This cacophony of almost random forecasts isn’t unique to public commentary on viruses – almost any area of national debate is drowning in confident predictions, very few of which have much value. Most of those you read suffer from one of two problems: the forecaster probably isn’t telling you what they really think, and even if they were, what they think might not be any good.

Accuracy, even among professional forecasters like intelligence analysts, has often been a target rather than an achievement. So anyone trying to work out who to listen to in order to understand the future always had their work cut out for them.

That’s why in 2011 the US Intelligence Advanced Research Projects Activity agency, IARPA, invested in a massive forecasting competition to try and make concrete improvements. Teams would compete to forecast hundreds of topics over four years, from a range of subjects in different countries, with the winning strategies studied and shared with the wider intelligence world.

The bestselling book Superforecasting went on to describe the techniques used by the winning team (of which I was a member) in that competition. We didn’t get everything right, but we learned that it’s possible to do a lot better than foreign policy experts by moving away from the traditional methods employed.

There were three key structural lessons from the competition:

1. Keep score
2. Identify the best forecasters
3. Work as a team

The most important item on this list is keep score. Most forecasts that we come across in the media are made by in-house pundits or people rung up by television stations or newspapers for a quote, and these aren’t scored – there’s no penalty for being wrong most of the time. Old articles are rarely revisited and if they are, the pundit can normally wave away criticism and continue along the same line. Some pundits have made a career out of political analysis despite consistently being wrong in their predictions.

If there’s no reward for getting things right, then accuracy will drop down the list of incentives behind the forecasts. Instead pundits will say what they think will cheer people up, what will promote their career via wacky, attention-grabbing predictions, what shows loyalty to their friends or signals political values and ideals to others — or any of the million other reasons someone might predict something without actually expecting it to happen.

It’s not just that pundits and others in the public eye are bad at predicting, they’re mostly not even trying to. And these predictions can influence politicians and decision makers.

The way to build a successful forecasting system is to reward people for accurate forecasts and disincentivise bad ones. Just that little step, of keeping score and letting everyone know at the start that what they say will matter, immediately changes everything. Now when someone is tempted to predict great success for a project led by their pal, they might think about how embarrassed they’ll be if it goes wrong, and drop their confidence.

More from this author

How the experts messed up on Covid

By Stuart Ritchie and Michael Story

Once people are actually trying to predict accurately, you can move on to step two — feedback. Scorekeeping doesn’t just incentivise people to be honest, it also gives them the opportunity to improve. Once you know what you’ve got right and got wrong, you can go back and review. As forecasts resolve, people tend to get better calibrated – that is, there develops a closer relationship between their confidence in their forecast and the outcome.

And once you have a regular run of forecasts, you can take one of the most important steps — stop listening to people who can’t or won’t predict accurately, and pay attention to the most accurate team members, the “superforecasters”.

One of our most striking findings was that the ability to forecast is a skill in itself, independent from subject matter knowledge. Successful forecasters aren’t perfect, of course, but a good track record of predictions is a strong indicator of future performance. And with these skills, superforecasters work for banks, governments and other organisations providing probability estimates for all kinds of risks as well as training analysts in forecasting.

In a system where forecasters are punished for getting it wrong and rewarded for getting it right, the most accurate people float to the top, and when they work together on a team, they become even more accurate.

The forecasts

Last week we gathered a small group of well-calibrated forecasters with good track records to look at the UK’s pandemic outlook over the next few months. With Britain facing both the worst period of the crisis but also an exit strategy with the vaccine roll-out, “when will this end?” is the most urgent question everyone is asking right now.

We can’t promise clairvoyance, but we made the best use of our team’s skills and produced the following set of estimates on three topics:

• How bad the daily death toll will get
• How many vaccines we’ll have administered at the end of February
• When daily deaths will first fall below 100

Superforecasters cannot predict exactly when the pandemic will be over, but these three questions act as proxies for the questions of how bad it will get, how soon an individual is likely to get a vaccine, and when the pandemic will recede. A double-digit daily death toll does not mean the end of the nightmare, but with large-scale vaccinations taking place it will certainly mean is in sight.

Suggested reading

When will the Covid-19 vaccine arrive?

By Saloni Dattani

These forecasts are presented as the median estimate of the group; the principle of the “wisdom of crowds”, the idea first suggested by Francis Galton that the median of any large number of guesses will usually be fairly accurate, applies to experienced forecasters as much as anyone.

We also include our “80% confidence” interval as an expression of the uncertainty surrounding our forecast. For example, our central estimate that the death toll will peak at 1,278 with an 80% confidence interval of 892 to 5,750, means we think there’s only a 10% chance it’ll be less than 892 and only a 10% chance it’ll be greater than 5750. (The confidence intervals are represented by the blue rectangles, and the central estimates with the red lines.)

The worst of the pandemic

The median estimate of the group was for the 7-day average of daily deaths to peak at 1,278, but with an 80% confidence interval of 892 — 5,750

Alongside the predictions, we note down anonymous comments from the forecasters in the team to explain their reasoning. Among the notes accompanying this question, perhaps the most optimistic was that there is a “small chance we are at the peak now”.

On the other hand, the worst-case scenario is if the hospital system collapses and we are forced to “en mass palliate anyone over 60/65” and the most pessimistic predicted 5,000 deaths a day. The big factors were new virus strains and “poorly managed vaccination centres causing infections among the most vulnerable and highest priority population”.

However, with 160,000 infections a day currently in the UK and a fatality rate just below 1%, 1,400 deaths a day seems likely.

Another forecaster also predicted that a third wave was “likely” once the current lockdown ended, “with quite possibly the highest caseload yet, but [was] unlikely to have as many fatalities due to prioritised vaccination of elderly and otherwise-vulnerable” individuals.

A lot of this depends on whether the vaccine protects against any new strains, and how quickly those vaccines are rolled out.

The February Vaccine Target 

Millions of people are currently waiting for a letter informing them of their vaccination appointment, a jab that will end a year of house arrest and often extreme anxiety. The Prime Minister has promised two million jabs a week, and tens of thousands of lives, and millions of livelihoods, depend on getting the jabs out as soon as possible.

We asked the team to forecast how many vaccine doses will have been administered in the UK by the end of February and the central estimate was 12 million, with an 80% confidence interval of 6 million to 20 million.

Among the big questions are the supply and how much political pressure there will be for the distribution network to move faster. Two more vaccines, by Novavax and Johnson & Johnson, are also expected to get approval next month and that will also increase supply.

One forecaster predicted that “Boris’s current target was 13 million for end of February: best guess is they undershoot that slightly and ramp-up only really gets going in March/April…. this has been the consistent pattern of undershooting a bit throughout the pandemic.”

So with a median estimate of 12 million, you can find out when you and your loved ones are likely to receive theirs here.

Suggested reading

Inside the Covid ward

By Jane Smith

In which month will deaths fall below 100 a day?

By estimating when deaths fall below 100 a day we can get some idea of when the virus is close to beaten.

Our group estimated April as the most likely month for daily deaths to fall below this level, with an 80% confidence interval of March to August.

Among the key unknowns are new strains, and also whether hospitals get overwhelmed “which results in cascading preventable deaths”. We can estimate when deaths will fall based on first wave patterns, when it took “two and a half months to get from 1k per day to below 100 in the March-July lockdown”, although the vaccination programme might accelerate that. Generally speaking, it takes about three weeks to half the deaths, and so following an expected peak in January, death rates will half four times by late April.

At this point, one forecaster suggests, the government will open up the economy and deaths will continue around 100 a day because this will be regarded as tolerable. Another believes this figure may continue until September, by which time a combination of vaccination and infection will have introduced herd immunity.

