15 March 2020

Describing uncertainty – why coronavirus is neither a ‘black swan’ nor a low probability event The manuscript of Radical Uncertainty was finished well before anyone had heard of the novel Coronavirus. However, the current pandemic (and attempts to deal with it) is a good example of the radical uncertainty we discuss. I will write more soon; for now, I’m posting the relevant extracts from the book here. If you’re interested (or need something to read when you’re self-isolating!) you can buy a copy here. Unknown unknowns (pp. 38-40) At the opposite pole of uncertainty from true randomness are the genuinely unknown unknowns.  Taleb’s metaphor of the ‘black swan’ describes the unknown unknowns of business and finance, which are no less important than those of aviation.  The origin of the metaphor is that Europeans believed all swans to be white – as all European swans are – until the colonists of Australia observed black swans.  A century ago, a telephone that would fit in your pocket, take photographs, calculate the square root of a number, navigate to an unknown destination, and on which you could read any of a million novels, was not improbable.  It was just not within the scope of imagination or bounds of possibility.  Before the wheel was invented (perhaps by the Sumerians, ancient Iraqis, around 3,500 BC) no one could talk about the probability of the invention of the wheel and afterwards there was no uncertainty to discuss; the unknown unknown had become a known known.  To identify a probability of inventing the wheel is to invent the wheel.  To ask, either before or after the event, ‘what was the probability of such an event?’ is not an intelligible question.[1]   True ‘black swans’ are states of the world to which we cannot attach probabilities because we cannot conceive of these states.  The dinosaurs fell victim to an unknown unknown – even as they died they did not know what had happened to them.  Human extinction will more likely come about in another way.  Martin Rees, a Cambridge scientist and former Astronomer Royal, has founded a Centre for the Study of Existential Risk, to identify such potential threats and suggest measures to mitigate them.  He warns of the possibility of runaway climate change, pandemics, artificial intelligence and robots which run out of control.  These are threats we can at least perceive.  But the observation of a black swan was not a low probability event; it was an unimaginable event, given European knowledge of swans.  As the convict colonists boarded the First Fleet, no one would plausibly have offered, or accepted, a wager of the kind “I bet you one thousand to one all the swans in Australia are white”.  Natural phenomena are more likely than social ones to be the result of stationary processes – the structure of the physical world changes less than do global business, finance and politics.  But the impact of a pandemic is determined as much or more by the state of medical knowledge than by the pathogens of disease.  The Black Death will not recur – plague is easily cured by antibiotics (although the effectiveness of antibiotics is under threat) – and a significant outbreak of cholera in a developed country is highly unlikely.  But we must expect to be hit by an epidemic of an infectious disease resulting from a virus which does not yet exist.  To describe catastrophic pandemics, or environmental disasters, or nuclear annihilation, or our subjection to robots, in terms of probabilities, is to mislead ourselves and others.  We can talk only in terms of stories.  And when our world ends, it will likely be the result, not of some ‘long tail’ event arising from a low probability outcome from a known frequency distribution, not even from one of the contingencies hypothesised by Martin Rees and colleagues, but as a result of some contingency we have failed even to imagine.  In 1895, Lord Kelvin, one of the greatest physicists of his age, wrote that “I have not the smallest molecule of faith in aerial navigation other than ballooning or of expectation of good results from any of the trials we hear of.  So you will understand that I would not care to be a member of the Aëronautical Society.”[2]  His observation was followed in an embarrassingly short time by the first controlled flight, covering 300 yards and just under a minute long.  And today, two centuries after the First Fleet reached Botany Bay with no expectation of observing black swans, an Airbus A380 weighing 360 tonnes can carry 550 passengers over 9000 miles from England to Australia.  Something that would have been incomprehensible even one hundred years ago.  The next hundred years will be no less radically uncertain. Modelling epidemics – look for critical parameters, not predictions (pp. 375-6) The first human exposure to HIV is thought to have occurred in the 1920s.  But it was not until 1981, when unusual clusters of PCP (a rare lung infection) in 5 gay men in San Francisco were reported, that the phenomenon we know today was identified.  Tasked with creating a model to guide policymakers as to how the disease would spread and the level of intervention necessary, the WHO designed a complex model informed by the latest country-by-country demographic data.   A far simpler model was developed by mathematicians Robert May and Roy Anderson, who came up with more pessimistic projections for the spread of HIV. Unfortunately their projections proved much closer to the eventual outturns.[3]  AIDS infections accelerated across the world, causing particular harm in Southern Africa: in 1990 there were estimated to be 120,000 people living with AIDS, a number which had grown to 3.4 million by 2000.  The number of new HIV infections had risen ninefold.[4]  The world was, it seemed, a much less stable place than the WHO model had assumed. Why did the (apparently) more sophisticated WHO model fail compared to May and Anderson’s simple one?  The key factors governing the spread of disease