How Algorithms Create and Prevent Fake News

What is fake news, really? Where does it come from? What are its consequences? Who or what is responsible? Why is this happening? Noah Giansiracusa’s How Algorithms Create and Prevent Fake News presents a systematic and accessible exposition of the problems, their causes and effects, and some approaches to solutions. 

 

The book begins with a discussion of Pageview, which sets up the economics of much of what follows. Further topics include deep learning and GPT-3, automated headline and news creation, Deepfakes, YouTube recommendation algorithms, biases in search algorithms, the role of automated advertising, and content moderation in social media. 

 

The complex issues are broken down and exposed in a way that the mathematically trained will appreciate but there is no formal mathematics in the book. The author does get into some optional details when he finds opportunities to get in and out quickly, but the focus is more broadly on what the algorithms do and how they are and could be used rather than on how exactly they work (much of which is proprietary anyway). There are plenty of notes and references for further investigation.

 

Actual solutions to the problems are a much taller order and the engaged reader will certainly learn the importance of finding some. The book lays out some attempted and possible avenues, often using the same technologies, but also points out the potential pitfalls. There is no avoiding political issues in this topic and the author does a good job staying above the fray, but ultimately the book must accept the existence of some sort of objective reality and that alone will bother some readers – indeed, that conundrum is a big part of why this book exists.

 

The book requires no technical background and is widely accessible. The narrative is compelling, with good use of historical and contemporary examples to help put the problems in context. We are reminded that disinformation and exploitation of confirmation bias are nothing new and it is interesting to see how many of our current issues are higher-tech versions of very old problems. For example, we can track the lineage of automated lie detection and deep fakes from their ancestors, the polygraph and simple photographic manipulation. These connections are interesting on their own and will help the less technically experienced reader understand the algorithm discussions.

 

The exposition hits a sweet spot – precise and thorough, yet brisk and not overly technical – that could make for some excellent supplemental reading for courses that look at data science, social media, technology in social sciences, or an assortment of interdisciplinary classes. The material is a snapshot of the rapidly-evolving situation right now (2021), although the general principles will remain even if some of the specifics may not age well. 

 

This book succeeds in its objective and is an easy recommendation for anyone looking to understand these issues at a deeper level than we find in the popular media. 

---

 

Bill Wood is an Associate Professor of Mathematics at the University of Northern Iowa. For now, anyway – after reading this book, he may soon look for a quiet place to ride out the collapse of society.