The most “optimistic” forecast suggested that we may have got deaths down by March but that might be because “we screw up” and “have killed everyone who could possibly die from it and therefore the virus has nowhere else to go”.

The most pessimistic suggested: “If there is a less successful vaccination programme, multiple policy errors such as keeping schools open once teachers have been vaccinated but not paying attention to cases/repeated attempted opening up, or mutations of the virus, then it might be December before deaths get below this level.”

But another thought that new drugs — the most recent breakthroughs came late last week — will have reduced the fatality rate fairly soon.

January and February 2021 are going to be very traumatic months for tens of thousands of people in the UK. But to be forewarned is to be forearmed, and if you’re worried about the next few weeks you might want to gain a better estimate of how bad it will get, when your vaccine will come, and most importantly when the nightmare will be over.

And perhaps next time that Britain faces a crisis of this magnitude, we’ll have systems in place that incentivise people to tell the truth, identifying those with strong track records of forecasting, rather than turning to those who tell us what we want to hear.

Back in March, pundits and experts were not hopeful: some were confident that it would take over 18 months, some thought at least two years, and others predicted several years. Five months on, with over 165 vaccines in development, Dr Anthony Fauci, the American epidemiologist advisor to the White House, says that an effective vaccine will be available in the US early or midway next year. And superforecasters – the paid experts whose job it is to successfully predict events – tend to agree.

Are we really on the verge of achieving something that was previously inconceivable: a vaccine for an entirely new disease, developed within a year? And who can make that prediction? On the one hand, there are scientists, who have deep knowledge of viruses, infectious disease, immunology, epidemiology and biology — but they may be unfamiliar with the global economic and political events that could influence vaccine development in a time of crisis.

On the other, there are superforecasters, who have broad knowledge of global issues, and skills and experience in their approach to making forecasts — but they may be unaware of certain quirks of viruses or the immune system that influence how difficult or easy it may be to develop an effective vaccine.

When estimating a vaccine date, the best approach might be to combine the knowledge and skills of both groups. But before that, we need to define the question we want to answer precisely. Do we want to know how long it will be until a Covid-19 vaccine is approved (for anyone? for the general public?). What about the vaccines that have been authorised in other countries, such as China and Russia? Perhaps we want to know how long it will be until there are a certain number of doses of vaccines that have been approved in the US and the UK? All of these questions might be interesting, but for the purposes of this piece, the last one may be most relevant.

At this point, it’s important to find an accurate ballpark estimate to work from, and it’s easy to be swayed by the predictions of other experts and media reports. As Philip Tetlock and Dan Gardner explain in their book Superforecasting: “When we make estimates, we tend to start with some number and adjust. The number we start with is called the anchor. It is important because we typically underadjust, which means a bad anchor can easily produce a bad estimate.” This effect, called anchoring bias, has been demonstrated in many large experiments.

Vaccines normally take a long time to develop, usually between 12-15 years from research to trials to approval and to distribution, and at each stage, the probability that any given vaccine candidate will make it through drops until only a few candidates, if any, actually succeed.

But today things are moving faster than ever before. This is partly because of improvements in biotechnology (such as cheaper genome-sequencing), computational speed, open data sharing, and preprints. But more significantly, the gravity and urgency of the situation have put the pandemic at the centre of attention, and there is more money for Covid-19 vaccines than for any previous types.

On top of this, scientists already have research from other similar coronaviruses (MERS and SARS), and some pharmaceutical companies already work with animal models of the disease and have factories to produce vaccines for their trials. What’s more, researchers don’t have to go through each phase of trials one-by-one, as they would usually, but can start multiple phases simultaneously to speed up the process.

More from this author

Why the Government changed tack on Covid-19

By Saloni Dattani

Some pharmaceutical companies have even started manufacturing some vaccines at scale so that they can be distributed swiftly, anticipating that they might be approved. The approval process, during which regulators scrutinise the evidence and inspect factories to find out how safe and effective a vaccine actually is, typically takes an additional year, but regulators have said this will be fast-tracked and begin while trials are ongoing. This means that a timeline that would usually take 12 years could be accelerated to 18 months or less.

