Too early to tell: AI and the logic of empire
AI – especially large language models and the infrastructures built around them – is already reshaping our societies, science, politics, and everyday life. But where will this lead?
In this review of Karen Hao’s Empire of AI, Tibor Dessewffy explores the origins of the prevailing AI model and the dominant ideas around its future course and assesses some of the normative proposals to tame its damaging effects on people and planet. He argues that while the current trajectory of AI development gives great cause for concern, the future form of this technology is anything but certain.
According to a frequently cited anecdote, when the Chinese foreign minister Zhou Enlai was asked in the early 1970s about the significance of the French Revolution, he replied: “It is too early to tell.” For a long time, this remark was taken as an illustration of Eastern wisdom and long-term thinking. In fact, however, it referred not to the French Revolution of 1789 but to the student uprisings of 1968.
Read this way, the statement does not concern the retrospective evaluation of a closed historical event, but rather a social transformation still in motion – one in which the temptation to pass judgment is strong, while the structural consequences have yet to become fully visible.
The contemporary discourse surrounding artificial intelligence (AI) is a similar situation. Expectations and fears orbit around extreme poles: on one side, AI appears as a technology promising the liberation of human labor and an era of boundless material prosperity; on the other, it is imagined as an existential threat, an autonomous machine intelligence hostile to humanity itself.
The critique that runs throughout Empire of AI is not directed at artificial intelligence as such, but at the dominant Silicon Valley vision to which AI development has become increasingly tethered. She rejects the idea that AI can only be pursued along a single developmental trajectory or, as she puts it in the epilogue: “the dangerous notion that broad benefit from AI can only be derived from […] a vision for the technology that requires the complete capitulation of our privacy, our agency, and our worth, including the value of our labor and art, toward an ultimately imperial centralization project.”
For Hao, the “AI empire” is not merely a descriptive metaphor, but a normative judgment. Taken as a whole, the book is less concerned with weighing the possible long-term futures of AI than with mapping the registers of unacceptable consequences that are already emerging along the dominant developmental path.
While Hao’s critical urgency is both understandable and legitimate, I remain more hesitant to treat these processes as settled or closed. It is far from clear to what extent this trajectory is truly inevitable, or what kinds of social transformations will ultimately result from the technological dynamic as it unfolds. Without disputing the epochal significance of AI, it is worth recalling the lessons of earlier waves of digital technology. Reality – though radically transformed – has so far fulfilled neither the darkest anxieties nor the most totalizing promises.
For this reason, I am more cautious than Hao when it comes to the alarmist tone, the determinism, and the sense of finality that pervade some of her claims. What she identifies as a decisive crossing of boundaries, I see rather as a trajectory whose outcome remains unresolved.
Karen Hao’s authorial position
Hao began her career with a background in the natural sciences before briefly working at a start-up after university. As an investigative journalist, Hao followed the evolution of the AI industry for years and she documents not only key events but also the communicative practices through which the industry operates.
One of the book’s notable strengths lies in its consistent refusal to drift into tabloid territory, even when the material itself would easily lend support to sensationalism. The focus is not on exposing private lives for their own sake, but on situating individual stories within broader contexts and showing how they contribute to an understanding of institutional dynamics.
Hao’s portraits of the central figures, such as Sam Altman and Ilya Sutskever, are vivid without lapsing into caricature. These actors are neither heroes nor villains, but individuals operating within institutional settings, marked by very human vulnerabilities. Their decisions and conflicts consistently exceed the bounds of personal biography, producing consequences that are structural in scope. Hao’s journalistic strength lies precisely in her ability to show that these decisions are often intelligible in light of personal values, fears, and ambitions, even as their effects prove durable and often difficult to reverse.
The OpenAI story as an organizational case study
The most compelling and persuasive layer of Empire of AI is its detailed reconstruction of OpenAI’s history. Hao traces how an idealistic start-up built around an altruistic mission gradually evolves into a mega-actor governed by market imperatives and capitalist logic. The strength of the book lies precisely in the fact that this transformation is not narrated through retrospective moralizing, but through concrete decisions, individual intentions, conflicts, and structural constraints.
