算法与预言

What is this book about?

The Italian network scientist Alessandro Vespignani asks whether big data and complex-network algorithms can predict epidemics, social crises, and even political upheavals. The book lays out the logic of algorithmic forecasting — how it reads macro trends out of vast numbers of small behaviours, and where it must fail. Drawing on forecasting work he took part in, on the Middle East and on epidemics, he shows both the power of these methods and their limits. A plain guide to prediction in the age of big data.

Why read it?

It pulls the word prophecy out of the occult and back into the lab: what algorithms can predict, from what data, and why they are sometimes startlingly right and sometimes wildly wrong. The author neither inflates nor dismisses; he explains the machinery of forecasting together with its uncertainty. After reading, you will be warier of both claims — that big data can do anything, and that prediction is all fraud — and better able to tell an oracle from noise.

Core Ideas

• Big data and network algorithms can read the trend of macro crises by capturing vast numbers of small behaviours.
• Social systems are full of uncertainty, so forecasting has firm limits and cannot compute a revolution or a war exactly.
• The power of prediction comes from data joined to models, and is bounded by the quality and bias of the data.
• Faced with oracle-style predictions, one should ask whether statistics or superstition lies behind them.

What questions does this book try to answer?

• Who gets to hear the oracle of the age — and who controls algorithmic forecasting?
• Can algorithms really compute the outbreak of a revolution, a pandemic, or a war?

Who should read it?

For readers drawn to data science, epidemiology, and social forecasting, and for anyone who wants to understand what big-data prediction really is. It asks for some patience with statistics and networks; if you want formulas or a programming tutorial, this is not that.

How to Read It

The author works in complex networks and computational epidemiology, and most cases come from forecasting projects he took part in — reliable, but seen from a researcher's own angle. The book is written for general readers with little mathematics and can be read chapter by chapter. A caution: the sections on epidemic forecasting read differently now, and their cases and figures are dated — do not apply those conclusions directly to today. A little statistics helps.