereby creating the conditions
so that large segments of the population for products and services; (4) technol-
are not left by the wayside. ogies should be fit for purpose and with
3
a human(ity)-centric perspective. Let us
Place people at the center and “in not forget that that technological break-
the loop” at all times. Privacy and throughs throughout history are often pre-
human rights should be core consid- cipitated by a crisis, and then adapted
erations. Simulations of unintended con- and reused elsewhere, both for good and
sequences from ethical and human rights bad. Good harmonization of regulation is
perspectives should be performed and key to ensure that initiatives can be scaled
potential risks minimized before imple- up quickly, as appropriate, and sustained
menting and deploying any technology. over time.
Social and behavioral responses to dig-
ital technology interventions need to be The COVID-19 pandemic—or syn-
anticipated and embedded in the design demic—presents a historic opportunity
of tools and apps. This requires large- for all parts of societies—the private and
scale public consultations, digital public public sectors in collaboration—to organ-
spaces such as “online parks”4, critical ise themselves and collectively build back
governance and accountability mecha- better following a human-centric approach
nisms, on line portals and local forums to to, and use of, digital data and technology.
ensure that citizens are informed and can Let us not miss it.
actively participate in outcomes.
Questions and comments about this
paper can be sent to eletouze@datapop
alliance.org.
4 https://www.wired.com/story/to-mend-a-broken-internet-create-
online-parks/ 5 Data-Pop Alliance Data Literacy White Paper, 2015Introduction 6
O
ver the past few months, many gov- to optimise this significant potential for of this crisis safeguard fundamental rights
ernments and public institutions change. A key question posed is how dig- and promote a renewed human-centric
around the globe have developed ital data and technologies can truly and vision rather than a techno-solutionist
or deployed initiatives leveraging digi- structurally improve our world by both approach that may enhance the very con-
tal technologies and privately held data fighting the pandemic and “building back ditions that contributed to the magnitude
in support of COVID-19 response efforts. better”, i.e. not satisfying ourselves with of the pandemic’s impact, such as struc-
Some resources aim to identify potential returning to business as usual, but rather tural inequalities. As we unpack these
hotspots or demonstrate the effectiveness capitalizing on this dramatic event and questions in a dire and urgent context, it
of containment policies, while others seek allowing novel, ambitious projects and is essential not to lose sight of the trade-
to trace infected individuals’ close con- ideas not only to emerge but also to garner offs and risks that putting our trust in tech-
tacts, amongst others. The usefulness and public support—pending the development nologies may entail, and how these tools
implications of these initiatives—notably of a vaccine or other effective treatment. could be leveraged to improve tomorrow’s
but not only contact tracing applications— world.
have been widely debated. Meanwhile, What makes infectious diseases
many countries are still struggling with the unique is that they thrive on human inter- With these points in mind, this paper is
first wave and several are in the midst of a action. In doing so they serve as a litmus structured as follows:
second—often in tumultuous socio-political test, revealing how societies function,
contexts. rendering visible the world’s inner work- Section I describes initiatives that use
ings and flaws. Thus, while data and tech- digital technologies and privately held data
Concomitantly, structural fault lines nology are seen as increasingly relevant as part of pandemic response strategies,
around the world have been laid bare in all for pandemic response strategies, crises unpacking how these initiatives work and
available data: far from hurting everyone offer an opportunity to step back, examine providing examples from several regions.
indiscriminately, the COVID-19 crisis and and hopefully improve our current systems Section II summarizes key questions and
its effects have disproportionately affected and societies. While there is no doubt that concerns these initiatives have raised
people governed by populists, the poor, using data has significant potential for across four main domains: technological
people of color, women, persons with dis- fighting COVID-19, challenges and ques- and scientific, commercial and economic,
abilities, migrants and children. Some con- tions about the requirements and long- ethical and legal, and political spheres.7
sider that COVID-19 has also “revealed term applicability of digital technologies Section III discusses recommendations to
a pandemic of poverty that benefits the must be identified and addressed. meet the challenges of today and tomor-
rich”.6 row by leveraging data and tech in the
Assessing the effectiveness, security, fight against COVID-19 and potentially
This historical context provides a privacy, ethical and trust implications of other pandemics, as well as the scourge
unique, perhaps once-in-a lifetime, oppor- these digital responses to the crisis is of global poverty and inequality.
tunity to reconsider our life styles and indispensable to combat the epidemic
and overcome it rapidly. However, it 7 Based on the taxonomy proposed in the publication Sharing
is Caring: Four Requirements for Sustainable Private Data
6 https://amp-theguardian-com.cdn.ampproject.org/c/s/amp. is equally essential to ensure that the Sharing and Use for Public Good co-developed and published
theguardian.com/global-development/2020/jul/11/covid-
19-has-revealed-a-pre-existing-pandemic-of-poverty-that-be-
longer-term impacts of the models, proto- by Data-Pop Alliance and the Vodafone Institute in November
2019. See https://www.vodafone-institut.de/studies/four-key-re-
nefits-the-rich cols and applications created in the midst quirements-for-sustainable-private-data-sharing/1.