At its founding, OpenAI defined itself as a nonprofit organization. Its stated mission was to achieve artificial general intelligence (AGI) before anyone else – defined by OpenAI as “highly autonomous systems that outperform humans at most economically valuable work” – and to do so in a way that would “benefit all of humanity.” In Hao’s account, this ethos initially appears not as a mere marketing slogan, but as a guiding star; a moral and technological reference point for its founders that, they believed, offered humanity its best chance of bringing AGI into existence safely.
Hao traces how an idealistic start-up built around an altruistic mission gradually evolves into a mega-actor governed by market imperatives and capitalist logic.
From the outset, OpenAI’s institutional identity was shaped in part through a negative point of reference. The organization explicitly positioned itself as an alternative to the dominant big-tech model – above all, to Google’s data-driven, profit-oriented approach. This self-definition, however, already contained the seeds of the tensions that would later come to define the organization’s trajectory.
Hao shows in detail how a fundamental dilemma gradually emerged and came to permeate OpenAI’s entire operation: how to remain faithful to a universal, long-term moral mission while the extraordinary resource demands of AGI research increasingly tethered the organization to powerful market and state actors. The nonprofit form proved insufficient on its own to sustain the scale of research required, and the growing appetite for computational capacity quickly reached a magnitude that structurally excluded traditional academic or civil-society models of operation.
One of the book’s key virtues is that it does not portray OpenAI’s transformation as a single dramatic “turning point,” but rather as the cumulative effect of layered decisions. The partnership with Microsoft, the introduction of the capped-profit model, and the gradual turn toward productization did not dismantle the founding ideals overnight. Instead, they incrementally reshaped the organization’s internal logic, generating persistent and often intense internal tensions along the way. In Hao’s narrative, these steps are largely intelligible and rational within their respective contexts. The actors involved appear not as cynical traitors, but as individuals acting in accordance with their own values and rationalities, within increasingly constrained circumstances.
Flawed and complex human beings
One of Karen Hao’s most striking achievements in Empire of AI is the way she presents OpenAI’s key figures not as icons or abstract decision-makers, but as idiosyncratic and often disconcertingly vulnerable individuals. These portraits are not incidental anecdotes. Hao makes it clear that the direction of the AI industry is shaped not only by institutional pressures and market logics, but also by very specific personalities, complete with their fears, obsessions, and worldviews.
Sam Altman, in Hao’s portrayal, emerges as an unsettling blend of extreme ambition and deep existential anxiety. He is not simply a technological optimist, but someone who takes the prospect of civilizational collapse seriously, even as he actively contributes to accelerating the forces that might bring it about. This apocalyptic mode of thinking cannot be disentangled from Altman’s technological visions. What takes shape is a worldview in which apocalypse and rapid growth – “scaling,” in start-up parlance – anxiety and messianic ambition are not opposites, but two sides of the same cognitive and moral framework.
Hao’s portraits resonate so strongly because they cannot be reduced to psychological curiosities. Individual flaws and eccentricities are channeled into institutional trajectories. Apocalyptic optimism and mystical fear of AGI are not simply personality traits; they are worldviews and habitus that structure OpenAI’s decisions, conflicts, and ultimately its global impact.
In this sense, one of the most uncomfortable lessons of Hao’s book is that the development of artificial intelligence is not a sterile technological process. While this is not unique to AI, the shaping of the future is once again being driven by deeply peculiar, flawed human beings – and Hao shows this with relentless consistency.
In Hao’s interpretation, the story of OpenAI shows how a value-driven initiative gradually adapts to structural constraints imposed by its environment, while its original goals acquire new meanings in the process. Such transformations inevitably generate deep organizational conflicts, in which former allies find themselves opposed precisely because of diverging values and ambitions. The shift does not necessarily entail a complete abandonment of founding principles, but it does alter their function: moral language increasingly becomes a tool for legitimizing institutional growth and survival.