And some previous global pandemics offer hope. In the spring of 1957, an American microbiologist called Maurice Hilleman, after reading a story in the newspaper, concluded that a deadly influenza epidemic in Asia could become a global pandemic. He sought out samples of the virus, and found it was a strain that most Americans had not encountered before, and therefore had no immunity to. Hilleman worked personally with vaccine manufacturers to scale up production, so successfully that the vaccine was trialed in patients in July that year; by the autumn, factories were ready with 40 million doses. This whirlwind speed massively reduced the impact of the Asian flu pandemic in the US and abroad – a timeline of only 4 or 5 months.

With other pandemics, though, we have not been so fortunate. During the Spanish flu of 1918, researchers were unable to identify the pathogen that caused the disease, while doctors often misdiagnosed patients, all of which interfered with the development of a vaccine. More recently, the HIV pandemic remains without an effective vaccine, decades since the virus was discovered.

Clearly, not all vaccines have turned out the same way — so how do we know whether a vaccine for Covid-19 can be developed reliably or not at all?

Fundamentally, an effective vaccine is a faithful representation of a part or all of the natural virus. It is administered by an injection, for example, and is absorbed by immune cells, which transport it to lymph nodes around the body. Here, little bits of the absorbed vaccine are shown to many immune cells, such as T-cells and B-cells, and some of these will be activated (if they match those bits of absorbed virus carefully). These cells will quickly multiply, producing antibodies and “memory cells” that can linger in the body for many years.

Later on, if the person is infected by the natural virus, a similar procedure occurs. The virus is absorbed by immune cells and transported to the lymph nodes, where the old memory cells are reactivated. These can quickly multiply and produce antibodies again, to fight off the virus before it can cause disease. Effective vaccines typically do not stop a recipient from being infected altogether, but they prevent a recipient from developing the disease caused by the virus.

Understanding the immune response to a virus can help direct development of the vaccine. In the case of Covid-19, there are several pieces of information we can make use of.

First, the way the virus mutates can tell us whether a vaccine will activate the same memory cells that can destroy the natural virus. If there are many strains of the natural virus, which appear very different to the immune system, the possibility that memory cells from the vaccine can recognise and destroy the virus becomes less likely. Effective vaccines for the coronavirus may be easier to develop than vaccines for many other viruses, because its mutation rate is fairly slow – thanks to special enzymes in coronaviruses (called exoribonucleases), which most viruses don’t have, and which proofread new mutations and fix most of them.

Second, the way the immune system usually responds to the virus can tell us whether the virus can be cleared from the body at all. A pessimistic scenario is one where the virus infects humans, finds cells to target quickly, and uses them to multiply very rapidly in the body, accumulating countless new mutations before immune cells can absorb it and take it to the lymph nodes. That is the case for HIV, which patients cannot clear from their own bodies, and for which a safe and effective vaccine has eluded scientists for decades. Even though there is still a great deal we are learning about the disease, this pessimistic scenario is less likely for Covid-19 because most patients do seem to be able to clear the virus from their body and recover from the disease.

These factors bode well for the prospects of an effective vaccine, but they are not all. We knew much of this in March, so why was it that even in late April, forecasters said there was only a 5% probability that a vaccine would be approved in the US before April 2021, with enough doses to vaccinate 25 million people, but they now believe that probability is 49%?

The fact that forecasts can shift so much so soon is more of a feature of forecasting than a bug. Many events that we forecast depend on whether other events happen, so just as my chances of being hospitalised are far greater if I’ve just fallen off a tree than if it was any random day, the approval of a vaccine becomes more and more likely if the vaccine succeeds in each phase of trials. When unlikely events come true, they make other events that are tied to them more likely.

