In Atlas of AI, Kate Crawford puts forward a timely, piercing analysis of Artificial Intelligence (AI) by inviting the reader on a compelling and, at times, deeply disturbing physical and intellectual journey that encompasses the multifarious, interconnected frontiers occupied by a small number of powerful technology firms. AI is emphatically declared to be neither artificial nor intelligent, and is defined broadly as “the massive industrial formation that includes politics, labor, culture, and capital.” At the outset, Crawford clearly establishes why it is essential to delve deeper than the ‘enchanted’ realm of algorithms, neural nets and statistical pattern recognition and ask overtly political questions, principally: “what is being optimised, and for whom, and who gets to decide”. The invocation of atlases sets up a colonial interpretation of dominance that comes to characterise the ensuing chapters. Atlas of AI is fundamentally a story about extraction, asymmetries of power and an existential crisis that must not be abstracted.
Commencing at the level of Earth, Chapter One underscores AI’s rapacious appetite for rare minerals, water and fossil fuels, dispelling any myths that the clouds of Big Tech bear a resemblance to their meteorological namesake. Essential to terrestrial extraction and production is labour and Chapter Two recalls the horrific realities of metal mining and iPhone factory suicides, from which it is all too easy to become detached as seemingly endless hardware and software innovations are launched on the global stage in increasingly spectacular displays of humanity’s apparently insatiable quest for something more. Crawford proceeds to undress the Mechanical Turks of AI, including Amazon’s own eponymous system, and, after paying a visit to one of the e-commerce giant’s fulfilment facilities, reveals the extent to which harsh labour extraction across supply chains is necessary to “support the illusion of automation”. Insights into the theories of figures such as Charles Babbage, Henry Ford and Frederick Winslow Taylor provide vital historical perspective on micromanagement and the dehumanisation of workers. Turning to an altogether less tangible form of extraction, Chapter Three elucidates how machine learning systems make inferences, how unfettered data collection on a planetary scale has become ‘moralised’ and why this might amount to a “capture of the commons”. By no means comprehensive, this Chapter offers a whistle stop tour through data, which is described as a “bloodless term”, and ends up at a conclusion that ultimately demands far greater substantiation than a book of this nature might reasonably be expected to provide. What Crawford does effectively is to lay down a challenge to the orthodoxy, which provokes the reader to think hard about quotidian data practices to which they are party.
The book proceeds to explore how data is ordered into training sets that form the ‘epistemic foundation’ that underpins how the world is classified by AI systems. In Chapter Four, Crawford amply illustrates the inherent limitations of and problems arising from classification technologies and arrives at a fundamental question: “What epistemological violence is necessary to make the world readable to a machine learning system?” Given that analysis of publicly available AI training sets has raised profound concerns relating to the dispensation of advantage, the reader is prompted to question what forces may be at play behind the closely guarded walls of the largest private technology firms and how they are shaping the lives of citizens across continents. Relatedly, Chapter Five critically assesses affect detection, a seemingly dystopian field of AI that is enjoying considerable growth despite strong challenges to its underlying theory from the scientific community. Shifting gear, Chapter Six sees Crawford plunge the reader headfirst into the fascinating world of state controlled AI. The book charts the relationship between the public and private sectors and, in so doing, maps out the complex geo-political landscape of the twenty-first century. Finally, Crawford deftly carves out a deeply satisfying Conclusion in the form of a pithy and plucky treatise on power that advocates for a fundamental recalibration of our collective approach towards AI regulation and resists “patch[ing] together legal and technical restraints on systems after the fact”. Crawford’s Coda, entitled Space, offers a refreshing exposition on the private space race and focuses attention on the Amazon founder Jeff Bezos, whose Wernher von Braun inspired endeavours are acerbically described as “a hedge against Earth”. Ultimately, Atlas of AI empowers the reader with a reasoned rejection of ‘technological inevitability’ and presciently signposts the choices that a new way of living on, and in harmony with, Earth might involve.
Although highly original and purposeful, the book is not without shortcomings. In particular, this reader would suggest that it feels like something of an oversight for a study concerned with asymmetries of power not to address the subject of algorithmic collusion, which is an emerging focus for research at the intersection between computer science and competition law. In addition, it is relevant to draw attention to a bold assertion made towards the end of Chapter One that “the scale required to build artificial intelligence systems is too complex, too obscured by intellectual property law, and too mired in logistical and technical complexity for us to see into it all.” The claimed negative impact of intellectual property (IP) law passes without so much as a reference, is not dealt with explicitly during the course of the book and, as such, risks being accepted as an unassailable truism. This reader would suggest that here Crawford falls back on a common trope that IP law inherently enables rather than restrains firms in pursuit of absolute power. Whilst IP and related law may be used by some firms to sew up sections of the fabric of AI’s proverbial veil, a more rigorous and nuanced analysis of the essential building blocks of AI reveals how the absence of a strong property rights regime may in fact have contributed significantly to a number of the grand problems that are the work’s central focus. Of course, Atlas of AI claims neither to be a universal nor a legal text per se and these shortcomings do little to detract from the work as a whole.
Atlas of AI is an important contribution that offers an insightful investigative account of how we have arrived at a critical moment in human history. In many respects Crawford’s work is a modern incarnation of Geoffrey Chaucer’s The Canterbury Tales, and this reader would urge all readers, from undergraduate students to Supreme Court judges, entrepreneurs at tech start-ups to government ministers, to dedicate their next available Sunday to devouring this refreshingly accessible and skilfully constructed book.
Ben Evans holds an LLM in IT & IP Law and is a Postgraduate Researcher at the University of East Anglia Law School. His research is focussed on the problem of data advantage, and he is a member of the Centre for Competition Policy and the Society for Computers and Law.
About the Book
- Hardback / Paperback
- Published: May 2021
- ISBN:9780300209570
- Yale University Press
- 336 pages
