5 minute read

Warren Graham

WHY LONG-TERM FORECASTS (ALMOST) ALWAYS FAIL - AND WHY PREDICTIVE MODELS, ARTIFICIAL INTELLIGENCE AND BIG DATA ARE NOT USEFUL...

ONE of the first things I do in the morning is to check the weather on my mobile to decide how to get dressed and what mode of transport (car or scooter) I will use. Then, I consult the same app before packing my suitcase or checking the wind on the beach in Epanomi where winter swimmers gather. Fortunately, in most short-term forecasts, the result is correct, allowing me to plan my life safely.

Wouldn’t it be possible, then, to use sophisticated algorithms to predict other future events besides the weather, as Asimov describes in the Foundation series? Why not entrust sophisticated predictions of the future to algorithms that, in combination with machine learning systems, artificial intelligence and supercomputers, will chart the possible trajectories of our lives or accurately predict the next pandemic?

Let’s first look at what is happening with weather forecasting.

Predicting the future of weather with mathematical models

Indeed, even with the limitations, we can predict the weather for the coming days with a fairly high degree of certainty. Weather prediction is a scientific achievement of the last century that originated and is still largely based on the work of the Norwegian physicists Vilhem Bjerkens (1862-1951) and Jacob (Jack) Bjerknes (1897-1975). Bjerknes, father and son, devoted their lives to observing weather phenomena and constructing the mathematical models that are still used today. Vilhem was the first to construct the theorem describing the motion of air masses, while the models were greatly improved by Jack’s groundbased observations.

By Warren Graham

WHY LONG-TERM FORECASTS (ALMOST) ALWAYS FAIL - AND WHY PREDICTIVE MODELS, ARTIFICIAL INTELLIGENCE AND BIG DATA ARE NOT USEFUL...

Bjerknes first understood and then synthesised models to predict the evolution of a seemingly chaotic system such as the atmosphere. The ability to predict the future of weather by solving mathematical equations was revolutionary, but it was not yet feasible in practice at the beginning of the last century. After Vilhem Bjerknes’ first publication in 1904, the English mathematician Lewis Fry Richardson spent three years developing the techniques for solving these mathematical equations. However, armed with only a logarithmic ruler and a logarithmic table and working on the battlefields of France during World War I, where he was part of an ambulance unit, Richardson only succeeded in predicting the change in pressure at a single point over a six-hour period. But as this calculation took six weeks, his weather forecast proved to be completely unrealistic.

However, Richardson envisioned the construction of a ‚forecasting factory’, where he estimated that 64,000 human ‚computers’, each responsible for a small part of the planet, would be needed to predict the weather. The factory, adapted from a book by Jules Verne, would be housed in a circular theatre-like hall, with galleries running around the hall and a map painted on the walls and ceiling. A conductor in the centre of the room would coordinate the calculations using coloured lights...

Although Richardson’s vision was never realised, the use of mathematics to predict the weather has developed over the years.

Photo 1. 269 “Bell telephone magazine”, 1922 (INTERNET ARCHIVE/PUBLIC DOMAIN)

Photo 2: Richardson’s weather forecasting „factory” (credit: Image courtesy of L. Bengtsson.)

NOTES:

n https://www.atlasobscura.com/articles/howa-father-and-son-helped-create-weatherforecasting-as-we-know-it.amp n https://earthobservatory.nasa.gov/features/ Bjerknes/bjerknes_3.php n https://celebrating200years.noaa.gov/ foundations/numerical_wx_pred/welcome. html n https://glowing-amaranth-camel-89. medium.com

Predicting the future, beyond the weather

Today, increased understanding of the atmosphere, along with advanced technology such as satellite data and the vast computing power available, have modernized weather maps beyond what Bjerknes could ever have imagined. It should be noted, however, that despite the technological evolution the theory behind these weather forecasts is the same, still based on the Norwegian cyclone model.

However, despite significant successes still the ability to see the future, of the weather, is limited to an extremely short-term horizon. Weather forecasting is only sufficiently accurate for the next 4-5 days, and errors or unpredictable developments are often observed even in the next 4-5 hours.

In other words, for a system such as the atmosphere, which we consider to have a sufficient understanding of how it works, and which is described by mathematical models that are constantly fed by big data collected by satellites or ground-based weather stations, and are continuously analysed in real time by powerful computers, we continue to limit ourselves to forecasts of only a few days.

So how can we possibly consider that algorithms used in more chaotic systems such as the economy, which are affected by even more unpredictable parameters such as a political decision or a new social trend, can provide any useful information about the long-term future and its evolution?

Could an algorithm in 1960 have predicted that in 1969 Neil Armstrong would step on the moon or predict the rapid growth in the US education and innovation ecosystem caused by the totally unexpected decision of one man, President Kennedy, to put NASA on the moon? In another example, what algorithm could have predicted that the self-immolation of the petty trader in Tunisia would lead to an avalanche of developments that triggered the Arab Spring?

Today, the world is more VUCA (Volatile, Uncertain, Complex, and Ambiguous) than ever before and accordingly the challenges for any kind of prediction require a new way of looking at the world and especially the development of an increased capacity to understand the system. In this direction, foresight mainly uses qualitative methods of systems analysis, while data analysis and mathematical models are only used as a supplementary tool in the process of trend identification.

Foresight provides exactly the tools we need to face the VUCA reality, to understand the system and possible changes, and finally to develop the required creativity and imagination in exploring the non-linear future.

REVIEW ROOM

This article is from: