Ubiquity, an ACM publication
October 2020
COVID-19 and Computation for Policy
by Jeffrey Johnson, Peter Denning, Kemal Delic, and Jane Bromley
Editor’s Introduction
Governments across the world are formulating and implementing medical, social, economic and
other policies to manage the COVID-19 pandemic and protect their citizens. Many governments
claim that their policies follow the best available scientific advice. Much of that advice comes
from computational modeling. Two of the main types of model are presented: the SIR
(Susceptible, Infected, Recovered) model developed by Kermack and McKendrick in the 1920s
and the more recent Agent Based Models. The SIR model gives a good intuition of how
epidemics spread; including how mass vaccination can contain them. It is less useful than Agent
Based Modeling for investigating the effects of policies such as social distancing, self-isolation,
wearing facemasks, and test-trace-isolate.
Politicians and the public have been perplexed to observe the lack of consensus in the scientific
community and there being no single ‘best science’ to follow. The outcome of computational
models depends on the assumptions made and the data used. Different assumptions will lead to
different computational outcomes, especially when the available data are so poor. This leads
some commentators to argue that the models are wrong and dangerous. Some may be, but
computational modeling is one of the few ways available to explore and try to understand the
space of possible futures. This lack of certainty means that computational modeling must be seen
as just one of many inputs into the political decision making process. Politicians must balance
all the competing inputs and make timely decisions based on their conclusions—be they right or
wrong. In the same way that democracy is the least worst form of government, computational
modeling may be the least worst way of trying to understand the future for policy making.
http://ubiquity.acm.org
1
©2020 Association for Computing Machinery
Ubiquity, an ACM publication
October 2020
COVID-19 and Computation for Policy
by Jeffrey Johnson, Peter Denning, Kemal Delic, and Jane Bromley
Worldwide the COVID-19 pandemic has been one of the greatest medical, social, economic, and
political shocks the world has experienced since the Second World War. Economies around the
world are reeling and taking on unprecedented levels of debt as they to try to survive the
pandemic. Worldwide, governments are adopting policies, which they say, “follow the science.”
That science includes traditional sciences such as biology, medicine, and epidemiology
combined with the approach of complex systems science and computational modeling. We
explain this new computer enabled science and how it is used to create computational models
that enable policy makers to forecast possible futures and possible outcomes of policy as they
try to manage this multi-trillion dollar global catastrophe. We also examine some of the
controversies associated with modeling and policy.
The Emergence of a Deadly New Virus
In the autumn of 2019 a new virus appeared in China. The world watched on as the epidemic
spread alarmingly through the eleven million people of the city of Wuhan in Central China’s
Hubei province. China’s new “flu” seemed far away in December 2019. Most countries around
the world did not see the significance of this until January 2020 and even then the scale of the
threat was not fully appreciated until two months later. Some countries were following their
experience with SARS and Ebola, where local containment had proved successful. COVID did
not follow the same pattern, taking everyone by surprise. For example, on January 24, 2020
Public Health England said the risk was low; by January 31st this was raised to the risk being
medium; by the 4th of March the British government issued instructions on hand washing and
social distancing by two meters in public places; and by March 26th it declared a hugely
expensive national lockdown policy in response to research from Imperial College London [1].
The spread of a new coronavirus, designated COVID-19 by the World Health Organization
(WHO), was declared on March 11, 2020 to be a pandemic: “Pandemic is not a word to use
lightly or carelessly. It is a word that, if misused, can cause unreasonable fear, or unjustified
acceptance that the fight is over, leading to unnecessary suffering and death. This is the first
pandemic caused by a coronavirus and we have never before seen a pandemic that can be
controlled … we have called every day for countries to take urgent and aggressive action. … We
have rung the alarm bell loud and clear.” [2]. “COVID-19 is transmitted primarily by particles
that can travel two meters or more when an infected person coughs or sneezes and
contamination of hard surface where the virus can survive for a few hours” [3]. The infection
transmission depends on number of viral particles multiplied by time of contact. Passing
http://ubiquity.acm.org
2
©2020 Association for Computing Machinery
Ubiquity, an ACM publication
October 2020
someone wearing a mask in a socially distanced store in a few seconds is less risky than
socializing in a noisy crowded pub. Frontline workers in the health system are particularly
vulnerable. In Europe the most severe early cases of the pandemic occurred in Northern Italy
where the health system was overwhelmed with severely ill people who could not all be
treated in hospital [4].
