The Atlas of AI

Prepare to have your understanding of artificial intelligence fundamentally reshaped. "The Atlas of AI" doesn't just explain what AI is; it unveils the hidden infrastructure and profound human costs behind the algorithms that increasingly govern our lives. Kate Crawford invites you on a journey that starts with the illusion of intelligent machines, like the famous horse Clever Hans, and quickly pivots to reveal the vast, often brutal, realities it conceals. You'll traverse the globe, from the gleaming headquarters of tech giants to the stark, resource-intensive landscapes where AI's physical foundations are laid, and delve into the often-invisible human labor that fuels its operations – the anxiety-ridden Amazon fulfillment centers, the data miners, and the workers whose lives are dictated by relentless metrics. Beyond the physical, this book excavates the intellectual and social bedrock of AI. You'll confront how historical prejudices, like those embedded in 19th-century craniology, are echoed in contemporary data classification systems, and understand the complex, often controversial, origins of affect recognition technology. Crawford pulls back the curtain on the shadowy, state-driven origins of AI, revealing its deep entanglement with power, inequality, and surveillance, dismantling the myth of its neutrality. Ultimately, "The Atlas of AI" offers not just knowledge, but a critical lens. You will gain a profound awareness of the societal structures, political forces, and ethical dilemmas that AI amplifies. This is an intellectually stimulating and emotionally resonant exploration that will empower you to see the world—and the future—with new eyes, questioning the promises of technological progress and demanding accountability for its true impact.

The author, Kate Crawford, invites us into a world where the lines between intelligence, illusion, and industry blur, beginning with the captivating tale of Clever Hans, the horse that seemed to possess extraordinary mathematical and communicative abilities. This nineteenth-century sensation, capable of tapping out answers to complex problems, captivated audiences and challenged prevailing notions of animal intellect. Yet, as Crawford reveals, a closer examination by Oskar Pfungst uncovered a subtle truth: Hans's "intelligence" was a product of unconscious cues from his questioners, a phenomenon now known as the Clever Hans Effect. This story, far from being a mere historical curiosity, serves as a powerful cautionary tale, illustrating how biases can insidiously infiltrate systems and how our desires can shape what we perceive as intelligence. It highlights a central dilemma: how intelligence is constructed and the inherent traps that arise in its making, a lesson profoundly relevant to the burgeoning field of artificial intelligence. Crawford argues that the mythologies surrounding AI – that nonhuman systems are mere analogues of human minds, or that intelligence exists independently of social and political forces – are deeply flawed. Instead, she posits that AI is not artificial, nor is it purely intelligent; it is deeply embodied and material, a vast industrial formation intricately woven from natural resources, human labor, historical context, and the very fabric of power structures. The ambition to "map the world" computationally, to become the ultimate arbiter of knowledge, is not a neutral scientific endeavor but a colonizing impulse, a "registry of power" designed to serve dominant interests. Crawford calls for a shift in perspective, urging us to view AI through the lens of an atlas, not as a singular, transparent black box, but as a complex, multifaceted terrain of interconnected material architectures, political forces, and human endeavors, from the extractive mines powering computation to the precarious labor that sustains it, and the classifications that enforce hierarchies. This expanded view reveals AI as politics by other means, a force reshaping our understanding of ourselves and the world, demanding that we confront its extractive nature, its power dynamics, and its profound implications for justice and equity, lest we become ensnared in its carefully constructed illusions.

