Nearly 10 years ago I published a paper in the history of science journal Osiris titled Reducing the Future to Climate: A Story of Climate Determinism and Reductionism.  It was a warning about viewing the future—and making decisions about the future—solely through the single lens of climate, in particular relying on predictions of climate futures emanating from models.  I believe similar caution is needed when it comes to present decision-making about dealing with Covid-19.  Good policy-making is rooted in wise human judgements, not in reading-off answers from a model.  Achieving this requires taking into account many more considerations about the future than only those offered by models.

Models are valuable tools to assist our thinking.  They are also alluring.  They are created by some of the cleverest people and often rely on some of the most advanced monitoring or simulation technologies available to us.  They appear to offer authoritative and quantified predictions of the future.  This is as true for climate change as it is for a pandemic.

In recent weeks we have seen many arguments (for example, here, here and here) about the value and role of epidemiological models in shaping and guiding public health policy.  These discussions often portray the issue of model choice as being a beauty contest (which model is the best), or a performance event (which modeller has the proven track record) or a power struggle (which modelling group has the ear of government).

Much more important than these questions however is the need to understand what a model is.   Instead of being impressed by the style, performance or influence of the model, it is important to understand what is inside the ‘black-box’–or rather what is left out.

Models by definition are simplified representations of reality.  They are designed with some goals in mind and not others.  Some physical processes are simulated, others are not.  Some simulated processes are guided by good data, others by expert assumptions or judgements.  For all sorts of good reasons, many important processes that bear on the question at hand—how will the pandemic spread? how will the climate change?–are left out. 

This means that model predictions of the future are always partial and conditional.  They may offer precision, but rarely are they accurate.  When a model’s prediction of ‘the future’ is over-weighted in decision-making, the future being managed has been reduced in dimensionality according to the design exclusions of the model in question.  This was the argument of my 2011 Osiris article: climate model predictions of the future dangerously narrow our field of view–just because they are predictions of future climate and not predictions of ‘the future’.

And it is true now of epidemiological models of Covid-19.  These models are partial and become dangerous if given too much weight in decision-making.  As Devi Sridhar and Maimuna Majumder have recently argued in the British Medical Journal, “all models are limited by the assumptions they make” and they warn that epidemiological modelling results “should be taken as just one input among many, one piece of a large puzzle”. 

Failing to heed this warning reduces the future to Covid-19—or to climate.  The policies that follow such reductionism mistake the model world for the real world.  They fail to accommodate the limitations and biases introduced by modelling, a point made recently by Ludovic Touzé-Peiffer and colleagues with respect to climate models.  The complex web of interacting processes, reflexive behaviours and competing values upon which future individual and collective well-being are based is mis-represented.

In the case of Covid-19, epidemiological models exclude the diversity of human behavioural responses to a virus or to policy measures.  They do not simulate the effect of a policy on cancer patients who are denied operations or on people’s decisions not to frequent A&E for emergency treatment; they do not simulate the differential effects of sustained lockdowns on domestic abuse, broader public mental and physical health or the economic and social consequences of businesses collapsing and of an economic depression; they do not simulate the effects of increased surveillance on cultural norms and democratic values.  And the list goes on. 

A reliance on a one-eyed model will lead to a one-eyed policy.  As I have argued elsewhere with respect of climate change, simply basing climate policies on securing a one-dimensional target like global temperature (or net-zero carbon emissions) would only make sense if one believed the future can be reduced to climate.  Future human and planetary well-being relies on many more factors than global temperature, factors which climate models do not simulate.

Similarly, basing public health policies on a one-dimensional target like the numbers of lives lost to Covid-19 would only make sense if this was believed to be the sole criterion of success.  Just because models can (crudely) simulate such a number is no reason to base policy upon it–any more than it is sensible to base climate policies on securing global temperature just because climate models can (crudely) simulate it.

All significant and important decisions—whether taken privately in the home or collectively in public life—are multi-dimensional.  They involve a multitude of (mostly unpredictable) factors.  And ultimately they require difficult trade-offs which in most cases revolve around challenging ethical judgements.  This is what we each do in our everyday lives.  And this is what politicians are paid to do when crises erupt.

We should not hand this task over to models.  Or even to a range of models.  Or even to a range of experts who try to interpret a range of models.  Or even to elected politicians if those politicians are attracted by the allure of those experts’ interpretations of a range of models. 

Decision-makers need a more diverse range of inputs and voices.  They need to consider many more pieces of “the puzzle” of the future, especially pieces that do not look or speak like ‘a model’.  Whether they are trying to predict the future of a pandemic or the future of climate, models offer us only one-eyed views of the future. 

We have heard a lot in recent weeks about politicians being led or “guided by science”.  And in the debates about climate policy we often hear pleas from advocates for governments to “listen to the scientists”, to ‘follow the science’.  No.  If such science, or if such scientists, merely interpret the reductionist models that their communities develop, public policy-making should not be led by them.  The future should not be handed over to models.

Mike Hulme, University of Cambridge, 27 April 2020