AGI as narrative, ideology, and organizing principle
Having traced the history of OpenAI, Hao’s book turns to the broader mode of thinking structured around AGI. Artificial General Intelligence – defined as a form of AI capable of approaching the full range of human cognitive abilities in an autonomous manner – is one of the most central yet least clearly specified concepts in contemporary AI discourse. Its potential significance follows from this ambiguity: were it to materialize, AGI would not merely constitute a new technological tool, but would carry with it the possibility of transforming the economic, political, and social order itself. In Hao’s account, however, AGI becomes influential less as a concrete technological future than as a narrative and ideological organizing principle within the AI industry.
The vagueness and uncertainty of AGI appears here not as a weakness but as a resource. This indeterminacy allows different actors to invest the concept with divergent meanings and deploy it for strategic purposes. Hao’s implicit claim is that AGI functions less as a technological reality than as a rhetorical and legitimating device: the decisive question is not when or in what form AGI will arrive, but what can be justified in the present in its name.
In this sense, the AGI mission performs a dual function. On the one hand, it operates as an internal motivational force, furnishing researchers and leaders with a moral horizon that extends beyond short-term market success and often imbues their work with existential significance. On the other hand, it also serves as an external instrument of legitimation, capable of justifying the imperative of rapid scaling, the extraordinary concentration of resources required for exponential growth, and a range of compromises that would be far more difficult to defend in other contexts.
According to Hao, the discourse organized around AGI generates a distinctive “end-goal logic” within OpenAI. If AGI is inevitable, then the moral evaluation of present-day decisions no longer rests on their immediate effects, but on their relation to a hypothesized future state. This mode of reasoning readily slides into what the book repeatedly characterizes as an “ends justify the means” logic, in which present harms, inequalities, or forms of exploitation are relativized in light of an imagined universal benefit. The goal does not merely provide direction; it gradually furnishes absolution. The more abstract and distant the endpoint becomes, the more effectively it can obscure the concrete social consequences of decisions made in the present.
Hao does not engage in any sustained assessment of the technological plausibility of AGI itself. Rather than a seriously evaluated future technology, AGI functions primarily as a political and organizational narrative in the book. This is consistent with the book’s overall approach, but it also introduces a degree of one-sidedness: technological optimism that takes AGI seriously tends to appear either naïve or manipulative.
This shift in emphasis becomes especially visible when Hao turns to internal debates within the AGI discourse. She offers a sensitive account of the tensions between so-called “boomers” and “doomsters,” as well as their organizational consequences, yet leaves relatively little room for the possibility that optimistic expectations may be informed not only by ideological constructions, but also by concrete technological experiences and incremental advances. In this sense, Hao’s interpretation of AGI is consistently sociological. Her concern is not with what will ultimately be realized, but with what people and institutions do with the idea of AGI.
In Hao’s analysis, the imperative of scaling and the ever-growing demand for computational capacity do not appear as some kind of technological “law of nature.” The appetite for giga-scale compute emerges instead from the differing motivations and strategic decisions of concrete actors. In the case of OpenAI, the logic of scaling is initially tightly bound to the pursuit of AGI – where ever-larger models are not merely more efficient tools, but prerequisites for achieving general intelligence itself.
Over time, however, this orientation increasingly intersects with another force of comparable strength: the logic of market competition. Through the figure of Sam Altman, Hao shows how scaling becomes, in part, a competitive necessity and, in part, a business strategy. Expanding computational capacity is no longer primarily aligned with the long-term vision of AGI, but with rapid productization, the pursuit of market dominance, and the expectations of investors. Scaling thus becomes a process with a dual genealogy – simultaneously ideological and market-driven.
This duality is central to Hao’s argument. The exponential demand for computational resources does not emerge as an inevitable feature of technological progress, but as the outcome of a contingent series of decisions that gradually harden into structural constraints. What begins as a choice later presents itself as a given, both for the organization and for the industry as a whole. In this sense, scaling is not merely a technical issue, but a point of convergence between the AGI narrative and market rationality – one that fundamentally shapes the future trajectory of the AI industry.
