Nilesh Jasani's Reviews > Algorithms for the People: Democracy in the Age of AI

An easy way to write a bestseller in recent years is to attack algorithms that drive our lives through the largest tech companies' apps, products, and associated ads. It did not matter if all the suggestions of such books were utterly impractical as long as they succeeded in arousing the hackles by invoking the loss of privacy and choice.

Josh Simons' book is exactly the opposite and hence is unlikely to be a bestseller! The author approaches the topic pedantically with the writing style also of a career academician. The analysis is scholarly, nuanced, and result driven. The technical details, discussions of obscure philosophical points, and a tendency towards accepting the powers of some new-age forces cause the book challenging to accept for anyone looking for quick fixes or points for dinner party conversations. If one throws in some of the conclusions that argue against needless political and other interventions, the practicality and sagacity make it perfectly serious but so unsalacious that it is not a surprise this review is the first on the book's page weeks after its publication.

The following are reviewers' notes based on some of the lessons from the book and others completely independent. Technology continues to increase its influence on how we live. As much as it works in most cases, no amount of customization will make it work perfectly for all in all cases. Many commentators love imposing arbitrary limits on the workings of technology based on these negative outcomes regardless of the substantial cumulative benefits.

The critics' most strident issues are with the biases; the author mostly keeps the discussions to those more relevant in the US, but the arguments are universally applicable in different forms. Models trained on past patterns, the argument goes, are unlikely to help us move away from the plagues of race, religion, gender, wealth, and countless similar bigotries of our past.

The calls to eliminate the biases invariably turn to suggestions that the book shows will do more harm than good. Any forced removal of certain information from the models is unlikely to change the conclusions in a world where machine learning algorithms no longer work on human-created parameters or classifications. The results from algorithms may not be any different (or even worse) if the methods used are so simplistic.

The solutions are partly in assertive, outside-the-model actions that improve the prospects of the disadvantaged. The solutions are partly in the continuous monitoring of the end results by all concerned in critical life areas to permit suitable feedback loops. The solutions are in competition, and more use of technologies for higher granularity, fewer regulations on processes, judicious use of fines, simultaneous human monitoring, etc.

This is a serious book for those interested in the topic. The conclusions are not all I summarise here. They are granular and arrived at with many weighty arguments of all kinds.