The three foundational components of Marxist philosophies of economics and politics are listed in the title of this book: labor, capital, and machines. They are all affected by the growth of artificial intelligence (AI) and, especially, machine learning. Marxism is neither a historical curiosity from the 19th and 20th centuries nor a political “boogeyman” used to scare voters. It is a school of philosophical thought that has many variants and active internal debates. This book shows how the question of AI generated debate within the Marxist philosophical community and the author’s evaluation of the different schools of thought as they apply to AI. Marxism does not offer a single narrative despite an 80-year history as the claimed governing philosophy of the Soviet Union. Many Marxisms have arisen because of the difficulties in understanding the classical writings of Karl Marx and Friedrich Engels.
The images of labor, capital, and machines in the 19th and most of the 20th centuries were those of large factories employing many workers in assembly line production. Although production processes are much more complex now, Marxist philosophy holds that the fundamental relationships among these three components retain the same qualities. The process of creating value (valorization) is the same now as it was 150 years ago. In Marxism, the surplus value of a product is captured and optimized by capital. It is the value beyond the value of the labor of the worker. Machines come into play as a tool by which the surplus value may be increased by greater productivity of the worker by lowering the labor component of the product. The more sophisticated the machine, the greater the reduction of labor. Hence, automation reduces the premium on human skills that must be on the labor side of the equation. AI should be thought of as automation on steroids--replacing not only physical skills but also mental skills.
Steinhoff’s book has eight chapters that are logically divided into three parts: Marxist evaluations of AI (three chapters), a history of and description of the AI industry (three chapters), and a concluding critique of the different Marxist evaluations in the first part of the book (two chapters).
The first part is a guide to the different Marxist interpretations of AI. The basic question in these chapters is “whether artificial intelligence is a tool used by capital in order to generate more surplus value to be captured, or an autonomous tool by which labor can liberate itself from capital’s pursuit of surplus value?” Steinhoff introduces the reader to three different, competing, and conflicting schools of Marxist thought: post-operaismo, labor process theory (LPT), and the new reading of Marx (NRM). Post-operaismo was originally an Italian school of thought that holds an optimistic opinion of the autonomy of AI that would liberate labor. LPT is based on observations that machines and their automation deskill workers and capture control of the production process in the hands of capital. The development of software as a productive process has not been studied extensively by the methods of LPT. (In the next part, Steinhoff begins to remedy this lack.) NRM emphasizes that value can only be realized within the context of social relations. The author rejects post-operaismo and favors an approach that combines LPT and NRM. He provides his evidence in the second part.
The second part of the book combines a history of AI and a large body of evidence, compiled by the author, of the actual process of working in AI. The history of this discipline is described in detail--all of the hyped expectations and crashes of disappointments, the shortcomings of expert systems, the roles of government and industry, and the evolution of AI into machine learning/deep learning.
This evolutionary process has taken AI and machine learning into an unknown area of “the automation of automating automation” implemented in very large artificial neural networks configured and running in the cloud. Steinhoff’s in-depth analysis of the work experience in machine learning is eye-opening. He describes the extreme stratification of the workers, the labor required to capture and prepare the enormous amount of data used, the long hours of work, and the effect of the amounts of money involved on the industries themselves, academic institutions, and national governments.
The third part of the book resolves the questions raised in the first part with the evidence reported in the second. There is nothing in machine learning or deep learning that contradicts the insights of LPT and NRM. The optimism of post-operaismo is unsupported by observation. There is no autonomous work that will liberate workers. The Marxist analysis holds well.
A few comments about the author’s style are warranted. The author is very clear and concise in showing the reader what his objectives are and are not, and in guiding the reader through the arguments and evidence. He is explicit in reviewing what he has completed and what the next steps will be. The reader should be aware that this book is basically a philosophical treatise. The first part may be heavy going because of the unavoidable philosophical language. It is somewhat like reading the first portion of an Umberto Eco novel. (A philosopher might be similarly discomfited in reading many computing monographs.) If the reader concentrates and tries to convert what is being said into his or her own frame of reference, it will be well worth the effort. The second part is truly masterful--thorough, lively style, careful reporting, cogent and clear analysis. Until I read the second part, I had no idea of the scope of machine learning in contemporary economics, how the industry was structured, and the evolutionary trends of this discipline. The third part pulls it all together, showing how the issues raised in the first part are resolved by his investigation.
It is worth the effort to read this book.