The other side of AI: Global inequalities
Another important layer of Empire of AI unfolds in the chapters where Hao juxtaposes the world of the Californian tech elite – engaged in the pursuit of AGI as a kind of technological Holy Grail – with the everyday realities of the global periphery. The universe of designers’ furniture and luxury offices, inhabited by technologists who often see themselves as bearers of a moral mission, stands in stark contrast to the places where the data labor required for AI systems actually takes place.
Hao’s empirical examples – stories of data workers, content moderators, and annotators in Latin America and Africa – render visible the darker side of the global division of labor underpinning the AI industry. Yet the primary causes of these precarious life conditions are not AI itself, nor even digital companies alone, but the long-standing structural inequalities of global capitalism. The AI industry does not “break into” this world; it enters it. It builds upon existing economic and social relations, exploits them, and in some cases intensifies them – but it does not create them from scratch.
This distinction becomes especially important in relation to Hao’s empathy-driven journalistic narrative. The Venezuelan example, in which data work tied to model training quite literally becomes a condition of survival in a collapsed, hyperinflationary economy, cannot be described simply as exploitation. Similarly, the story of the lesbian couple in Kenya resists reduction to a straightforward “victims of Big Tech” narrative. In such cases, data work plays an ambivalent role: it simultaneously entails vulnerability and – limited but real – forms of agency.
Hao’s sensitivity and empathy are unquestionably among the book’s greatest strengths. At the same time, they raise the question of whether this empathy occasionally leads to a kind of analytical short circuit. The dramatic presentation of global inequalities can easily create the impression that these conditions are direct “crimes” of the AI industry, when in fact they are better understood as outcomes of AI’s embedding in an already asymmetrical world economy. It remains essential to distinguish between causes, mediating mechanisms, and consequences.
This section of the book is therefore both highly compelling and analytically problematic. It is compelling because it makes the global operation of the AI industry visible through concrete human lives; it is problematic because the weight of these empirical cases sometimes obscures structural analysis. Data labor, invisible labor, and environmental burdens are indeed integral to the AI industry – but they become analytically productive only if we clearly identify the alternatives available to different actors and distinguish between the general logics of global capitalism and the AI industry’s specific contributions to them.
A recurring theme in the book is that the energy footprint of a single hyperscale data center is comparable to that of urban infrastructure, not research laboratories.
Hao approaches the environmental dimension of AI not through an abstract sustainability discourse, but through striking comparisons of scale. The energy and water demands of data centers required for large-scale AI models are rendered tangible through analogies that clearly exceed intuitive notions of a “digital” industry. A recurring theme in the book is that the energy footprint of a single hyperscale data center is comparable to that of urban infrastructure, not research laboratories.
In several places, Hao cites estimates suggesting that AI mega-campuses may require between 1,000 and 2,000 megawatts of power – levels comparable to the annual electricity consumption of major cities. These comparisons are not meant to establish technical precision, but to convey scale: to make clear that AI development is an industrial, not marginal, energy issue.
When it comes to water use, Hao offers even more concrete cases. The reliance of data centers on potable water for cooling becomes especially contentious in regions already affected by water scarcity. In the Chilean case, the book provides a detailed account of a proposed data center near Santiago, where plans to consume hundreds of liters of drinking water per second in a drought-stricken area provoked local resistance and ultimately led to the project’s rejection.
In Hao’s narrative, these examples do not appear as isolated controversies, but as illustrations of how the AI industry imposes global infrastructure demands on specific places and communities. The emphasis is not on the precise accuracy of individual figures, but on the way resource use enters into competition with other fundamental social needs: drinking water, residential energy supply, agricultural use.
The strength of Hao’s environmental argument lies precisely in this shift of scale. AI definitively loses its image as a “weightless” digital phenomenon and appears instead as an industry that intervenes in social processes through decisions about energy, water, and land use. At the same time, these costs become truly intelligible only when situated within the existing structural inequalities of global capitalism. AI does not create these conflicts, but it intensifies them and gives them a new scale.
