Hello World: Being Human in the Age of Algorithms

By Hannah Fry, W.W.Norton & Co. 2018, 246 pages, ISBN 978-0-393-63499-0, $25.95 hardcover.

This book is about computer algorithms that, according to the author, have a large and growing influence on nearly all aspects of our lives. Hannah Fry is Professor of the Mathematics of Cities at University College in London and also a regular presenter for the BBC. Hello World covers algorithms as relatively straightforward as the statistical analysis of crime rates and as powerful as the machine learning techniques used to play chess, make medical diagnoses, and more. The book’s emphasis is on what these algorithms are designed to do, how well they achieve their goals, and the effect of these increasingly invasive algorithms on our lives. There is minimal technical discussion about the mathematics and programing of computers. The book is well written and easily accessible to non-technical readers.

One of the book’s strengths is its wide range and number of topics. Chapter one is a general discussion of how ubiquitous these algorithms have become and the frighteningly important role they sometimes play in critical decisions, such as when to launch a retaliatory nuclear missile attack. The author then turns to a very different topic: the enormous power people can have over us by using data mining procedures and how much they can glean about intimate aspects of our lives. For example, a Target store was able to determine that a teenager was pregnant from the items she had bought there. Her parents, with whom she was living, first learned she was pregnant from the coupons Target started sending her. Fry describes how the Chinese government, through its Sesame Credit citizen scoring system, uses similar techniques. Based on a wide range of data gathered from online activity, Sesame Credit determines a single score for each person. When such scoring becomes mandatory in 2020 it will give the government enormous power over people’s lives.

Chapter two reviews algorithms that calculate the most appropriate sentences for convicted criminals or predict which defendants will skip bail or commit a crime while on bail. A concern expressed throughout the book is our bias toward believing computerized results, even though we know that dubious input data or computational methods imply dubious results. Moreover, widely used algorithms such as COMPAS are proprietary so that only the creators of the algorithm know how it works. This makes it particularly difficult to understand COMPAS’ known biases. Fry believes judges rely on the output of COMPAS more than they should. One reason for this over-reliance is the judges’ desire to be held less accountable. But the author also notes that human judges often make terrible mistakes; for example they can reach different conclusions from essentially the same data depending on the time of day they judged a case.

The chapter on algorithms in medicine focuses mostly on medical diagnosis. It includes pattern recognition to detect cancerous cells in biopsy slides, as well as more challenging tasks such as predicting which cancers might kill you. Part of the difficulty in making accurate diagnoses lies in the poor quality and accessibility of many people’s complete medical data. Another concern is more familiar: maintaining the privacy of medical data such as DNA test results while at the same time making these data available to those doing R&D in, for example, medically-related artificial intelligence. The author emphasizes how difficult it is to strike an appropriate balance between such competing demands, particularly when the benefits of an algorithm are overstated and the risks are obscured.

Another chapter discusses semi-autonomous and completely driverless car development. Fry describes using Bayes theorem to improve the decision-making of algorithms that drive these vehicles. Critical issues include: what degree of good performance is good enough for us to allow driverless cars on the streets? How will this technology interact with such random phenomena as unruly people? What about the moral decisions involved in prioritizing the safety protection for people involved in a potential accident? Fry also discusses the interesting problem of the large time lag between when a semi-autonomous vehicle senses trouble and when the backup human driver becomes sufficiently engaged to deal with the accident about to occur.

One chapter deals with predicting where and when crimes might occur and who might commit them. Fry is concerned about the efficacy and fairness of the mathematical models in these algorithms, as well as the growing use and accuracy of facial recognition software. The final chapter discusses computer programs that produce such artistic works as music in the style of Bach.

This is an interesting and enjoyable read. Fry’s main concern is finding a way to use algorithms to improve our lives while recognizing their weaknesses and strengths. She ultimately concludes that the best result is a partnership between human and algorithm with the final decisions made by the human.

Martin Epstein
California State University, Los Angeles
epstein@calstatela.edu

These contributions have not been peer-refereed. They represent solely the view(s) of the author(s) and not necessarily the view of APS.