COVID-19
7
digital initiatives
T
he way that data and technologies smartphones has also emerged not only
are leveraged and positioned in the as a promising and rich source providing
COVID-19 response presents a real the public with information about the virus,
opportunity for greater visibility, collabo- but also as a critical source of information
ration and evidence of impact for digital for decision makers and authorities.
solutions. However, it also harbors sig-
nificant risks given the speed with which The COVID-19 context has also opened
governments and companies are obliged discussions on using data collected by
to react and make decisions about data additional technologies such as smart-
use, privacy, oversight and accountabil- phone apps (e.g. Facebook, Google
ity in developing and implementing these Maps), search engines (e.g. Google
solutions. The scope of data considered in searches) or social media platforms (e.g.
this paper centers mainly on that gener- Twitter feeds), facial recognition sys-
ated and/or enabled by interactions with, tems, satellite and surveillance devices,
or between, mobile phones (both feature bank and credit card transactions, public
phones and smartphones): a highly sensi- transportation systems, electronic health
tive issue with almost all publics. records and funeral homes to aid gov-
ernments in their responses to contain
Digital technologies based on the the spread of the virus. In this context,
analysis of large-scale human behavioral expectations as to what technologies and
data are being touted for their prospec- applications can really do to enable better
tive usefulness to combat the pandemic. responses and policies remain high. How-
Given the ubiquity of cell phones, mobile ever, given that these applications fre-
phone network data has been one of the quently rely on collecting, sharing, storing
first sources of privately-held data that and analyzing personal, and often quite
many countries—both developed and sensitive data, it is critical to assess the
developing—have turned to in COVID-19 possible unintended consequences that
response efforts. Moreover, data captur- may arise from sharing and using such
ing the interactions with, and between, data.A. Types and Box 1. Location and proximity
taxonomies of data: how mobile devices can
privately held be used to infer your position8
data sources GPS: Mobile devices can determine their location using the
global positioning system (GPS) through the device’s GPS chip
The functions and promise of many of which receives signals from satellites orbiting the earth. Accuracy
the mobile phone applications evoked of GPS signals is variable and tends to be less so in urban areas or
above for combating the spread of COVID- indoors. GPS signals are detected primarily through the device’s
19 are grounded in their ability to make use operating system or through mobile applications where the user is
of mobile data in order to map hotspots of asked to opt-in to sharing their location. They can also be detected
infection, determine changes in mobility by wearable devices or navigation systems to provide location
patterns, or track contacts between at-risk data. When analyzed individually, GPS location data is subject to
or infected individuals. The ubiquitous privacy regulations, given its sensitive nature. 8
nature of our mobile devices and the fact
that human mobility is one of the key fac- Base transmitter stations (BTS): BTS—or cell towers—facil-
tors in the spread of an infectious disease itate signal reception of cell phones and other wireless devices.
make these devices a formidable tool to Thus, carriers are able to know where devices are, based on which
understand and measure our movements. tower they connect to for services as well as the signal strength of
the connection. Given that each tower has a unique ID, from the
The most widely used types of loca- tower ID and the signal strength one can infer a device’s location.
tion and proximity data collected by cell BTS location information is useful for inferring aggregate mobility
phones in the context of the pandemic are patterns but not highly accurate in location tracking of individuals,
summarized in Box 1. Each of them has its as their spatial granularity depends on the density of cell towers
strengths and weaknesses, with varying in a region. For instance, two devices connected to the same rural
degrees of privacy implications. BTS could in fact be kilometers apart.