Empire as a theoretical framework
The final layer of the book marks a qualitative shift – a synthetic conceptual move captured by the notion of an “AI empire.” Here, Hao seeks to grasp the underlying logic of the AI industry through an overarching metaphor. In this sense, the concept of empire is not a mere rhetorical flourish, but an interpretive attempt. Hao argues that the operations of OpenAI and similar actors exhibit patterns that are structurally reminiscent of historical empires. This is not empire in the classical sense of direct political rule or military domination, but a configuration of power exercised through economic, technological, and cultural means. The book identifies four such dimensions which, taken together, constitute the logic of the “AI empire.”
The first dimension is ideological justification. According to Hao, the universal and deliberately vague mission of AGI – artificial general intelligence that benefits all of humanity – functions in a way analogous to the “civilizing missions” of historical empires. This narrative simultaneously mobilizes talent and capital while providing a moral framework for decisions that, in the short term, generate tangible social harms. The familiar “good empire versus bad empire” framing – often articulated through competition with China – is not a peripheral rhetorical device, but a central source of institutional legitimation.
The second dimension concerns the appropriation of resources. Here Hao includes not only the large-scale extraction of data from the internet, but also natural resources: energy and water, as well as the spatial concentration of infrastructure. In this reading, the AI industry is built upon resources over which it holds no direct ownership, yet which it incorporates into its operations on a global scale. This process is not necessarily illegal, but it is structurally similar to the ways in which historical empires drew resources from their peripheries to sustain their centers.
The third dimension is the structural exploitation of labor. Hao addresses both global data and content moderation work and the longer-term labor market effects of automation. “Ghost work” is not an isolated anomaly, but a foundational condition of AI development, even as the resulting systems are premised on promises of labor displacement. In this tension, Hao identifies a power asymmetry that goes beyond individual abuses and points toward a reconfiguration of the global division of labor.
The fourth dimension is the monopolization of knowledge and interpretive frameworks. According to Hao, the increasing concentration of AI expertise and research capacity entails not only economic power but cultural authority as well. When key researchers and dominant narratives are absorbed into corporate settings, public understanding of the technology is inevitably filtered through institutional lenses. This knowledge monopoly does not operate through formal censorship, but through unequal distributions of visibility, access, and epistemic authority.
In this sense, Hao’s empire metaphor does not assert a historical equivalence between past and present, but draws attention to structural similarities. At the same time, the book does not fully develop the analogy. The only explicit historical parallel is the British East India Company, and even this comparison remains sketchy. While this approach allows the concept of empire to function as a provocative interpretive frame, it also leaves open the question of whether we are truly dealing with a new form of empire, or rather with familiar logics of digital capitalism repackaged in a more dramatic register.
The metaphor becomes most incisive – and most vulnerable – when set against the solutions Hao proposes. It is here that the distinction between diagnosis and normative overreach comes most clearly into view.
Beyond diagnosis
Hao’s proposed remedies are organized around dismantling the concentration of power that underpins the AI empire. Her starting point is that power in the AI industry coalesces along three main axes: knowledge, resources, and influence. The dissolution of the empire, therefore, can only be imagined through the simultaneous redistribution of all three.
According to Hao, one of the central pillars of the AI empire is the corporate concentration of knowledge production and interpretation. To counter this, she calls for strengthening independent research. This includes supporting AI research that is not funded or controlled by corporations – particularly institutions that do not regard the scaling of large language models as the sole path forward, but instead pursue alternative, more data- and energy-efficient approaches.
Hao also emphasizes the importance of community-based, task-specific AI applications that do not promise universal technological solutions, but respond to the needs of particular communities. Closely related is the demand for independent evaluation of corporate models, so that society does not rely solely on companies’ own claims regarding the capabilities and risks of AI systems. The redistribution of knowledge thus presupposes transparency: public insight into training data, computational infrastructure, and the real-world impacts of deployed models.
The second cluster of proposals targets the material foundations of the AI industry. Hao argues that AI empires secure their power through the accumulation of resources – data, labor, energy, and capital – and that without redistributing these, all other reforms remain superficial. A key element here is the strengthening of labor protections: not only for data annotators and content moderators, but also for creators whose copyrighted work has been absorbed into training datasets. Trade unions and collective bargaining, Hao suggests, play a crucial role in establishing negotiating power in the face of automation.