COVID-19 is a new virus and in January 2020 almost nothing was known about it. In the months
that followed policy makers were faced with the completely new problem of estimating how
many people would be infected, how fast the virus would spread, how many acute hospital
beds would be required, how much PPE (personal protective equipment) would be required,
how many testing kits would be required, and so on. To answer these urgent questions they
turned to their scientists and computational modelers.
Policy Following the Science
In the early days of the pandemic politicians in many countries repeatedly said that they were
“following the science.” For example, on March 11, 2020 the U.K. Minister of Health said in the
House of Commons,“Our approach will be guided by the best scientific evidence and medical
advice, and we will take all necessary measures to deal with this outbreak” [5]. Most countries
have scientific laboratories and university departments with expert knowledge on epidemics. In
the U.K. the government appeals to its Scientific Advisory Group for Emergencies (SAGE) to
advise it on COVID-19. However what appears to be a sensible partnership between objective
scientists sharing scientific facts and politicians deciding the best course of action ignores the
essential nature of scientific knowledge as contingent and contested. As time went by it
became clear there were disagreements on the science between members of SAGE and other
scientists, and even scientific disagreements within SAGE. In May 2020 a former government
chief scientific advisor set up an “alternative SAGE” [6]. As it became clear that there was not a
single definitive science to follow, political decisions had to be made which, if they went wrong,
could not be blamed on “the science.” It is the nature of science that its progress depends on
fierce battles between alternative theories and viewpoints. Although there is great consensus
across vast swathes of science, the scientific community rarely has a single view on anything
new.
Lockdown
Sooner or later governments worldwide imposed different variants of “lockdown” requiring
people to stay at home isolating themselves, with most business closed and many people being
“furloughed.” In the U.K. this meant people receiving a reduced government-funded income
from their employer instead of being let go so that the business could start again when the
epidemic was over. Even so many people were laid off and many furloughed people will still
lose their jobs as businesses adjust to the new economic situation after lockdown. For example
http://ubiquity.acm.org
3
©2020 Association for Computing Machinery
Ubiquity, an ACM publication
October 2020
on July 1st 12,000 U.K. job losses were announced including many from iconic department
stores such John Lewis and Harrods [7]. In Britain the government has committed a budget of
billions of pounds to keep the economy on life support until it has recovered. The social and
economic cost of lockdown is enormous. Also lockdown has increased the number of nonCOVID deaths including untreated cancers and suicides.
Policy Needs Forecasts of the Future
Policy makers need to know what might happen next, and what will be the likely outcomes of
their policies. The conventional scientific approach cannot answer these questions. Pandemics
are complex, highly dynamic multilevel systems. They do not respect traditional discipline
boundaries between the biological, medical, social, economic, and other sciences. Global
spread and infection numbers emerge at the macro level from the micro-level activities and
interactions of people in a very wide variety of individual social situations.
Most systems administered by politicians cannot be predicted in a conventional scientific sense.
Prediction is the gold standard of conventional science. Theories live or die by their ability to
predict events in the future and by experiments showing the predictions are correct. Generally
these are point predictions that a particular event will occur at a particular point in time. Such
predictions are rarely possible in complex systems. There are many interrelated reasons for this
but together they make the dynamics of complex multilevel systems “sensitive to initial
conditions.” Change the starting position slightly at any level and the outcome can be very
different. For example, in the U.K. the apparently minor decision not to maintain a complete
stock of protective personal equipment (PPE) in 2009 contributed greatly to the inability to
protect frontline medical and service staff in 2020 [8]. Complex systems cannot be predicted in
a deterministic sense. At best future events can be identified and approximate estimates made
of their likelihood. Apart from guesswork and blind certainty, the only scientific way to forecast
possible behaviors of complex systems is mathematical, statistical, and computer modeling and
computer simulation building on conventional scientific knowledge.
