Out of these unprecedented ownership claims over the means of digital production, a new system of class relations could now arise. This article analyzes the social divisions associated with the ascendancy of the tech industry and the reorganization of social processes through algorithms2 —sets of instructions written as code and run on computers. In the next section, The Rise of the Coding Elite and the Pre-Automation of Everyone Else, we argue that the core divide in digital capitalism opposes what we call the coding elite, who hold and control the data and software, and the cybertariat, who must produce, refine, and work the data that feed or train the algorithms, sometimes to the point of automating their own jobs and making themselves redundant. We also show that claims of technical and economic efficiency, as well as fairness, are an important component of the coding elite's societal power. [...] In the section titled Classifiers and Their Discontents, we show that algorithmic processes also structure how people come to know and associate with one another, and how technical mediations intersect with the perception and production of self and community. [...] # The Coding Elite: Power at Scale A new elite occupies the upper echelons of the digitized society—a class or proto-class that, in a self-conscious nod to Mills (2000), we call the coding elite. The coding elite is a nebula of software developers, tech CEOs, investors, and computer science and engineering professors, among others, often circulating effortlessly between these influential roles. [...] Professors circulate between their own start-ups, key positions in large firms, government-sponsored research labs, and classrooms. Most valued in this world are those people who touch and understand computer code. Most powerful are those who own the code and can employ others to deploy it as they see fit. Mastery of computational techniques bestows special kinds of powers. These powers are at once cultural, political, and economic. [...] no profession, no matter how prestigious or how high the barriers to entry, is exempt from having its judgment subject to a second (algorithmic) opinion, if not wholly supplanted by it. Legitimacy has been displaced from the professional to the coder-king—and, increasingly, to the algorithm. [...] # Cybertarians of the World, Disunited If industrial capitalism concealed labor's existence through the fetishism of commodities, digital capitalism intentionally conceals it through the fetish of artificial intelligence (AI) and feigned automation. The smooth functioning of on-demand apps, search engines, mapping sites, social media websites, and even autonomous vehicles and many other products all depend on the collective intelligence of armies of humans performing ghost work.[...] What stands beneath the fetish of AI is a global digital assembly line of silent, invisible men and women, often laboring in precarious conditions, many in postcolonies of the Global South. A new class of workers stands opposite the coding elite: the cybertariat. [...] One of the distinguishing features of digital capitalism is its reliance on free labor. [...] It may be more difficult for the cybertariat to resist the extraction of their labor by undertaking shop-floor organizing, as did the proletarians of the past. The material basis of their work situation precludes it. Resistance is actively underway nonetheless, for example, in online forums organized off-site. Coders allied to cyber workers' rights also play a role, creating tools to aid cybertarians in information sharing and self-organizing. [...] Fairness may be the most hotly debated topic in machine learning today, which often leads to complex arguments about which statistical criterion best fits the situation: false negatives versus false positives, or demographic parity versus predictive rate parity (Weinberger 2019, Narayanan 2019). Some critics reject outright the claim that mathematical objectivity is inherently better at guarding against social inequities than human judgement, however subjective the latter may be. Eubanks (2017, p. 168) insists on the fundamental role of empathy in the delivery of social services: “the assumption that human decision-making is opaque and inaccessible is an admission that we have abandoned a social commitment to try to understand each other.” Echoing this sentiment (and turning Max Weber on his head), Pasquale (2019) concludes that a rule of persons is better able to guarantee legal due process than a rule of machines. # The Algorithmic Dominion Inclusion into some identification database has long been a prerequisite of modern citizenship [...] In India, Aadhaar, an integrated identification system that stores fingerprint and iris scans along with demographic data for each citizen, was originally publicized as a tool for graft elimination and the efficient delivery of welfare services. It has swiftly become required for interactions with both public and private institutions, anchoring an emergent mass surveillance infrastructure. (Rao & Nair 2019). In South Africa, the postapartheid government similarly sought to implement a nationwide biometric identification system to improve the uniformity of social welfare grant disbursement. In typical Weberian fashion, the government claims that the system's universalism and standardization guarantee equal treatment (Donovan 2015). Despite the country's history of oppressive information infrastructures, most notoriously its passbook system, this new citizen database was embraced by postapartheid leaders. In China, both municipalities and the central government have partnered with private sector firms to develop social credit systems oriented to improving the financial behavior and civic-mindedness of individuals and organizations (Ahmed 2019, Liu 2019, Ohlberg et al. 2017). By linking algorithmically produced social credit scores to tangible outcomes (conveniences and perks, public praising or shaming), these systems foster rule compliance (e.g., using crosswalks to cross the street) and obedience to social expectations (e.g., taking care of one's parents, doing volunteer work). While Western commentators have often interpreted the development of social credit through the lens of China's political authoritarianism, it is useful to remember that private data infrastructures elsewhere can feel similarly oppressive and inescapable. O'Neill (2016) describes, for example, how a job applicant was shut out of work in a sizeable portion of the American retail industry when he failed a hiring prescreening test designed by a software company with contracts throughout the sector. Other data systems operate ubiquitously across national borders. Euro-centric assumptions built in to cybersecurity tools that automate the identification of fraud, for example, have become a ubiquitous part of the global infrastructure (Jonas & Burrell 