Hao argues that AI empires secure their power through the accumulation of resources – data, labor, energy, and capital – and that without redistributing these, all other reforms remain superficial.
These proposals are closely connected to updating intellectual property regulation and rendering supply chains transparent. Hao argues that public disclosure of training data sources and greater visibility into infrastructure investments would make extractive, “invisible” forms of resource appropriation far more difficult. Finally, she also calls for consumer-protection–style oversight of AI products – regulatory mechanisms focused not on abstract, hypothetical risks, but on the systems’ actual social impacts.
The third axis concerns the question of influence. According to Hao, the persistence of the AI empire is sustained not only by the concentration of resources, but also by the “magical” aura that surrounds the technology. Dismantling this aura requires broad-based education – not training people to code, but enabling them to understand how AI systems work, what their limits are, and which worldviews and assumptions shape their development. The aim is to strip AI of its unquestionable authority and to render it an object of democratic contestation.
In this sense, Hao treats AI governance as fundamentally political rather than technical. The central question is whether artificial intelligence concentrates power or redistributes it. The dismantling of imperial ideology, she argues, is only possible if this question is made explicit within regulation, education, and public discourse alike.
Normative force, structural constraints
Hao’s proposals are internally coherent and normatively compelling. Yet even on a first reading, the question arises whether they amount to a realistically actionable program, or rather to the delineation of a critical horizon. The redistribution of knowledge, resources, and influence is not merely a matter of policy design; it confronts deeply entrenched economic interests, regulatory limitations, and – no less importantly – widespread social indifference.
In normative terms, Hao’s prescriptions are difficult to dispute. A program centered on redistributing knowledge, resources, and influence follows logically from the problems she diagnoses. If AI indeed operates according to an imperial logic, then breaking the concentration of power is not an optional ambition, but a necessary one.
However, a tension emerges between diagnosis and therapeutic realism. Hao’s proposals implicitly assume that power redistribution is primarily a matter of regulatory and institutional will. The difficulty is that the imperial logic she herself describes explains precisely why this assumption does not hold automatically.
The issue is not whether alternative forms of AI research can exist, but under what conditions they could become relevant counterweights
Consider the redistribution of knowledge. Proposals to strengthen independent research encounter structural limits in a scientific and funding environment where the largest infrastructures, data assets, and computational capacities are already concentrated in corporate hands. Supporting independent research is normatively desirable, but it remains structurally marginal as long as it cannot meaningfully compete with the logic of scale. The issue is not whether alternative forms of AI research can exist, but under what conditions they could become relevant counterweights – a question Hao addresses only indirectly.
Hao is right to stress that the aura of technological magic sustains AI’s social authority. Yet she reflects less on the extent to which broad-based technological literacy and critical understanding collide with public apathy. For most users, AI is not a political issue but a convenience tool, and this indifference itself becomes a stabilizing force.
It is at this juncture that one of the book’s core tensions becomes fully visible. Hao persuasively demonstrates that the AI industry operates as a system driven by structural constraints. Her proposed remedies, however, often read as though these constraints could be partially circumvented through normative interventions. This does not render the proposals irrelevant, but it shifts their register: less a concrete program of action than a moral compass.
In this sense, Empire of AI offers not so much a roadmap as a critical horizon. The book’s greatest strength lies in making clear that the future of artificial intelligence is not a technological inevitability, but the outcome of political and social decisions, conflicts, and struggles. Where those decisions will ultimately lead – however strong our intuitions may be – remains, indeed, “too early to tell.”
Overall, Empire of AI is a rare book in its ability to address multiple audiences at once. For readers newly encountering artificial intelligence, it provides an accessible and compelling entry point into the social and political dimensions of the industry. For researchers already immersed in the field, it remains a stimulating read thanks to its empirical richness and deliberately provocative arguments. One may not agree with all of Hao’s claims – but it is precisely the combination of thorough documentation and sharply articulated positions that makes this a book worth engaging with seriously.
Featured image from CANVA.
Blog post licensed under CC BY-NC-SA 4.0
Comments