Happily, the events of the last few months mean we can predict with some more confidence whether the trials will pan out as planned, and if the vaccines will be effective and safe.

More from this author

Why are minorities so hard hit by Covid-19?

By Saloni Dattani

Often, large obstacles in vaccine development have come during the undertaking of trials. During phase III trials, participants are randomly given a vaccine or placebo, and researchers note the difference in the chances of disease between the two groups. But that difference would be undetectable if the outbreak was contained or sporadic, because few participants would be exposed to the virus at all.

We now know that is unlikely to be a problem for Covid-19, since the pandemic is still pervasive and participants are being recruited in hotspots around the world as they develop. Many thousands of people have already joined for the phase III trials of many frontrunning vaccines and recruitment is almost complete for Astrazeneca’s trial.

Even if a trial were feasible, though, it’s not guaranteed that a vaccine candidate would be effective or safe. It will be months until we know if the frontrunner vaccines will actually prevent the disease, when results from phase III trials become available. So far, we can only base our predictions on the data that is available from animal research and some phase I and II trials, which tell us whether the vaccines have triggered some immune response and what their side effects were, for a limited demographic.

When it comes to Covid-19 vaccines, some reviewers and experts (such as Hilda Bastian, Florian Krammer and Derek Lowe) have written detailed summaries of the data from phase I and II trials so far, and the results are broadly encouraging. The frontrunner vaccines trigger an immune response, people produce working antibodies and T-cells to them, and don’t exhibit major side effects.

But there are still caveats: these trials were very small, most vaccines required boosters, and mild side effects were common. Since the participants were mostly young and healthy, that could become a problem when the vaccine is then tested in a wider demographic. Fortunately, some vaccines, such as by Pfizer and Moderna, have been tested in elderly people, and they did not show significant side effects. Still, it is difficult to compare the results of phase I and II trials between different vaccines, because the numbers of participants were low and the doses of each vaccine varied in each trial.

But, judging from the results so far, we can rule out some worst-case scenarios: that vaccines wouldn’t trigger any immune response at all, or that they would frequently cause serious side effects. The results so far provide some cautious optimism about the prospects of these vaccines; but squeezing much further interpretation out of this data, which is limited and preliminary, is probably premature.

Having examined all this evidence, how do we put it together to answer the original question: how long will it be until there are enough doses of an approved vaccine for Covid-19 for 25 million people (in the US)?

Useful forecasts tend to not give single specific estimates of when events might occur (such as simply saying they will happen in December), but give a series of predictions. Even though some time periods will be most likely, we would also want to find out when things will occur if not then. So a useful exercise may be to imagine three forecasts – an optimistic timeline, a pessimistic timeline, and in-between “Goldilocks” timeline.

Here’s how those might play out.

In an optimistic timeline, a majority (say 60%) of the early vaccine candidates are on track to succeed. Even though this likelihood is far higher than the average proportion of all trialled drugs that are approved over all phases of trials (around 10%), the biology of the virus and promising results so far suggest that this anchor should be updated. Moreover, the pandemic continues to be pervasive, so a smaller number of participants need to be recruited to detect the effect of a vaccine, and phase III trials take much less time than usual.

Suggested reading

The ZeroCovid debate: can the disease be eliminated?

By Freddie Sayers

So perhaps one or two of the vaccines that started phase III trials very early (Astrazeneca, Pfizer or Moderna) are demonstrated to be safe and effective in key arms of the trials. And owing to demand and perhaps political pressure, they are authorised for emergency use soon after the companies stated they would submit applications (between September and December). Vaccine manufacturers quickly manage to produce tens to hundreds of millions of doses of these vaccines as they have claimed they will have by the end of the year. And the approvals process takes only weeks to a couple of months, because of rolling reviews.

This would place the date that 25 million doses of approved vaccines are available in the US sometime around January 2021. Although not everything went perfectly in this timeline and lots of vaccines failed, most things went well for at least one or two early vaccines. What would happen if they didn’t?