2019). [...] In these examples, fair allocation is not the only issue. The inability of those so forcefully governed to shape the terms of the algorithmic dominion, or to evade the rule of the code, raises fundamental questions about democracy and human autonomy (Amoore 2020, Aneesh 2009). [...] A proper critique must thus begin with the recognition that algorithms are ethico-political entities that generate their own “ideas of goodness, transgression and what society ought to be” (Amoore 2020, p. 7). In other words, algorithms are transforming the very nature of our moral intuitions—that is, the very nature of our relations to self and others—and what it means to exist in the social world. The next section examines this shifting terrain. [...] the digital infrastructure operates in increasingly totalizing, continuous, and dynamic ways. Not only do digital data traces allow for intrusive probing by institutions far afield from the data's original collection site (e.g., credit data matter to landlords and to prospective romantic partners, and police departments are hungry for social media data), but they also enable the guiding or control of behavior through reactive, cybernetic feedback loops that operate in real time. The more one interacts with digital systems, the more the course of one's personal and social life becomes dependent on algorithmic operations and choices. This is true not only for the kinds of big decisions mentioned above but also for mundane, moment-by-moment actions: For instance, each online click potentially reveals some underlying tendency or signals a departure from a previous baseline. As new data flow in, categories and classifications get dynamically readjusted, and so do the actions that computing systems take on the basis of those categories. This has important implications for how people ultimately perceive themselves and for how social identities are formed. [...] Inferences made about us are often fed back to us as visualizations, assessments, scores, or recommendations and, in turn, reconfigure how we understand ourselves in real time. [...] the metrics, calculators, and visualization tools that presumably tell our personal truth are not of our own making. However inaccurate, daily step counts, menstrual cycles, heart rate, emotional states, social networks, and spending patterns are reflected back to us, to institutional others (e.g., doctors, insurance companies, welfare agencies), and to the world as undeniable evidence of who we are over and above subjective self-assessment or the old techniques of analog self-presentation [...] Monitoring and investigating our sleep patterns, eating habits, and social relations this way is slowly becoming second nature. What may sometimes feel like playful self-diagnosis is really no play at all, however—rather, it is a permanently self-probing condition, powered by incessant feedback loops between human and machine. [...] The endgame of the coding elite, the ultimate goal of their professional project, like the algorithms they build, remains opaque.[...] AI's trajectory in society, however, is not simply a question of whether humanity will benefit or not but, rather, who will benefit. A new division of learning opposes the knowers against the known (Zuboff 2019); those who make AI work face those who make AI work for themselves. Unlike the mass of those surveilled, those misrepresented and alienated, the data capitalists may be able to correct, control, or improve their personal data representation; to buy themselves entirely out of surveillance regimes; or to benefit from AI in new ways. [...] For now, dominant industry talk promises a gentler, more acceptable, less biased kind of AI, compliant with best practices and ethically infused (Crawford & Calo 2016). But for all the great chatter about equity and value alignment (Gabriel 2020), the established technological trajectory has remained secure: Venture capital offices and start-up firms continue to roll out a world of ubiquitous computing, located in everything from human bodies and mundane objects to city infrastructures and other large works of engineering. [...] ____ Tratto da https://www.annualreviews.org/doi/full/10.1146/annurev-soc-090820-020800 Si tratta di una analisi molto chiara e ben strutturata, sebbene non consideri aspetti tecnici e politici essenziali. Si confonde algoritmi e software. Si fraintende il loro funzionamento ("a new class of algorithms (deep learning, an evolution of neural network models) exploits this abundance by drawing direct inferences from the data"). Si nota come questa nuova "coding élite" dipenda fortemente dal lavoro non pagato, ma non si menziona l'open source. Il movimento hacker non viene contrapposto a queste "coding élite" (ed anzi viene riproposta la propaganda della Silicon Valley che si giustifica e nobilita ponendosi in continuità con esso) e la nostra azione politica si riduce a "Coders allied to cyber workers' rights". Così come non viene fatta menzione delle implicazioni geopolitiche, degli squilibri e delle tensioni che caratterizzano la società cibernetica globale contemporanea. Si tratta insomma di una ottima analisi che però prescinde dalla comprensione della tecnologia di cui cerca di descrivere l'impatto. Nonostante ciò, già solo basandosi sulle osservazioni esogene delle dinamiche di sociali osservabili, ne trae conclusioni che credo tutti i tecnici qui condivideranno (a riconferma delle conclusioni stesse): ``` It would be a mistake to uncritically embrace the fever dreams of the coding elite's most fervent boosters, to treat computing “theologically rather than scientifically or culturally”. In that respect, sociology offers a useful reality check. Ethnographers who observe digital technology in action have brilliantly tackled the unglamorous everyday realities of algorithms. They have documented considerable resistance to algorithmic systems, frequent errors and breakdowns, and variations in meaning and effect, both across and within societies. We can both reject magical thinking about machine intelligence and acknowledge the enormous economic, political, and cultural power of the tech industry to transform the world we live in. Beyond futurism and hype, existing AI is actually quite mundane. It is designed by the coding elite, sustained by the cybertariat, fueled by personal data extracted by (mainly) large digital firms, frequently optimized for profit maximization, and supported by a contingent set of legal institutions that authorize (at the time of this writing) continuous data flows into corporate as well as state servers. Like prior control innovations, AI surveils, sorts, parses, assembles, and automates. And like prior forms of social surveillance and discipline, it weighs differently and more prejudicially on poor and minority populations. Far from being purely mechanistic, it is deeply, inescapably human. ``` Giacomo