Models and Computation
A model is a description of a system or process to show how it works. We have mental models
about all aspects of our lives and we use them to try to understand things and decide what to
do. The idea of a scientific model is that the “real world” is mapped to an abstraction in words,
numbers, pictures, etc. Within that abstraction there are transition rules that map the model
from its state “now” to another state in future time (Figure 1). In traditional science if the
modeled dynamics and the real dynamics give the same outcome, the modeled dynamics are
accepted as a viable theory of the actual dynamics. This validation has to be repeated many
times for people to trust it.
http://ubiquity.acm.org
4
©2020 Association for Computing Machinery
Ubiquity, an ACM publication
October 2020
Figure 1. A simplified view of modeling.
Typically modeling proceeds in two stages: (i) The model is compared with reality to see how
well it predicts outcomes when the only information provided is parameter measurements
from the actual world. Then (ii) forecasting is done with a validated model where the main
uncertainty is the values of the parameters in the future time period. The modeler has to make
assumptions about those future parameter values, which then go into the model to yield the
forecast. Most science aims to work with validated models and validated methods of parameter
forecasting, and the best science makes clear the underlying assumptions.
With complex systems the modeled dynamics are not expected to produce point predictions. In
fact the science of complex systems allows—or expects—the modeled dynamics to produce
different forecasts of future modeled states. A further complication is that complex systems
may undergo local or global phase transitions so assumptions that are valid at one time become
invalid at another [9]. Indeed one of the defining features of complex adaptive systems is that
they can reconfigure themselves and behave in new and different ways [10].
Complex systems science is computer enabled and one of its most powerful tools is simulation
by iterated computation. This involves computing the system state at the next tick of the clock
from its current state using transition rules. For systems that are sensitive to initial conditions
this kind of computer modeling involves running simulation programs many thousands or
millions of times to produce distributions of outcomes. This gives a view of the space of future
events rather than saying that any particular event will occur. A single run of a computer
simulation tells you almost nothing. Many runs from many starting positions are needed to
populate the space of future possibilities so that its shape and the likelihood of alternative
events can be evaluated. For these reasons real-world computer modeling often requires large
powerful computational infrastructure. Nonetheless useful computer modeling can be done on
standard PCs and notebook computers. Also modeling embraces AI, IoT, and social media, e.g.,
the new apps being developed using mobile telephony to collect contact information to support
the test-trace-isolate policies used to try to contain the pandemic at local levels.
http://ubiquity.acm.org
5
©2020 Association for Computing Machinery
Ubiquity, an ACM publication
October 2020
Macro-level Epidemic Models
A key parameter for epidemics is the basic reproduction number, R0, which is defined to be the
number of people an infected person will infect at the beginning of an epidemic. This depends
on many things, some biological and some cultural, such as shaking hands, kissing, hugging, and
close contact at shared meals, parties, and musical events. As an epidemic develops the
reproduction number, R, may change according to the number of susceptible people, local
circumstances, and policy interventions.
In 1927 Kermack and McKendrick proposed one of the first models of epidemics [11]. It was
called the SIR model and could be used to investigate the relationships between the numbers of
susceptible (S), infected (I), and recovered (R) people in a population. Interest in the SIR model
was rekindled by a paper by Anderson and May in 1992 [12]. The model’s abstraction is that
there is a population made up of susceptible people, infected people, and recovered people. In
the basic SIR model it is assumed that when people have recovered they are not susceptible to
reinfection. In the SIR model the reproduction number R is defined as the infection rate divided
by the recovery rate. When the infection rate is greater than the recovery rate (R > 1.0) an
epidemic can occur. When the recovery rate is greater than the infection rate (R < 1.0) the
epidemic dies out. This model is based on differential equations that link the increases and
decreases in the numbers of susceptible people (blue curve), infected (red), and recovered
(green) as shown in Figure 2.
The basic SIR model (https://cs-dc.uk/sir.html) is easy to program and gives results similar to
those in Figure 2(a). The red curve shows the proportion of the population infected at any time.