Given that location and proximity data Wi-Fi: Wi-Fi signals tend to provide more accurate indoor
can be key sources of information for location data and can often generate more granular data. Mobile
understanding the spread of a pandemic, devices can scan for nearby Wi-Fi networks and crowdsource
analyzing how digital technologies and location. Nearby networks or “access points” can include any
applications can be used to safely collect Wi-Fi signals in the vicinity, such as that in cafes and shops or
and harness data, and studying the ways neighbors’ homes.
they are being used—or proposed—is
key to gage the opportunities and risks Bluetooth: Bluetooth technology is common in portable
of these solutions for COVID-19 response devices and can be thought of as a beacon that broadcasts one-
efforts. way signals which other devices can pick up (think of connecting
your phone to wireless headphones) when enabled. This occurs
through bursts of information packets dispersed into the electro-
B. Technologies, magnetic spectrum, which other Bluetooth-enabled devices then
detect. No direct connection has to be established, as devices
applications and exchange identifiers. Bluetooth can be used to infer location or
proximity. In the case of location, a registered device in a known
uses location can infer the locations of other devices that are visible
to it via Bluetooth with a certain signal strength. In the case of
In this paper we consider technological proximity, Bluetooth-based signals can be sent to other devices
tools developed from the application of within a certain range to collect proximity data rather than absolute
scientific knowledge to raw materials for location. Bluetooth-based proximity information is generally more
practical purposes, i.e. digital technolo- privacy preserving than absolute (e.g. GPS, Wi-Fi) location data.
gies. Several applications have been iden-
tified below within the scope of pandemic Many devices use a combination of GPS with other forms of
containment using combinations of tech- location signals such as Wi-Fi and Bluetooth to improve the preci-
nologies and the different data sources sion of the devices’ location.
detailed above.
8 https://fpf.org/2020/03/25/a-closer-look-at-location-data-privacy-and-pandemics/; https://theintercept.
com/2020/05/05/coronavirus-bluetooth-contact-tracing/.
https://gimbal.com/location-data-guide/These include: websites can also be a tool for citizens to
1
access reliable, trustworthy information
Self-assessment / regarding symptoms and next steps to
symptom tracking take when experiencing them.
Self-assessment and symptom track-
ing websites or apps allow users to report While these applications do not nec-
their symptoms and get instant feedback essarily require personal data in order to
on their assessed risk through interactive fulfill their promise, they do often collect
forms and surveys. In many markets with information from users, including home
low smartphone penetration, symptom address, phone number and location.
tracking messages can be displayed via These apps have been deployed widely
a USSD menu (the user opens a menu by local and national governments in, for
and picks from a range of options) which example, Afghanistan, Colombia, Kenya,
enable two-way flash messages. These Singapore and Turkey. A recent study
types of applications can help govern- by Zoe Global, Massachusetts General
ments to better handle citizen demand Hospital and King’s College, which
for trustworthy feedback and information tracked self-assessment applications in
when faced with symptoms and to moni- Sweden, the UK and the US found that
tor disease outbreaks. This in turn enables these apps could be “remarkably effective
a better use of resources and medical in predicting coronavirus infections”.9 Nev-
services: to an extent, these apps and ertheless, while self-reporting apps can be
websites relieve some of the strain put on very useful there are caveats to consider,
hospitals and facilities as self-screening such as a high variance in self-reporting
can help rule out infection and reduce the or misreporting due to a misperception
need for patients to seek a formal diagno- of users’ own realities. Awareness of the
sis. It can also suggest containment and human factor in these types of applica-
control measures to individuals at risk. tions is important.
Given the large amounts of misinformation
9 https://www.nytimes.com/2020/05/11/health/coronavirus-sym-
surrounding COVID-19, these apps and ptoms-app.html?auth=login-email&login=email
9
Impact Assessment Determine wheth-
er—and how—various interventions affect
the spread of COVID-19 and identify ob-
stacles hampering achievement of objec-
tives of particular interventions.
Prediction Leverage real-time popula-
tion counts and mobility data to enable
new predictive capabilies to assess future
risks, needs and opportunities.
Cause and Effect Identify key drivers and
consequences of implementing measures
to contain the spread of COVID-19. Estab-
lish variables with incidence on a problem.
Situational Awareness Better under-
stand COVID-19 trends and geographic
distribution.
Centralized / Decentralized Epidemiological
Contact Tracing
Surveillance and
Social Media Analysis / Enforcement
Citizen Surveys
Self Assessmnt /
Flows Modelling / Symptom Tracking
Mobility Mapping
Figure 1. Purpose and applications of digital technologies for
COVID-19 response and recovery efforts
Source: Data-Pop Alliance, 20202
Contact tracing applications
Contact tracing is a central techni-
que often applied in epidemiology
that has gained widespread attention
amid COVID-19 response efforts. The
objective here is to quickly identify poten-
tially at-risk individuals who have been in
close contact with a recently diagnosed
positive case of an infectious disease
requiring compulsory reporting, as is the
case for SARS-CoV-2. Once these people
have been identified, the main goal is to Box 2. CoronaMadrid10 10
quickly test and isolate those with a con-
firmed coronavirus infection so as to break and COVID-19 CDMX11
the chain of transmission.