In a pessimistic timeline, only a minority (say 10%) of early vaccine candidates will succeed. It turns out that several of the vaccine candidates show severe side effects in the elderly in phase III trials. A few of the vaccines, alarmingly, also cause rare side effects (in trials or after emergency use authorisation). Pharmaceutical companies and regulators decide to scrutinise vaccine applications for much longer than in the optimistic timeline, and err far more on the side of caution until they are unequivocally shown to be safe.

Simultaneously, there are geopolitical setbacks or disasters that affect manufacturing and transport, or perhaps far larger doses are required than anticipated. Even more unluckily, the vaccines that do eventually get approved turn out to be the ones that are more difficult to manufacture at scale.

This would place the 25 million doses date sometime around June 2021, if not later. Distressingly, lots of things went wrong in this timeline, but they were still somewhat realistic. Fortunately, because so many vaccines were in trials and vaccine funding was at its highest level ever, at least some vaccines succeeded less than two years after the disease was identified, but this was still much later than people had hoped.

Now we should synthesise the two scenarios, but striking a balance between them could be difficult. Perhaps we can pick apart the chances of each of the events in both scenarios to try to figure out the most likely timeline. Here is my thinking:

So far, the speed of trials and dose manufacturing has been much faster than anticipated, the preliminary results are quite promising, there are very many different types of vaccines in trials, and the ongoing pandemic means it will be unusually easy to recruit enough participants into trials. The biology suggests that an effective vaccine is very achievable, the worst-case side effects are likely to be rare, and the urgency has already sped up the approvals procedure. At the same time, low efficacy and mild-to-moderate side effects could be a problem for several of the vaccines in trials, and unforeseen setbacks in manufacturing or approvals will probably occur for some of the vaccines.

Suggested reading

Covid could become the new common cold

By Tom Chivers

My view, then, is somewhere between these scenarios, with the most likely outcome being that vaccines will be approved and available in that quantity around February 2021, given what we know or can estimate so far.

Note that this doesn’t mean it would take until then for the public to receive Covid-19 vaccines. If a vaccine is cleared for emergency use authorisation, it could be prescribed to treat or prevent infections as a sort of experimental drug, before it might be approved.

The chances that no vaccine is approved before 2022 is probably fairly low, given the large number of vaccines in trials, but it also includes scenarios where vaccines would take many years to be approved, because of unforeseen or rare events. All in all, these are the chances that I would place on vaccines becoming approved and available in those large quantities in each time period: before 1 October 2020 (1%), between 1 October 2020 and 31 March 2021 (58%), between 1 April 2021 and 30 September 2021 (32%), between 1 October 2021 and 31 March 2022 (6%), not before 1 April 2022 (3%).

With a concrete set of predictions, then, this finally brings us to the end of the forecast, for the time-being. Events of the coming months, including results from phase III trials, will mean we can revise and refine our predictions; but many questions will remain.

How many vaccines will be approved? How effective will the first-approved vaccines be? Will people need to take boosters or combinations of vaccines? How will the vaccines be distributed? How long will it take to achieve herd immunity with vaccination? How long will it take for the disease to be eradicated globally; and is that even possible in our lifetimes?

The momentous day a vaccine arrives will feel revolutionary – a triumph of modern science over suffering. But with all the questions that remain and all that remains to be done, history suggests that it will be just one more hard-won battle in humanity’s long war against infectious disease.

Richard Dawkins does not have the same rule, I think it’s fair to say. “It’s one thing to deplore eugenics on ideological, political, moral grounds,” he tweeted recently. “It’s quite another to conclude that it wouldn’t work in practice.”