The worry for policy makers is that the peak of this red curve for infected people may swamp
the available healthcare provision, as was seen in Italy in March 2020 [4].
susceptible
recovered
recovered
cumulative
distribution
function
infected
(a) The SIR model
(b) The cumulative distribution S curve fits the SIR model
Figure 2. An example of the SIR model.
Although the SIR model gives a good understanding of the shape of epidemics and explains why
vaccination works (the number of susceptible people in a population decreases so that the
http://ubiquity.acm.org
6
©2020 Association for Computing Machinery
Ubiquity, an ACM publication
October 2020
infection cannot reach exponential growth) it has limited value for forecasting because
externally induced changes are likely. For example, it cannot support an analysis of the WHO’s
test-trace-isolate policy which, in the absence of a vaccine, is seen by many as the way to
control a pandemic.
Figure 2(b) shows the black S curve of the cumulative distribution function
𝑓 𝑥 = 1 (1 + 𝑒 !(!!!)/! ) is a good fit to the green curve of recovered people generated by
the Kermack-McKendrick model. Related S curves occur in other well-known applications such
as the Bass Diffusion model [13], a differential equation for the S-curve, which is a universal
feature of adoption and spread of technology in populations. The Bass model can forecast the
total number of adoptions and the inflection point of the S-curve. Using calibrated curves, such
as the S-curve models, works well in diffusion models when there is no change in the underlying
dynamics. Social distancing or lockdowns changes the underlying dynamics and makes these
models inappropriate for forecasting.
Figure 3. The phases of the empirical curves do not fit the S model.
Two major problems make forecasting epidemics difficult: changes in social behavior altering
the underlying dynamics and the availability of reliable data.
The first is illustrated by the different shaped curved in Figure 3. On June 30th India was still in
the exponential phase of the S curve. However after March 28th the USA no longer had an S
curve but had an almost linear trajectory. In contrast the number of new infections in the U.K.
tails off in June while in Italy the total number of people newly infected tails off towards zero.
The reason these curves deviate from the S curve is that policy and changes in public behavior
have changed the reproduction number. Although policy makers hope their policies will work—
and these curves give retrospective evidence that some policies have worked—modeling by
curve fitting has limited value for forecasting when there is no way of knowing when one kind
of curve will transform into another. For example, in September 2020 many countries have
http://ubiquity.acm.org
7
©2020 Association for Computing Machinery
Ubiquity, an ACM publication
October 2020
experienced new exponential growth and are now preparing for a “second wave” not
evidenced in the graphs of Figure 3.
The second problem with these models is that the data in many countries can be very poor. For
example in the U.K. there is systemic underreporting at weekends as new cases wait to be
added to the statistics for Monday or Tuesday. Worse is that infections may be missed or
misclassified. Even worse is in the absence of comprehensive testing we do not know how
many people have had the virus, how many people are asymptomatic, and how many people
are naturally immune. Thus the proportion of susceptible people in the population may be
much lower than 100 percent with a significant impact on the model. Even the best models
cannot give meaningful forecasts in the absence of reliable data.
In epidemiology with a new disease, such as COVID-19, there is no way to validate the model.
The best that can be done is taking models from similar past diseases and modifying them
according to your assumptions for future parameters. This is much more challenging than
standard scientific modeling because the model and the parameter forecast methods cannot be
validated beforehand.
Parameter forecasting for the Kermack-McKendrick model is a particular difficulty. For example
initial estimates of the initial reproduction for COVID-19, R0, were in the range 1.9 to 5.7 [14].
This makes a huge difference to the simulated evolution of an epidemic.
Agent-Based Modeling
One of the major tools of complex systems science is agent based modeling for computer
simulation. Instead of trying to model macro-level statistics such as the number of people
infected in a nation, in this approach these numbers emerge from the micro-level interactions
of simulated agents (people in this case). In the simplest cases people are represented by
colored dots on a computer screen: red for infected, blue for susceptible, and grey for
recovered. The dots move around the screen and when a red infected dot gets close to a blue
uninfected susceptible dot, with a given probability the susceptible agent becomes infected.