In early March, the autonomous community of Madrid in Spain
Traditional contact tracing involves released CoronaMadrid, a self-assessment application available
carrying out epidemiological interviews for Android and Apple phones, as well as in web form. One of
(typically performed over the phone) to the main objectives with this application was to reduce traffic and
collect relevant data about the symptoms, demand on mobile hotlines, while allowing officials to continue
mobility and social behavior of patients. providing instructions and recommendations to citizens. Individ-
Personal information is commonly col- uals are only prompted to use this application if they experience
lected in these interviews, including the symptoms. Users can opt-in to share their locations to provide
phone numbers of all the people with public health organizations with more precise information to inform
whom the patient has been in close con- their responses. Users must share their phone numbers.
tact within the past N days (for COVID-19
the latest recommendation is 48 hours). Conversely, the tool created by the government of Mexico City in
Traditional contact tracing has four intrin- March is a uniquely web-based form. It requires users to share per-
sic limitations where digital tools might sonal information such as their cell phone number and complete
help. First, it relies on the patient’s mem- address. After assessing symptoms, the tool will classify individuals
ory. Second, all the close contacts need according to their risk factor and recommend a series of actions
to be known to the patient such that (s)he to take. The privacy notice for this tool establishes that the data
can share their contact information with collected may be used by judicial and administrative, federal and
the contact tracer. Third, it does not work local authorities.
well across borders. Fourth, it is expen-
sive, human-resource intensive and time 10 https://coronavirus.comunidad.madrid/
consuming. Despite these limitations, it is 11 https://test.covid19.cdmx.gob.mx/
an effective tool to help contain the spread
of an infectious disease, assuming the
contact tracing teams are properly scaled
to the incidence of the disease and the
information they collect is in digital form,
ideally using state-of-the-art tools, so that
it is readily available for analysis and deci-
sion making.
Since the outbreak of COVID-19 has
exceeded the capacity of most manual
contact tracing teams worldwide, public
officials in many countries are turning to
smartphones as a key tool to complement
these existing initiatives. Thus, we are
witnessing the emergence of dozens of
smartphone-based contact tracing appsglobally. If indeed smartphone apps were
able to passively record close contacts
between individuals, they could automat-
ically generate the necessary contact
traces, such that at-risk individuals who Centralized Decentralized
had been in close contact with an infected A central system John and Jane’s
person could be notified, tested, and generates a series phones generate
isolated if positive. This process would of user-specific a series of user-
enable the transmission chain to be bro- anonymous IDs specific, anonymous
ken and prevent community transmission and sends them to IDs.
of the disease. John and Jane’s
phones.
Contact tracing applications rely
on proximity technology and/or loca- 1 John and Jane
tion traces to identify potential contacts don’t know each
between individuals. First efforts on this other but chat for
front—such as those deployed in China 10 minutes in a
or South Korea—leverage the GPS loca- park.
tion alone or in combination with other
data, such as credit card transactions
or visual surveillance camera footage.
These applications infer close contacts if 2 Their
individuals have been within a radius of smartphones
1.5-2 meters of each other and for at least exchange their
15 minutes. However, limitations associ- anonymous
ated with GPS—including imprecisions in ephemeral
indoor (e.g. buildings) and transport (e.g. identities over
subway, planes) environments, as well pri- Bluetooth LE.
vacy concerns, have led technologists and
governments to turn towards Bluetooth as 3 A few days later 11
the main sensor to detect close contacts John tests positive
between individuals via smartphone apps. for COVID-19
and, via the
Bluetooth-based contact tracing apps app, consents to
enable devices to share “digital hand- sharing his status
shakes” by sharing encrypted, unique as well as his test
identifiers (referred to in the literature as results.
tokens, beacons, pseudonyms, temporary
exposure keys (TEKs) or temporary con-
tact numbers (TCNs) to record contacts
that last “longer than a few minutes”12 and
located within a 1.5-2m radius.
The success of these apps depends
on many factors, including high adop-
tion rates and tight integration with public
health systems, such that both doctors and 4 John’s phone 4 John’s phone
infected individuals can report positive sends the sends his own
cases and at-risk individuals can be duly anonymous anonymous
notified. Unlike manual contact tracing, identifiers of identifiers (or a key
these applications record contacts that a people he has that can derive
person may not remember or know they met to a central them) to a central
have come in close proximity with. How- database. database.
ever, these applications are not exempt
from their own limitations and challenges, 5 The central 5 Jane’s phone
including difficulties in reliably detecting database downloads the
close contacts via Bluetooth, battery con- matches entire central
sumption, trolling and hacking scenarios,13 the reported database
human-centric challenges,14 privacy and identifiers to and checks
security risks and low adoption rates, par- John’s contacts for matching
ticularly within the most vulnerable groups and sends them identifiers.
(e.g. the elderly, low socioeconomic levels an alert.
and skeptics).
6 Jane’s phone
12 https://venturebeat.com/2020/04/13/what-privacy-preserving- alerts her that
coronavirus-tracing-apps-need-to-succeed/ someone she
13 In the UK, for example, hackers successfully launched
phishing attacks with the National Health Service’s app.
met has tested
A phishing message redirected victims to a fake website whe- positive.
re they were asked to type in their personal details; www.itwire.
com/guest-articles/guest-opinion/how-hackers-can-abuse-con-
tact-tracing-apps-91032.html
14 The app’s design needs to be based on how people can,
need and want to perform tasks, rather than expecting users to
Figure 2. Centralized vs.decentralized
adjust and accommodate their behaviors to the product. contact tracing app-based models+50 % more movement than usual
Feb. 16 Feb. 23 March 1 March 8 March 15 March 22 March 29
March 19
-50 % less movement than usual
March 16
-100 %
Average change for wealthiest and poorest
Figure 3. “Location Data Says It All: Staying at Home During Coronavirus Is a Luxury”
Source: NY Times, 3 April 2020, https://www.nytimes.com/interactive/2020/04/03/us/coronavirus-stay-home-rich-poor.html
3
Modelling and mapping
population flows Box 3. Bluetooth-based
Mapping flows, or mobility of peo- 12
ple over time and space, has been one of contract tracing app models
the more common applications of private
data for social good initiatives in the recent The debate over app-based contact tracing models centers fore-
past. Mobile phones can often act as indi- most on what data is captured so as to ensure that only strictly
cators of human mobility and give insights relevant information is collected. Second, there are debates over
into behavior. Aggregated and anonymi- where this data should be stored. Concerns also exist around the
zed location data can be sourced from aggregation of data, underlying privacy configurations and who
various technologies such as GPS, mobile should have access to this data—including public authorities. For
cell towers, Wi-Fi networks, Bluetooth con- the apps that exchange tokens via Bluetooth, two main architec-
nections, surveillance video, credit card tures have been proposed: centralized and decentralized. In both
records and wearables, as well as many cases, the token exchange takes place locally in the phones. The
other devices and apps. In the case of main difference stems from (1) who provides the phones with the
COVID-19 responses, the analysis of these initial seed used to generate such tokens, and (2) what information
data enhances findings by identifying risk the phones send to a centralized server when their user is tested
and potential hotspots, assessing public positive for coronavirus.
responses and the effectiveness of social
contact and mobility contention policies, In the centralized model, the initial seed to generate the to-
and detecting where more resources may kens is given by a trusted, centralized server typically controlled
need to be channeled. Moreover, human by administrators or public health authorities. Moreover, when an
mobility is a valuable input to computa- individual tests positive and upon recording this event in the appli-
tional epidemiological models. While the cation, their phone sends all tokens of the devices it has had close
aim of flow modelling is descriptive, its use contact with (e.g. over the preceding 14 days) to a centralized
can cross the line and be used as a tool server. The central server matches the tokens and alerts users to
for control by authoritarian governments— a potential contact. Resulting aggregated, anonymized data can
and, even more surprisingly, by others help experts fine tune the risk calculation to determine whom to
perceived to be considerably less so—by send a notification to and also allows administrations to detect
applying stringent enforcement policies. infection patterns in society, which is crucial input when designing
policies and measures aimed to curb the spread of a disease.