The author, Kate Crawford, embarks on a journey that begins in the gleaming heart of Silicon Valley, revealing the iconic headquarters of tech giants like Apple and Google, but quickly pivots, driving east into the stark landscapes of Nevada. This shift is crucial, as Crawford seeks to uncover the material origins of artificial intelligence, a quest that leads her to the remote Clayton Valley and the Silver Peak lithium mine. She explains that the seemingly ethereal nature of AI is deeply rooted in the earth, mirroring the historical extraction that built cities like San Francisco from the spoils of gold and silver. This historical parallel highlights a core tension: the immense technological progress we celebrate often masks a history of environmental devastation and displacement, a pattern repeated from the 19th-century gold rushes to today's demand for lithium, the critical element powering our rechargeable batteries and electric vehicles. Crawford vividly describes the iridescent green ponds of the lithium mine, a stark contrast to the alien-looking pipes siphoning brine from the earth, illustrating how the 'stuff' of AI is extracted. She reveals that the tech industry, much like historical mining operations, operates under a strategic amnesia, overlooking the true costs—environmental degradation, resource depletion, and the human toll—in its relentless pursuit of innovation, embodying a different era's 'move fast and break things.' The narrative then expands globally, touching upon the extraction of rare earth minerals in places like Baotou, China, where toxic black lakes are formed from hazardous processing, and tin mining in Indonesia that scars landscapes and endangers communities. Crawford introduces the concept of the 'planetary mine,' emphasizing that AI's material demands extend far beyond discrete mining locations, permeating the entire globe through complex supply chains, from the guttapercha harvested for Victorian telegraph cables to the cobalt and germanium essential for modern devices. She underscores that the cloud, often perceived as intangible, is in reality a material infrastructure powered by immense energy, often from fossil fuels, and demanding vast amounts of water, as seen with the NSA's data center in Utah. The chapter culminates by framing AI as a 'megamachine,' akin to the Manhattan Project, a complex system of industrial infrastructures, global logistics, and human labor kept opaque, demanding enormous resources while concealing its true environmental and human costs. Crawford urges a shift in perspective, moving beyond the myth of 'clean tech' to confront the material realities and interconnected systems of extraction that underpin our digital world, suggesting that understanding these hidden costs is vital for achieving greater justice.

Kate Crawford's chapter on 'Labor' invites us into the stark reality of modern work, beginning with a visceral tour of an Amazon fulfillment center. Here, the air hums not just with the whir of Kiva robots, but with a palpable anxiety among the human 'associates' whose every second is timed and tracked, their bodies often bearing the silent testament of injuries from repetitive, fiddly tasks that machines cannot yet perform. Crawford reveals a core tension: while AI systems like the 'matrix' algorithm meticulously optimize packaging to prevent breakages, human bodies become mere 'connective tissue,' an afterthought in the relentless pursuit of efficiency. This isn't a future of robots replacing humans, she argues, but a present where humans are increasingly treated *like* robots, a regression to older industrial practices of exploitation, amplified by AI's capacity for granular surveillance and algorithmic assessment. She traces this lineage back through figures like Charles Babbage, who viewed labor as a problematic source of value rather than its creator, and Frederick Winslow Taylor, who quantified human movements with stopwatches, demonstrating how the drive for efficiency and profit has historically devalued human experience. Crawford illuminates the hidden 'ghost work' and 'human-fueled automation' that underpins AI, where underpaid workers perform rote tasks, often without credit or fair compensation, creating the illusion of artificial intelligence. She calls this 'Potemkin AI'—facades designed to impress investors while relying on unseen human effort, a modern echo of Samuel Bentham's inspection houses and the panopticon, itself inspired by early manufacturing facilities. The chapter underscores that contemporary AI is deeply rooted in the exploitation of human bodies, a continuation of industrial capitalism's logic where value is extracted not from labor itself, but from its control and commodification. The narrative builds tension as Crawford highlights the privatization of time, from the railroad magnates' imposition of standardized time zones to Google's proprietary 'TrueTime' protocol, all designed to centralize control and maximize profit. This relentless rhythm of the 'rate,' dictated by executives far removed from the warehouse floor, is the central dilemma. Yet, Crawford offers a glint of resolution, pointing to worker resistance, from the smashing of time clocks at Fordlandia to modern protests demanding fair treatment and the refusal to be treated 'like robots.' She concludes that AI and algorithmic monitoring are merely the latest tools in a long history of controlling labor through time and surveillance, emphasizing the urgent need for cross-sector solidarity to fight for more just and humane working conditions for all.