This is illustrated by the sequence in Figure 4. At the beginning (top left) no agents are infected,
at the top right the infection is in the exponential phase, at the bottom left the red peak of the
infection has been passed, and at the bottom right the epidemic is almost over.
http://ubiquity.acm.org
8
©2020 Association for Computing Machinery
Ubiquity, an ACM publication
October 2020
Figure 4. An agent-based simulation of an epidemic.
A model like this is makes it easy to understand the epidemic process and the lay person can
see the exponential growth of the epidemic and how epidemics can jump from one country to
another. However, although these animations can be alluring they depend on the assumptions
built into the simulation. Lay people may be willing to trust that the modelers’ assumptions are
reasonable, but they have no way check. Modelers do not always present the results of
sensitivity analyses (small changes in modeling assumptions), making it harder for lay people to
be skeptical and question a model’s output.
In the absence of a vaccine, as lockdown is eased the number of susceptible people will remain
high (although without adequate testing exactly how many cannot be known). This means there
could be sufficient susceptible people in the population for the epidemic to re-emerge, possibly
requiring another even more socially and economically damaging lockdown. In many countries
the epidemic is being controlled by the well-established test-trace-isolate procedures of
epidemiologists. Computer modeling has enabled this policy to be tested in silico and the
results suggest it will be very successful—as it has been in the countries than have
implemented it. However some countries do not have effective testing and this makes
controlling the epidemic more difficult.
http://ubiquity.acm.org
9
©2020 Association for Computing Machinery
Ubiquity, an ACM publication
October 2020
Are the Models Wrong and Dangerous?
Here is where the controversies exist. A modeler might say the parameters are based on real
data (evidence based) but that does not overcome questionable assumptions. Some policy
makers are challenging the models because they are not sure the right “evidence” is being used
or the assumptions are right. Other policy makers are making value judgments about tradeoffs
between what the models say and the social and economic costs.
Alex Berenson gives a scathing account of the Imperial College model that led to the lockdown
policy in the U.K. and the University of Washington Model that has informed policy makers in
the USA. In his view these models are indeed wrong and dangerous [15]. Sharon Begly also
criticizes the Washington University model, which she says is flawed and should not be used to
guide U.S. policies [16]. Certainly some models proposed by scientists are contradicted by the
available evidence, and some arguments made by scientists are flawed logically, empirically, or
both. Modern science can be big business and human weaknesses can overcome scientific
integrity. Consulting for the government can fund laboratories and make successful careers, but
it can also compromise objectivity.
Computational Modeling and Future COVID-19 Policies
Democracy has been described as the least worst form of government. In a similar vein,
computational modeling could be described as the least worst way of anticipating the future for
policy purposes.
At the time of this writing (September 2020) citizens and business in many countries are
desperate to end lockdown and governments are responding. In the U.K. the 2-meter
separation guidance has been replaced by a 1-meter rule suggesting that political pressure may
override evidence-based modeling. It appears politicians are diverging from their scientific
advisors and making decisions to relax the lockdown on political as well as scientific grounds.
However some question whether England began ending its lockdown in July too soon [17]. It
remains to be seen if the risks taken for political reasons pay off, or if they release new waves
of infection in the great majority of the U.K. population believed to be uninfected and
susceptible. The stakes are high. In August 2020 the USA and the U.K saw new epidemic
hotspots emerge and it is clear that the pandemic is not close to being over, as the World
Health Organization warns. In September 2020 the British Prime Minister said the U.K. is "now
seeing a second wave" of COVID-19 and that "It's been inevitable we'd see it in this country"
[18].
There are many subtle patterns of infection dynamics that could still occur, including
transmission on “star networks” as local towns and cities become infected. How can we know
whether or not this will happen? Without modeling we go into the unknown. With modeling—
assuming transparency about its methods, data, and assumptions—we have a vision of possible
http://ubiquity.acm.org
10
©2020 Association for Computing Machinery
Ubiquity, an ACM publication
October 2020
futures that can help us take the safest paths with the least risk and best chance of good
outcomes.