These analyses can also simply shed
light on the effects and effectiveness of In the decentralized model, the initial seed to generate the
containment measures, especially across tokens is given by the operating system (in the case of the Apple/
different demographics, potentially pointing Google API) or by the app itself. When an individual tests positive,
to enabling and constraining factors. For upon recording this event in their phone, their app only sends to
example, a US study revealed that, when the central server their list of tokens. All the phones running the
on March 16th people were asked to stay app periodically poll the central server for the list of tokens of pos-
at home, those living in richer areas had itively diagnosed individuals. Given that the phones have the list
already reduced their mobility by nearly of contact traces, they locally check if there is a match between
half whereas people in poorer areas did not their contact traces and the list of tokens associated with recently
substantially reduce theirs until three days diagnosed individuals. If a match is found, the app triggers a noti-
later, suggesting structural impediments fication with indications of what to do next. In this case, no central
for the latter to staying at home and limiting authority has visibility on how many users have been notified for
their exposure to the virus. each registered positive case.Box 4. Debate over contact
tracing apps: moving from
centralized to decentralized
Descriptive tools such as flow mapping
are used to look at people’s movement approaches
patterns locally to gage risks or potential
hotspots, as well as to assess how people Early on, the general public in several European countries
are responding to the virus and response showed support for centralized models using pseudonymized
measures to inform public response. In the proximity data. For example, the Pan-European Privacy Preserving
Valencian region of Spain, a team of Proximity Tracing (PEPP-PT) initiative developed an open protocol,
experts has been working closely with defining standards for tracing apps built on it and uses a blend
the president of the region on a variety of centralized and decentralized methods.18 The UK and France
of data-driven tasks related to COVID-19, have also developed their own centralized apps, with France
including quantifying and modeling human being the first to launch its voluntary app StopCovid, using a pro- 13
mobility captured by the mobile network tocol known as Robert to complement existing manual contact
infrastructure. In a pioneering collabora- tracing. While the data protection authority, CNIL, has not raised
tion between the Spanish National Office any major flags, concerns are being voiced over the use of pseud-
of Statistics and the three largest telecom- onymized data which necessitates a certain level of trust that the
munication companies in Spain, experts government is indeed respecting the limitations around data col-
have been able to assess the success of lection it has detailed.19 Supporters of these models ensure that
containment measures and their impact fully privacy-preserving techniques are in place, along with ready-
on the spread of the pandemic, estimating to-use, well-tested, properly assessed mechanisms, and support
that over 40,000 lives were saved in the for interoperability.
process.15
Other countries, like the US, quickly turned to decentralized
In many countries, pre-existing tools models, using e.g. the Apple/Google API that combines Bluetooth,
designed to look at flows of people for cryptography and location tracking. The debate between cen-
applications in the public transport or tour- tralized and decentralized contact tracing models has continued
ism sectors have also been adapted to within and across countries and, as response efforts develop, more
the current COVID-19 context. In Austria, countries, including Germany, are choosing to pursue decentral-
for example, Invenium, an existing col- ized models, mainly due to fears of function creep.20
laboration between A1 Telekom Austria
Group and a local university, developed a Researchers at MIT have created Private Kit: Safe Paths,21 a
motion analysis application that was used free and open-source application that uses both Bluetooth and
to model human mobility flows for applica- GPS tracking based on the decentralized model. One key feature
tions in traffic congestion or tourism sites is its interoperable standards to ensure usability between differ-
to assess the effectiveness of response ent apps within or across different countries based on an open
measures to reduce movement and social source Temporary Contact Number protocol to ensure interopera-
contact.16 The COVID-19 Community bility. Decentralized Privacy-Preserving Proximity Tracing (DP-3T),
Mobility Maps generated by Google are developed by researchers in France, Germany, the Netherlands
based on users’ aggregated location data and Switzerland, creates a virus contraction risk score gener-
and reflect community-level behavior such ated from an algorithm running on the user’s data locally on their
as travel, for example, to grocery stores, device. Decentralized apps are the subject of criticism particularly
parks and public transport centers.17 Many with respect to making it more difficult for health authorities to have
of these mappings of concentration and the necessary data regarding how many close contacts receive
movements of people use aggregated and a notification for each positive case; agility and practicality rely
anonymized data, further calibrating policy on cryptography which is complex and requires challenging and
response and containment measures such frequent updates of parameters, especially at the scale that would
as social distancing and contact tracing. be needed to be effective in this epidemiological response.22
18 https://www.pepp-pt.org/
15 https://www.gva.es/es/inicio/area_de_prensa/not_detalle_area_ 19 https://www.france24.com/en/20200602-france-rolls-out-covid-19-tracing-app-amid-privacy-debate
prensa?id=858477 20 https://techcrunch.com/2020/04/27/germany-ditches-centralized-approach-to-app-for-covid-19-
16 https://www.reuters.com/article/us-health-coronavirus-europe- contacts-tracing/
telecoms/european-mobile-operators-share-data-for-corona- 21 https://www.technologyreview.com/2020/05/07/1000961/launching-mittr-covid-tracing-tracker/
virus-fight-idUSKBN2152C2 22 https://venturebeat.com/2020/04/13/what-privacy-preserving-coronavirus-tracing-apps-need-to-
17 https://www.google.com/covid19/mobility/ succeed/Figure 4.
HU
IT
Predicted baseline
NL
mobility patterns
O
PL
for 28 January to
N
18 February 2020.
PT
Individual prob- FI RO
ability of moving ES
between the top SI
20 European
EL TR
countries with the UK
greatest outward
mobility.