The author, Kate Crawford, embarks on a profound exploration of 'data' as the foundational, yet often invisible, engine of artificial intelligence, revealing how its acquisition and application have fundamentally reshaped our world. She begins by unearthing a disturbing truth within the archives of the National Institute of Standards and Technology (NIST): mug shots of individuals, stripped of their names and contexts, are used as technical baselines to test facial recognition software. This practice, a chilling evolution from Alphonse Bertillon's identification system to Francis Galton's 'Galtonian formalism,' reduces human beings to mere data points, their vulnerability and pain disregarded in the relentless pursuit of algorithmic accuracy. Crawford argues that this mindset, where everything is data 'there for the taking,' is the urtext of modern AI, a logic that pervades the tech sector and normalizes the mass extraction of information without consent or consideration for its origins. She meticulously traces the historical demand for data, from Vannevar Bush's early visions of data-hungry machines to the probabilistic approaches of the 1970s and 80s, highlighting how IBM's speech recognition group, led by figures like Robert Mercer, shifted focus to statistical methods requiring vast amounts of training data, famously proclaiming, 'There's no data like more data.' This led to the repurposing of massive text corpora, like those generated from a thirteen-year antitrust lawsuit against IBM, and later, the Penn Treebank and the Enron email dataset, all treated as interchangeable linguistic material, often obscuring the biases and specific contexts from which they were drawn. The chapter then pivots to the capture of the human face, detailing the Face Recognition Technology (FERET) program as a precursor to internet-scale data collection, where consent was still a factor, though diversity was already a concern. The true paradigm shift, however, arrived with the internet, transforming it into a 'natural resource' for AI research. Platforms like Facebook and Twitter became endless pipelines of images and text, fueling the creation of massive datasets like ImageNet. This monumental effort, driven by Fei-Fei Li's vision, relied heavily on Amazon Mechanical Turk, deploying low-paid crowdworkers to label millions of images, often resulting in offensive and racist classifications imported from lexical databases. Crawford exposes the erosion of consent in this era, citing cases of secret camera installations at universities and the harvesting of celebrity photos for datasets like Microsoft's MSCeleb, which ironically included critics of surveillance. She argues that the commodification of data, framed as 'the new oil' or a capital asset, justifies this relentless extraction, creating 'data subjects' devoid of subjectivity or rights, and sidestepping ethical review processes that were designed for biomedical research, not for abstract computational endeavors. The chapter culminates in the concept of the 'capture of the commons,' where publicly available data, once considered part of a shared good, is privatized by tech companies, enriching them while diminishing the spaces free from surveillance and data collection. Ultimately, Crawford reveals that the way data is understood, captured, and classified is not a neutral technical act but a profound act of world-making, a political intervention that shapes AI's vision and disproportionately affects vulnerable communities, often under the guise of benevolent innovation.

The author, Kate Crawford, invites us into a room filled with human skulls, a stark reminder of how classification has long been entangled with power and prejudice. She guides us through the work of Samuel Morton, a 19th-century craniologist who, with meticulous, though flawed, measurements, sought to rank human races, his 'objective' data tragically used to legitimize slavery and segregation. This historical echo resonates deeply into our present, as Crawford reveals how the very act of classification in artificial intelligence, from labeling images to assessing risk, carries similar political weight. She illustrates this with Amazon's failed attempt to automate hiring, where a system trained on predominantly male résumés inadvertently penalized women, demonstrating how past biases become encoded into future technologies. The chapter then probes the limitations of simply 'debiasing' AI, highlighting IBM's 'Diversity in Faces' dataset, which, despite good intentions, reinforced a binary understanding of gender and reduced diversity to measurable facial features, akin to Morton's craniometry. Crawford argues that bias isn't merely a bug to be fixed, but a feature of classification itself, deeply embedded in the 'epistemic machinery' that constructs our understanding of the world. She uses the vast ImageNet dataset as a case study, showing how its hierarchical structure, imported from WordNet, contains latent biases, categorizing people in ways that naturalize gender as binary, essentialize race, and even include deeply offensive slurs, all presented under a veneer of technical neutrality. The core tension, Crawford explains, lies in the inherent power of classification: the ability to define, categorize, and thus, control. When AI systems classify people, they don't just reflect existing social orders; they actively construct them, often flattening complex human identities into quantifiable, and frequently harmful, categories. The chapter concludes by emphasizing that true justice in AI requires more than technical fixes; it demands a fundamental shift in perspective, acknowledging that classification systems are not static geology but dynamic, often invisible, forces that shape our world, and that challenging these systems requires sustained political will, much like the historical struggles against oppression. The real work, she suggests, is in understanding that every classification has a consequence, and power, in this domain, concedes nothing without a demand.

The author delves into the complex origins and pervasive influence of affect recognition technology, tracing its roots back to the controversial research of Paul Ekman and the foundational theories of Silvan Tomkins. In 1967, Ekman journeyed to Papua New Guinea, hoping to find universal human emotions displayed by the isolated Fore people, using flashcards to test his hypothesis that a small set of innate affects are recognizable across all cultures. Despite initial frustrations and methodological challenges, Ekman's work, bolstered by Cold War-era U.S. intelligence funding, eventually laid the groundwork for an industry now worth billions. This chapter reveals how Ekman's presuppositions, despite significant scientific doubt and numerous critiques, became deeply embedded in artificial intelligence, particularly in areas like national security, airports, education, and hiring. The narrative highlights the inherent circularity and questionable scientific validity of systems trained on posed or simulated facial expressions, often labeled using schemes derived from these same artificial displays. For instance, companies like Human and HireVue use AI to analyze video interviews, scoring candidates on traits inferred from facial cues, while giants like Affectiva and Microsoft offer emotion detection APIs built on large databases of expressions. The chapter underscores a critical tension: the desire to automate the understanding of human emotion, driven by military priorities, corporate profit, and a fear-based political climate, clashes with the deep complexity and cultural variability of human feeling. The author explains that foundational to this AI industry is the idea, championed by Tomkins and Ekman, that emotions manifest in universally recognizable facial expressions, a notion challenged by anthropologists like Margaret Mead and psychologists such as James Russell and Lisa Feldman Barrett, who emphasize the profound influence of context and culture, arguing that a scowl is not necessarily anger. This has led to systems that are not only scientifically dubious but also prone to racial bias, as seen in the critique of the SPOT program for airport security, which disproportionately flagged certain groups. The core dilemma lies in the AI field's adoption of a simplified, measurable model of emotion, often based on Ekman's taxonomies, ignoring the nuanced, emergent, and relational nature of human experience. The chapter concludes by urging a critical examination of the origins and consequences of these emotion recognition systems, cautioning that their widespread deployment in high-stakes contexts risks unfair judgments and reinforces existing power structures, all while operating on a flawed premise that human emotion can be accurately and universally decoded from facial movements alone, a notion increasingly disproven by scientific consensus.

Kate Crawford, in her chapter 'State,' pulls back the curtain on the shadowy origins of artificial intelligence, revealing a parallel sector forged not in open labs but within the classified confines of intelligence agencies. From an air-gapped laptop, poring over the Snowden archive, she uncovers how programs like TREASUREMAP and FOXACID, developed by entities like the NSA and GCHQ, laid the groundwork for the pervasive data extraction and surveillance systems we see today, far predating the public-facing AI boom and operating with a starker, more acquisitive logic. The author explains that the U.S. intelligence apparatus, alongside agencies like DARPA, has been a foundational driver of AI research since the 1950s, shaping its priorities around command, control, and surveillance, a legacy that infused AI with classificatory thinking and the notion of 'knowing them by their metadata.' This state-driven AI, once confined to national security, has now seeped into municipal services and law enforcement, exemplified by companies like Palantir, which apply militarized pattern detection to civilian contexts like deportation and policing, often amplifying existing inequalities and creating a 'techno-washed' inequity. Crawford highlights the dramatic shift with the 'Third Offset' strategy, a conscious effort to integrate AI into military operations, leading to initiatives like Project Maven, where the defense department partnered with tech giants like Google and Microsoft, igniting internal ethical debates and revealing the complex entanglement of corporate and military interests. This collaboration, driven by a desire for national control and international dominance, has blurred the lines between civilian and military applications, transforming once-secretive tools into widely accessible, often unregulated, surveillance infrastructures like Vigilant Solutions' license plate readers and Amazon's Ring network. The chapter's central tension lies in how these state-born, militarized logics of targeting and risk assessment, once used against perceived enemies, are now being 'outsourced' and applied to everyday citizens, creating 'terrorist credit scores' and 'social credit scores' that judge individuals based on data signatures and patterns, often with severe, inaccurate consequences, as seen in the automated austerity programs in Michigan. Ultimately, Crawford argues that the myth of 'sovereign AI' contained within national borders is dissolving; the reality is a hybrid, planetary-scale computation where state, municipal, and corporate logics are deeply intertwined, creating a 'fever dream of centralized control' that has privatized public surveillance and recalibrated state sovereignty around corporate algorithmic governance, leaving individuals subject to opaque systems of tracking and scoring. The narrative moves from the tension of hidden state power to the insight of its pervasive, outsourced application, resolving with a cautionary note on the profound power imbalance and lack of accountability in this new era of data-driven governance.

The author, Kate Crawford, unpacks the pervasive myth of artificial intelligence as an objective, neutral force, revealing instead that AI systems are deeply embedded in and shaped by human social, political, cultural, and economic structures, designed to amplify existing hierarchies and inequalities for the benefit of states, institutions, and corporations. Crawford challenges the narrative of 'algorithmic exceptionalism' – the idea that AI's computational prowess makes it inherently superior and unbiased – by highlighting how spectacular feats, like AlphaGo Zero's game mastery, are often presented as magic, obscuring the immense human effort, capital investment, and, crucially, the power dynamics at play. This mystification, termed 'enchanted determinism,' creates a dual narrative: one of utopian salvation through AI, the other of dystopian peril, both of which falsely locate power solely within the technology itself, thereby distracting from systemic forces like neoliberalism and racial inequality. Crawford then pivots to the material reality of AI, illustrating with the blueprint of a Google data center how its vast energy consumption and expansion are publicly subsidized and dependent on public utilities, reminding us that AI's 'immateriality' is a carefully constructed illusion. This leads to a core insight: AI is fundamentally an extractive industry, drawing on planetary resources, human labor (from miners to crowdworkers to shadow workforces within tech companies), and vast amounts of data, often harvested without consent, to serve the interests of a powerful few. The author argues that the conventional understanding of AI, represented by abstract diagrams, misses the 'wider landscape' of interconnected global systems of extraction and power. She emphasizes that AI's datasets are not neutral raw materials but political interventions, shaping how the world is seen and classified, often through reductive and harmful taxonomies that mirror historical practices of essentializing identity. This process, where complex human experiences are transmuted into 'computable sameness,' is a form of epistemological violence, reducing the multifaceted world to machinereadable tables based on proxies. Crawford critiques the pervasive 'collectitall' mentality and the reliance on 'operational images' – representations of the world made solely for machines – as inherently political acts that claim scientific neutrality while embedding bias. The chapter builds tension by detailing how these systems, particularly when entangled with state power and military applications, lead to profound surveillance, labor exploitation, and the amplification of existing injustices, effectively redrawing civic life to strengthen centers of power. The author proposes a resolution not in 'democratizing AI' as a panacea, akin to democratizing weapons manufacturing for peace, but in a 'politics of refusal' – questioning *why* AI ought to be used, rather than merely accepting that it *can* be. This requires centering the lived experiences of those most harmed by AI, understanding that ethical principles alone are insufficient without addressing the power structures that shape technology. Crawford concludes by advocating for united justice movements that connect issues of capitalism, computation, and control, urging a shift from technological inevitability to a vision of a more just and sustainable world, where we chart a course beyond extraction and discrimination, recognizing that the greatest hope lies in collective action that challenges the fundamental patterns of domination.