Managing the pandemic on the ground is very complicated. Although we have good models of
how COVID-19 is spread the number of variations for any particular case can be large. For
example, consider children returning to school. The interactions between children can be very
varied as they move through corridors between classrooms, interact in class, and travel to and
from school. Even though most schools have their own individual layouts it would be possible
to build computer models and simulate movements and infections starting from many sets of
initial conditions. Similarly, simulations of movements and interactions could reassure workers
that it is safe to go back to their city center office or inform managers that further measures are
required to make the workplace safe. Such models let us run “what if” scenarios that reveal
possible futures based on a given set of assumptions. Policy makers have to decide which
scenarios to take more seriously and implement policies that will maximize the chance of the
desired future actually occurring. The use of models to run scenarios is not the same as running
them to make predictions.
Political Leadership
In the end good policies result from good leadership. The leaders decide what to do and use the
models for advice. This is what Winston Churchill was driving at when he said, “scientists should
be on tap, not on top.”
When the pandemic unfolded many leaders were criticized for lack of preparedness. But
governments find it constitutionally challenging to spend resources on possible events that
seem remote and unlikely—there are so many urgent things to do in the present. The public
will support attending to the urgent but not preparing for the rainy day. Some scientists were
warning that pandemic danger was getting high because of international travel and the
increasing number of labs experimenting with genetically modified viruses. For example, in
2015 Bill Gates gave a TED talk warning of an imminent pandemic [19]. But such warnings could
gather no following and few governments were prepared. The ones that did the best (Taiwan
and South Korea) had already had previous close brushes with the SARS coronavirus and took
steps to be prepared for another surge. It is likely that governments now have the public
support they need to prepare for future pandemics. This preparation can be informed by
computer modeling.
Summary
Computational modeling is an essential input to policy for managing major disruptive events
such as the COVID-19 pandemic. Two major approaches to modeling have been presented:
curve fitting to macro-level models based on differential equations and agent-based modeling
http://ubiquity.acm.org
11
©2020 Association for Computing Machinery
Ubiquity, an ACM publication
October 2020
based on interactions at the micro level. The former are aggregate macro-level numerical
models while the latter disaggregate micro-level models based on hypotheses of agent
behavior within their environment. While politicians like very much the simplicity of R to make
policy arguments, experts warn about important weaknesses of calculating/estimating of this
number at the country and regional level [20].
Macro-level models are useful for providing forecasts of future numbers. Agent-based models
go further because they give insights into the space of future possibilities. The macro-level
models are susceptible to error due to externally induced phase change—for example,
lockdown completely changes the shapes of the curves. Agent-based models can model phase
changes yielding a better insights into the possible future. All these models can give poor
results, especially when the data are poor.
There is an inevitable tension between politicians and scientists since the former want simple
single answers to their questions, while the latter present many alternatives reflecting the way
that science works. Science does not present a single view but has many competing ways to
model new phenomena. The argument “we follow the science” is undermined when the
scientists can’t agree among themselves.
We believe computational models are the best we have to guide policy. Since models can give
bad forecasts, modelers should make their assumption clear (including data assumptions) so
that others can decide. Ultimately policy makers must make decisions, including which models
to accept or reject within a wider political context. Furthermore, policy decisions must take into
consideration a delicate trade off between public health and the economic and social damage
of pandemic catastrophe leading to deep recession and social unrest.
Computational modeling will be important in the recovery from the pandemic over the next
few years for meta modeling pandemics waves, economic impact, and social change. These are
highly intertwined and immensely complex problems. They are singular, history-making events
deserving better attention and the deeper understanding that computational modeling can
provide.
Further information
More detail can be found in the free online course ‘COVID-19: Pandemics, Modelling, and Policy
produced by the UNESCO UniTwin Complex Systems Digital Campus.
http://ubiquity.acm.org
12
©2020 Association for Computing Machinery
Ubiquity, an ACM publication
October 2020
References
[1] Walker, P.G.T., Whittaker, C., Watson, O. et al. The global impact of COVID-19 and strategies
for mitigation and suppression. Imperial College London (March 26, 2020). DOI:
https://doi.org/10.25561/77735.