Lots of people got very cross about that, and I was naively surprised; surprised enough to break my own rule. I thought he was saying something fairly obvious, which is that if you bred humans for some trait then it would probably make that trait more common, and that he was explicit in saying he wasn’t talking about whether it was morally acceptable. But other people thought — they told me so — that he was literally in favour of eugenics. (Dawkins later clarified that no, he was not: “I deplore the idea of a eugenic policy [but just] as we breed cows to yield more milk, we could breed humans to run faster or jump higher. Heaven forbid that we do it.”)

The analyst John Nerst, who writes a fascinating blog called “Everything Studies”, is very interested in how and why we disagree. And one thing he says is that for a certain kind of nerdy, “rational” thinker, there is a magic ritual you can perform. You say “By X, I don’t mean Y.”

Having performed that ritual, you ward off the evil spirits. You isolate the thing you’re talking about from all the concepts attached to it. So you can say things like “if we accept that IQ is heritable, then”, and so on, following the implications of the hypothetical without endorsing them. Nerst uses the term “decoupling”, and says that some people are “high-decouplers”, who are comfortable separating and isolating ideas like that.

More from this author

How daughters turn fathers into feminists

By Tom Chivers

Other people are low-decouplers, who see ideas as inextricable from their contexts. For them, the ritual lacks magic power. You say “By X, I don’t mean Y,” but when you say X, they will still hear Y. The context in which Nerst was discussing it was a big row that broke out a year or two ago between Ezra Klein and Sam Harris after Harris interviewed Charles Murray about race and IQ.

As a high-decoupler, Harris thought that it was OK to talk about what-ifs; if there are genetic components to racial differences, then we still need to treat everyone with equal dignity, etc: “I’m not saying there are, but if there are…” He thought he’d performed the ritual.

But for Klein, the editor of Vox, the ritual was not strong enough. Murray’s ideas are reminiscent of a grim history, in which pseudoscientific ideas about a hierarchy of humans were used to justify slavery or Jim Crow laws. For Klein (a low-decoupler, in Nerst’s taxonomy), you can’t simply take an idea out of its context like that. The context comes with it.

These two paradigms are very hard to square. Harris thought he was having a coolly rational debate in the philosophy-seminar style, so was baffled to find he was being accused of racism; Klein thought Harris was trying to sneak racist ideas in under an academic smokescreen, and couldn’t believe Harris claiming otherwise. Their models of the world were so different they just couldn’t understand each other. So obviously it descended into a massive online row with accusations of bad faith and racism.

That’s what I think was going on with the Dawkins tweet. Dawkins thought he’d performed the magic ritual – “It’s one thing to deplore eugenics on ideological, political, moral grounds. It’s quite another to conclude that it wouldn’t work in practice” =  “By X, I don’t mean Y.” He is a nerdy, high-decoupling person, a scientist, used to taking concepts apart.

But many people reading it are not high-decouplers; they hear “eugenics” and “work” and immediately all of the history, from Francis Galton to Josef Mengele, is brought into the discussion: you can’t separate the one from the other. 

More from this author

Could the Cummings nerd army fix broken Britain?

By Tom Chivers

I think a lot of arguments in society come down to this high-decoupler/low-decoupler difference. And while I hope I’ve done a good job of putting the case for low-decoupling, I am very obviously a high-decoupler, so often I find myself thinking “but they performed the magic ritual! They said they didn’t mean Y!” and being really confused that everyone is very angry that they believe Y.

Dawkins, to my knowledge, never explained why he suddenly brought up eugenics out of a clear blue sky, but the word is in the news at the moment because Dominic Cummings hired the weirdo he wanted to hire, a man called Andrew Sabisky. Inevitably enough, the media has gone through his old social media posts and found various things he’s said, and he has since quit.

For the record: some of it is genuinely unsettling, for instance that “there are excellent reasons to think the very real racial differences in intelligence are significantly — even mostly — genetic in origin”. This is, I think, not true — I don’t know how you square it with the simple fact that poor black boys in the UK get better educational outcomes than poor white boys, for instance. And none of this should be seen as a defence of Sabisky’s hiring. (Watch me perform the decoupling ritual: will it work?)