AT
DE
BE
CH
BG
Source: https://science.
sciencemag.org/content/
sci/369/6510/1465.full.pdf
4
Surveillance and enforcement
In general, tools for surveillance and Box 5. Applying research on
quarantine enforcement analyze
sensitive personal data. More granular, malaria-related mobility
often pseudonymized (but in the case of
COVID-19 also increasingly sensitive) flows to COVID-19: the case
data sources have emerged to monitor 14
people’s movements with a view to con- of Mozambique23
taining the spread of the disease. This
exceptional circumstance has led many Novel mobility analyses by mobile networks have proven
governments to consider loosening or themselves useful for mapping the spread of many diseases. A part-
sacrificing individual privacy during the nership was established between Vodafone, the University
response period for the sake of curbing of Southampton, the Clinton Health Access Initiative and the
contagion and saving lives. Facial rec- National Malaria Control Program; it was backed by the Bill and
ognition systems, mobile tracking apps, Melinda Gates Foundation. Together they analyzed mobility flows
wearables, geolocation, credit card and in Mozambique, a country where malaria poses a great burden
financial transactions and transport data on the economy and the general well-being of the population.
are being used for real-time monitoring of By examining aggregated and anonymized population flows and
compliance with response policies. malaria incidence in the country, the analysis allowed a better
prioritization of resources and geographically stratified actions
Examples of crisis response involving by providing malaria “sinks” and “sources” – thus showing how
both new technologies and sensitive per- the disease moves across the country with population flows. The
sonal data along with emergency public lessons learned from this analysis were then quickly applied to
health policies and law enforcement mea- COVID-19 and, by leveraging the global reach of Vodafone’s
sures can be seen in several countries mobile networks, mobility insights were extracted. They were used
globally. Apps deployed in China, such not only for tracking how populations were responding to govern-
as Alipay Health Code, are grounded in ment measures, but were also fed into an epidemiological model
technology for symptom tracking, assign- on the effects of travel restrictions and lockdown behaviors during
ing a color-coded QR code to the user the spread of the disease.
indicating their risk level.24 This tool goes
beyond self-assessment as all citizens 23 https://www.vodafone.com/perspectives/blog/world-malaria-day-2020-vodafone-fighting-malaria.
are required to use it and personal data is
sent to law enforcement bodies to enforce
quarantine measures based on an individ-
ual’s risk level. Amongst others, this app
has been criticized for its lack of auditabil-
ity as the rules behind the assignment of
risk are not widely known. Though appar-
ently not compulsory, access to many ser-
vices and activities in China is dependent
on receiving a green code.25
Taiwan was one of the first countries
to lead the development of technologies
for quarantine enforcement using mobile
24 https://www.nytimes.com/2020/03/01/business/china-corona-
virus-surveillance.html
25 https://www.afr.com/world/asia/how-china-s-health-code-app-
is-used-to-fight-infection-20200424-p54mzkdata, implementing a “digital fence”.26 This
integrates location data from cell phones
to trigger an alert system if anyone moves
too far from their home and issues a fine
for breaking quarantine restrictions.27 Box 6. Trade-offs in quick
Hong Kong has introduced wearables in
order to enforce the 14-day quarantine for containment and digital rights
anyone arriving at the airport: an electronic
tracker wristband, paired with a mobile In Israel, the response efforts went beyond introducing new
app used to calibrate the wristband, using technologies using sensitive personal data; the authorities also
geofencing technology.28 In the case of the implemented an emergency law passed to specifically track
tracker wristband, the technology is said to infected individuals and their contacts in order to enforce individ-
preserve privacy as it does not track indi- ual quarantine measures. Importantly, there were time limitations 15
viduals’ exact location, but simply signals involved in the implementation of this technology set out in the law,
whether an individual is inside or outside initially to thirty days. As of September, the security service pro-
of their home. gram used for contact tracing was still in place.30 However, other
key questions such as who has access to this data for analysis,
There has been mixed public percep- what other types of analysis may be performed and when the data
tion of the development and use of these will be deleted have not been specified. Digital rights advocates
applications. China, which has seen have pushed back on these measures, warning of the risks not
breaches of data collected for the COVID- only of mass surveillance but also of targeted law enforcement
19 response entailing negative conse- action, as there are fears of a slippery slope as these methods
quences such as discrimination or stigma, unfold.