The narrative opens with the thunderous spectacle of a Saturn V rocket launch, a scene imbued with the fervent ambition of tech titans like Jeff Bezos, who, inspired by the dawn of space exploration, sees humanity's future not on a constrained Earth, but among the stars. This vision, amplified by the soaring rhetoric of private space companies like Blue Origin and SpaceX, presents a compelling, albeit selective, narrative of progress, juxtaposing images of boundless cosmic potential against the stark realities of Earth's congestion and resource limitations. The chapter reveals a core tension: the deeply intertwined ideologies of artificial intelligence and space exploration, both fueled by extreme wealth and a desire to transcend earthly constraints. These tech billionaires, leveraging public sector innovations and government incentives, are not merely seeking to explore; they aim to extend extraction and growth across the solar system, mirroring a colonial impulse to claim new frontiers. Kate Crawford unearths the intellectual lineage of this ambition, tracing it back to Gerard K. O'Neill and his response to the Club of Rome's 'Limits to Growth' report, a seminal work that predicted resource depletion and population collapse. While 'Limits to Growth' advocated for sustainable management and equitable resource distribution, O'Neill, and by extension Bezos and Musk, proposed an escape route through space colonization, a vision that sidesteps the complex systemic issues of consumption and inequality on Earth. This "escape" narrative, however, is exposed as a dangerous fantasy, a hedge against planetary limits rather than a genuine solution, especially when considering the historical legacy of figures like Wernher von Braun, whose contributions to rocketry are shadowed by his use of slave labor. Crawford underscores the ethical chasm, highlighting how this pursuit of space mirrors other immortality-focused fantasies of the tech elite, while simultaneously undermining international agreements like the Outer Space Treaty, which designates space as a common heritage of humankind. The author’s own journey to Blue Origin’s West Texas facility provides a visceral counterpoint to the grand pronouncements, revealing a guarded, provisional infrastructure—a private dominion in the making, shadowed by a history of colonial violence and a chilling sense of being surveilled. Ultimately, the chapter concludes that this relentless push into space is less about genuine human progress and more about an existential fear of stasis, mortality, and the undeniable, encroaching limits of our own planet, a fear that drives a powerful few to seek an ultimate escape.

Kate Crawford's "The Atlas of AI" offers a profound and unflinching demystification of artificial intelligence, dismantling the illusion of its abstract, disembodied nature. Instead, it reveals AI as a material formation, a planetary extractive industry deeply entwined with the exploitation of natural resources, the hidden labor of countless human workers, and the intricate, often unjust, socio-political structures that govern our world. The core takeaway is that AI is not a neutral tool but a 'registry of power,' a computational colonizing impulse that maps and defines reality according to dominant interests. Crawford masterfully illustrates this through concepts like the 'planetary mine,' where the seemingly intangible 'cloud' is shown to have a vast, tangible environmental footprint, and 'ghost work,' the unseen human labor that underpins the facade of automation. Emotionally, the book evokes a sense of disquiet and urgency. It highlights the 'strategic amnesia' of the tech sector, which conveniently forgets the human and environmental costs of its relentless pursuit of profit, mirroring historical patterns of industrial exploitation. The narrative challenges the myth of 'clean tech' and exposes how AI can amplify existing inequalities, particularly in the dehumanization of workers, reducing them to data points and abstracting their labor. The practical wisdom lies in Crawford's call for a 'politics of refusal' – a move beyond superficial ethical guidelines and technical fixes to fundamentally question the necessity and purpose of AI applications. She urges us to confront the material realities, the hidden labor, and the inherent biases embedded in AI's classification systems and datasets, which are not neutral resources but political interventions. The book underscores the interconnectedness of capitalism, computation, and control, advocating for united justice movements that link labor, climate, and racial justice to dismantle the power structures AI reinforces. Ultimately, "The Atlas of AI" is a vital call to awareness, urging us to see AI not as a futuristic marvel, but as a deeply material and political force that demands critical engagement and a commitment to building a more equitable and sustainable future, both on Earth and, as the coda suggests, by challenging the extractive ideologies that drive even the ambition for space colonization.