Software: The greatest software error in history?

In The Price of Panic, a detailed, albeit dated, overview of the world’s Covid-19 response, the authors note the use of Neil Ferguson’s ICL Covid simulator software – and its rather outrageous errors – could be classified as the most expensive software error in history.

Ferguson’s model predicted massive numbers of deaths from Covid-19 including 2.2 million in the U.S. through spring of 2020. In his 20 years as a disease modeler, each of his predictions has been off by orders of magnitude.

In the case of his model output last March, his model was the basis for adopting lock down measures in the UK, the US and other countries. Yet his model was completely wrong.

(The argument that because the countries did lock downs they did not have so many deaths is to assume that which you are trying to prove – you are assuming that lock downs work and then using that assumption to prove that lock downs work.)

The concept of a lock down was first proposed by a 14 year old high school student in a high school science fair project. This was then modeled by her Dad, who worked as a systems modeler for Sandia Laboratories. His model produced an output showing that a lock down would have saved huge numbers of lives in a previous influenza epidemic. Except, it was a model based on assumptions. Yet this model became the basis for using lock downs in this pandemic, even when research published by WHO in October 2019 noted that lock downs generally do not work and their costs exceed their benefits.

Thus, we used a flawed model from the ICL to select a lock down policy, which itself was based on a flawed model – and which is expected to have cost the world trillions of dollars, saddling all of us with debts for the remainder of our lives.

Several newer studies find that lock downs appear to have little long term benefit. In the short term, they may temporarily delay a viral outbreak. But they are not sustainable, and as the lock down gradually erodes – people get on with their lives anyway – their benefit goes away and the virus resumes.

The message of this post is that we must be far more careful and wise in our use of models. And recognize that errant model output such as that of Neil Ferguson’s ICL model may have just been the most expensive software error in history. (And needless to say, he seems to face no consequences for the errors.)