has strengthened public debate around
privacy in the country.29 Similar concerns South Korea has gone even further. It has not only used sen-
and debates have been raised about pri- sitive personal data from mobile phone tracking, credit card
vacy and digital rights globally, such as transactions, as well as face-to-face interview data with patients,
the examples detailed in Box 6. but used this information to publish a publicly available map to
allow citizens to verify their potential contacts and the patterns of
Transparency in the design and use of those infected as well. While the data does not include personal
many of these apps across several coun- identifiers, there is a high potential for re-identification of individ-
tries has been called into question as uals due to the granularity of location data, mobility patterns and
there is often a lack of clear data privacy even personal descriptions of those infected. Though the trans-
policies, communication with the public, parency and openness of the government has allegedly increased
or limitations on who has access, for what trust in its containment efforts, the fear of social stigmatisation is
purpose and for how long. Complicating high, given the amount of information usually released about con-
this is the ambiguity in legal and regula- firmed patients.
tory frameworks on data protection and
privacy that has given some governments India has become the only democratic nation in the world to
ground to implement measures that may require a majority of its citizens to download and use its tracking
infringe on digital and human rights in app, Aarogya Setu, with threats of fines, losing jobs, or jail if non-
cases of emergency, with unclear limita- compliant. While official policy maintains that the application is
tions on these provisions. It is not impossi- voluntary, all government employees, many large private compa-
ble to imagine that certain abusive policies nies, landlords and even city governments are mandating its use.
may linger long after any justification for The technology underpinning the application differs from many
them has disappeared. others as it allows for enforcement as well, in that it goes beyond
exposure notifications from proximity data to assigning color-coded
risk badges, similar to China’s Alipay Health Code app. Other con-
26 This recent AI&I exchange with Audrey Tang, Taiwan’s cerns have been raised about the lack of legal frameworks around
Digital Minister can be accessed here: www.youtube.com/ data privacy and lack of transparency around data access or use
watch?v=sfNESpLr0pk
27 https://qz.com/1825997/taiwan-phone-tracking-system-
from the app as the developers’ profiles are not fully disclosed to
monitors-55000-under-coronavirus-quarantine/ the public and include many private companies.31
28 https://qz.com/1822215/hong-kong-uses-tracking-wristbands-
for-coronavirus-quarantine/ 30 https://hamodia.com/2020/09/08/contact-tracing-app-prevent-infection-spread-ineffective-kosher-
29 Tracing.Testing.Tweaking. Approaches to data-driven phones/
Covid-19 management in China (Meric paper) 31 https://www.technologyreview.com/2020/05/07/1001360/india-aarogya-setu-covid-app-mandatory/5
Epidemiological modelling
Both metapopulation and individual 1M
computational epidemiological mod- 100K
els benefit from high quality, real-time data
about the number of people infected, hos- 100
pitalized or in intensive care. Moreover,
0
2 cycles
human mobility (computed from, for exam- 3 weeks
ple, the mobile network infrastructure) and
quarantine information enable building 1M
more accurate models and predictions. 100K
Having an underlying model enables run-
ning it under different scenarios—such as 100
different social contention, mobility and
0
contact tracing situations—to assess the 4 weeks
impact that different non-pharmaceutical
interventions (NPIs) might have on the
incidence and spread of the disease. 1M
100K
Other technologies such as smart
thermometers and AI-based diagnostic 100
tools have been providing different ways
0
3 cycles
to map and predict the evolution and 3 weeks
spread of the virus. Data from smart ther-
mometers were used in the US to create 1M
HealthWeather, a map which visualizes 100K
seasonal illness linked to fever, based on
aggregated, anonymized data from the 100
thermometers and mobile applications.32
0
In the fight against COVID-19, the benefits 4 weeks
of real-time data drew attention to similar 16
sources of data such as wearable fitness
and health devices, encouraging users to 1M
synchronize their existing devices to spe- 100K
cific apps such as MyDataHelps to pool
data. However, concerns were raised 100
around the accuracy of these efforts as
0
4 cycles
they are based on information about the 3 weeks
behavior of flu-like illnesses, as well as the
representativeness of these initiatives as 1M
their data collection appears to be biased 100K
towards people who have access to wear-
Number infected
able devices. 100
0
Real-time modelling, and therefore real- 4 weeks
time data access, are critical to enable 1 2 3 4 5 6
timely response policies particularly in the
case of outbreaks. Data access is also key Time since beginning of simulation in month
to accelerate scientific research in order
to better diagnose, treat and develop vac- Asynchronous NPIs Synchronized NPIs
cines. (non-pharmaceutical interventions)
32 https://www.washingtonpost.com/national/health-science/ Figure 5. Cases over time, when lockdowns are synchronized
start-ups-health-weather-map-may-help-forecast-spread-
of-diseases-like-covid-19/2020/03/26/36c069b8-6ef0-11ea-
or unsynchronized across all European countries.
a3ec-70d7479d83f0_story.html Source: (N. W. Ruktanonchai, 2020) https://science.sciencemag.org/content/369/6510/1465You can also read