But I thought a few of them were interesting from this high/low decoupling point of view. For instance, he described female genital mutilation (FGM) as a “moral panic”, because the actual numbers were low, so the risk to girls in the UK was pretty tiny. The full quote was “It is still unclear to what extent FGM represents a serious risk to young girls, raised in the UK, of certain minority group origins. Much of the hue and cry looks more like a moral panic.”

It seems fairly obvious to me that Sabisky is saying that FGM is rare in the UK – a view, incidentally, shared by at least one consultant obstetrician in the BMJ, who said “there is no evidence that large numbers of girls living in UK are having FGM”. He’s not belittling the impact itself.

Similarly, I’ve said before that claims about a wave of teen suicides are “absolute bollocks” and that the “epidemic of loneliness” is not real; that doesn’t mean I don’t think that suicide and loneliness are terrible things. You can decouple the frequency of the event from the seriousness of the event when it does happen. But for low-decoupling people, you can’t; by downplaying one, you are downplaying the whole package.

Similarly, Sabisky said that “women’s sport is more comparable to the Paralympics than it is to men’s”. But the context was that of Caster Semenya, a female runner who has an unusually high level of blood testosterone, and whether she should be allowed to compete in women’s sport. Sabisky is suggesting, as I understand it, that there is a need to draw boundaries around women’s sport in a way that you don’t around men’s, but you do around the Paralympics; if anyone could compete in women’s sports or the Paralympics, then lots of mediocre able-bodied men would win them and there would be no women’s sports or Paralympics left. The concept of “boundaries around the sport” has been separated from the idea of the sports themselves. But to a low-decoupler, it sounds, I think, as though he is saying they are inferior. 

Suggested reading

Test article

By Graeme Archer

Even his “eugenics” argument, in an interview with Laura McInerny in SchoolsWeek in 2016, strikes me as quite a separating-the-feelings-from-the-facts, decoupling-style situation. In fact he even says so, explicitly: “You have to separate yourself from what you feel and from what are the facts of the matter.” He asks whether parents should be able to screen IVF embryos, whether for disability or schizophrenia, or for intelligence. The former is not a million miles away from our current situation, where pregnancies at high risk of chromosomal disorders can be terminated if the mother wishes; the latter will probably be happening in China in a few years. The fact is that it’s possible; that doesn’t say whether it’s right. 

Sabisky himself, according to that interview, seems to be against it: “He is personally uneasy about the ethics of many of the things he talks about and as a Christian — he married in church last year — he has moral views on the topics that may not be what people expect.” It seems to me that he is explicitly decoupling what he thinks is true, or what he thinks is effective, from what he thinks is right.

I don’t think, as some people do, that these remarks have been “taken out of context”, as such. I think that even with the context, lots of people would still assume that when he says “FGM isn’t a major risk” he means “we don’t need to care about FGM”. Similarly, Dawkins’s tweet wasn’t taken out of context: the context was all there, but the magic ritual didn’t work. It’s a translation problem: some people think they’re having a cold, rational discussion; other people are alive to the implications; there will be frequent occasions when the two groups will hear the same words and yet understand totally different things by them.

Again, to perform the decoupling-ritual one more time, none of this is to defend the hiring of Sabisky — who, full disclosure, I met in 2017 and interviewed for my book. It wouldn’t surprise me, to decouple again, if he was a useful person to have around government: he is a “superforecaster”, which means that of the 20,000 people making predictions about the future for the Good Judgment Project, he was one of the 150 who scored best for accuracy. (Read a bit more about that here, if you like.) I suspect having people who are really good at predicting the future is useful, but that is decoupling his abilities from his views, and maybe that’s not a good thing to do.

But I think the decoupling thing makes me understand a bit more why Dawkins’s tweet got people so angry. Sometimes the ritual fails, and the spirits break through the warding circle.