[2] WHO Director-General’s opening remarks at the media briefing on COVID-19 - 11 March
2020. World Health Organization, March 11, 2020.
[3] World Health Authority ‘Coronavirus disease 2019 (COVID-19), Situation Report–73. World
Health Organization. April 2, 2020.
[4] Horowitz, J. Italy’s health care system groans under coronavirus a warning to the world. New
York Times. March 12, 2020.
[5] House of Commons. Statement by The Secretary of State for Health and Social Care.
Hansard, Volume 673. U.K. Parliament. March 11, 2020.
[6] Stone, J. Top scientists set up ‘shadow’ SAGE committee to advise government amid
concerns over political interference The Independent Newspaper. May 3, 2020.
[7] BBC News. Coronavirus: U.K. firms slash more than 12,000 jobs in two days. July 1, 2020,
[8] BBC News. Coronavirus: U.K. failed to stockpile crucial PPE. April 28, 2020.
[9] Solé, R. et al. Phase transitions and complex systems. Complexity 1 (1996), 13–26.
[10] San Miguel, M, Johnson, J. H., Kertesz, J., Kaski, K., Díaz-Guilera, A., MacKay, R. S., Loreto,
V., Érdi, P., and Helbing, D. Challenges in complex systems science. The European Physical
Journal Special Topics 214 (2012), 245–271.
[11] Kermack, W. O. and McKendrick, A. G.A contribution to the mathematical theory of
epidemics. Proc. R. Soc. A. 115, 772 (1927).
[12] Anderson, R. M. and May, R. M. Infectious Diseases Of Humans: Dynamics and control.
Oxford University Press, 1992.
[13] Mahajan, V., Muller, E., and Bass, F. M. Diffusion of new products: empirical
generalizations and managerial uses. Marketing Science 14, 3 (1995).
[14] FutureLearn. Pandemics, Modelling, and Policy, Step 1.7. The Kermack-McKendrick SIR
epidemic model. April 2020.
[15] Berenson, A. The failure of expert predictions and models. Hillsdale College. April 28, 2020
[16] Begly, S. Influential COVID-19 model uses flawed methods and shouldn’t guide U.S.
policies, critics say. STAT. April 17, 2020.
[17] Is England ending its lockdown too soon? The Economist. June 4, 2020.
[18] BBC News. COVID: U.K. seeing second wave, says Boris Johnson. September 18, 2020
[19] Gates, B. The next outbreak. We’re not ready. TED Talk. March 15, 2015.
[20] Adam, D. The Limits of R. Nature 583, 7816 (2020), 346–348.
Biographies
Jeffrey Johnson is Professor of Complexity Science and Design at the Open University in the
U.K., Deputy President of the UNESCO UniTwin Complex Systems Digital Campus and PastPresident of the Complex Systems Society. His research interests are in the dynamics of
http://ubiquity.acm.org
13
©2020 Association for Computing Machinery
Ubiquity, an ACM publication
October 2020
complex multilevel social and environmental systems, and in systems thinking and
computational complex systems science in policy and management. He has many years
experience of online education. He is an associate editor of ACM Ubiquity.
Peter J. Denning is past president of ACM (1980-82) and is Distinguished Professor, Chair of the
Computer Science Department, and Director of the Cebrowski Institute at the Naval
Postgraduate School in Monterey, California.
Kemal Delic is a senior visiting research fellow at the Center for Complexity and Design at the
Open University. He is the co-founder of AI-Inc, Ltd. and a lecturer at the University of Grenoble
and University of Sarajevo. He is advisor and expert evaluator to the European Commission. He
previously held positions as a senior enterprise architect and senior technologist and scientist
at Hewlett-Packard.
Dr. Jane Bromley is lecturer in Computing and Cyber Security at the Open University. Her
research is at the interface of AI and psychology, including how we learn and make machines
that learn. Her current project investigates how humans carry out difficult image classification
tasks and how machines can be made to do this.
DOI: 10.1145/3427634
http://ubiquity.acm.org
14
©2020 Association for Computing Machinery