# Artificial Intelligence Hub > [!info] Short version > New to the topic? [[Artificial Intelligence]] is the one-page orientation: the working definition, the key distinctions, and where to go next. This hub is the full map. Artificial intelligence is not just a technology. It is a power-concentrating force — one that channels capital, compute, data, and decision-making authority toward an extraordinarily small number of companies and individuals, many of whom are documented extensively elsewhere in this vault. That makes AI both a subject worth understanding on its own terms and a thread that runs through nearly every other subject area here: the [[Economics Hub]]'s story of wealth concentration, the [[Authoritarianism Hub]]'s analysis of surveillance and control, the [[Disinformation Hub]]'s mapping of synthetic media, the [[Media Theory Hub]]'s account of algorithmic attention capture. To understand the contemporary right-wing ecosystem without understanding AI is to miss the infrastructure. This hub does two things. First, it provides the technical grounding — enough to follow the debates, read the papers' abstracts, and understand what the technology actually does and does not do. Second, it maps the political economy: who builds AI, who funds it, who deploys it, who regulates it (or refuses to), and how those decisions connect to the power structures this knowledge base documents. The discipline pages in `/Subjects/Artificial Intelligence/` explain the field honestly; the political analysis lives in `/Concepts/`, `/Companies/`, `/Ecosystems/`, and `/Ideologies/`. For this knowledge base's purposes, AI sits at the intersection of three dynamics documented elsewhere in the vault. It is a tool of economic concentration (the [[Economics Hub]]'s concern), a tool of surveillance and social control (the [[Authoritarianism Hub]]'s concern), and a tool of epistemic disruption — the capacity to manufacture synthetic reality at scale (the [[Disinformation Hub]]'s concern). The people building it are, in disproportionate numbers, the same people funding the political movements this vault tracks. That overlap is not incidental. It is the subject. ## The technology landscape The field divides historically into two paradigms whose competition organizes its entire intellectual history (see Connectionism). [[Symbolic AI]] — "good old-fashioned AI" — treated intelligence as the manipulation of explicit symbols and logical rules. It dominated from the Dartmouth Workshop in 1956 through the Expert Systems boom of the 1980s and the subsequent collapse known as [[AI Winter]]. [[Machine Learning]], the rival paradigm, treats intelligence as patterns learned from data rather than hand-coded by programmers. Its [[Deep Learning]] branch, built on multi-layer [[Neural Network]]s trained via Backpropagation and Gradient Descent, drove the breakthroughs of the 2010s and underlies essentially all of today's frontier AI. The war has an unresolved third act: [[Neurosymbolic AI]], the attempt to bolt symbolic reasoning back onto neural systems in order to recover the auditability that scale gave away. The modern landscape is organized around a handful of capabilities, each with distinct political implications. [[Natural Language Processing]] underpins the large language models — the [[Large Language Model]]s and [[Foundation Model]]s — that power conversational AI, code generation, and increasingly autonomous agents. These models are the most visible face of the current AI boom, and the most immediately consequential for labor markets, education, media, and law. [[Computer Vision]] enables [[Facial Recognition]], autonomous vehicles, satellite imagery analysis, and the battlefield perception systems built by [[Palantir]] and [[Anduril]]. Of all AI subfields, computer vision is the one most directly implicated in state surveillance — the technology that turns cameras into instruments of identification and control. [[Generative AI]] produces text, images, audio, and video, creating both extraordinary creative tools and an industrial-scale capacity for synthetic disinformation. The political implications are explored in the Disinformation subject area, but the technical capabilities — voice cloning, face synthesis, video generation — originate here. Reinforcement Learning drives game-playing systems, robotics, and the [[RLHF]] technique used to align language models with human preferences. [[Robotics]] is where AI meets the physical world — manufacturing, logistics, military, and the open question of what happens to human labor when machines can manipulate objects as fluently as they now manipulate language. The architectural revolution behind the current boom is the [[Transformer]], introduced in 2017. Combined with massive training datasets, enormous compute budgets, and techniques like [[Attention Mechanism]]s, [[Embeddings]], and [[Tokenization]], the Transformer architecture enabled the [[Scaling Laws]] that showed performance improving predictably with more data and compute — a finding that turned AI development into a capital-expenditure arms race. The Convolutional Neural Network cracked computer vision; the Recurrent Neural Network was the earlier approach to sequential data; the Diffusion Model powers modern image generators; and the Generative Adversarial Network was the original engine of synthetic media. For the full technical vocabulary, see the [[AI Lexicon]]. Understanding the history matters because the field's past failures are instructive about its present hype. The two historical [[AI Winter]]s — the mid-1970s collapse after the Lighthill Report and the late-1980s collapse after the Expert Systems bubble burst — share a common pattern: extravagant promises, institutional overinvestment, and a correction that destroyed funding for even the productive research. Whether the current boom follows the same trajectory is the most important open question in the field. The [[Scaling Laws]] that have driven progress may plateau; the economic returns may disappoint the investment; the regulatory environment may shift. The History of Artificial Intelligence page and the [[Artificial Intelligence Timeline]] document the full arc. ## The political economy of AI The most important thing to understand about AI in 2026 is not how it works but who controls it. Training a frontier model requires hundreds of millions of dollars in compute, proprietary datasets of unprecedented scale, and a talent pool so small that a few hundred researchers moving between a handful of companies can reshape the entire field. This means AI development is concentrated — structurally, inevitably, and increasingly — in the hands of a very small number of corporations: [[OpenAI]], [[Google]] (via [[Google DeepMind]]), [[Meta]] (via [[Meta Superintelligence Labs]]), [[Anthropic]], [[Microsoft]], and [[xAI]]. The hardware layer is even more concentrated: [[NVIDIA]] holds a near-monopoly on the [[Graphics Processing Unit]]s required for training, and the semiconductor supply chain funnels through a handful of fabrication facilities whose geopolitical significance is documented in [[Chip Export Controls]] and [[Compute Sovereignty]]. This concentration maps directly onto the vault's broader analysis of [[Oligarchy]] and [[Corporate Capture]]. The people who control frontier AI are not anonymous engineers — they are, in many cases, the same figures who appear throughout this knowledge base as political actors, ideological entrepreneurs, and donors. [[Elon Musk]] funds [[xAI]], controls the platform formerly known as Twitter, leads [[DOGE]], and sits at the center of the [[Musk-Thiel Funding Map]]. [[Peter Thiel]] co-founded [[Palantir]], bankrolls the [[Thiel Fellowship]] and [[Founders Fund]], and helped build the political careers of [[JD Vance]] and [[David Sacks]]. [[Marc Andreessen]] published the [[Techno-Optimist Manifesto]] and runs [[Andreessen Horowitz]], one of AI's largest venture investors. [[Sam Altman]] steered [[OpenAI]]'s transition from nonprofit research lab to capped-profit corporation — a trajectory documented in the [[OpenAI Timeline]] and contested in the [[Musk v. OpenAI Trial]]. [[Jeff Bezos]] funds [[Anthropic]] through [[Amazon]]; [[Mark Zuckerberg]] has bet [[Meta]]'s future on open-weight models and defense contracts. The pattern is not subtle: the same people building the most powerful technology in a generation are also funding political campaigns, lobbying for deregulation, and in some cases running government agencies. The [[PayPal Mafia]] genealogy, the [[PayPal Mafia to Pentagon Pipeline]], and the [[Silicon Valley AI Ecosystem]] map trace these connections in detail. The [[Open Source AI]] debate adds another dimension. [[Meta]] has released large open-weight models, positioning itself as the democratic alternative to closed labs. But "open" is a spectrum — weights can be released while training data, RLHF procedures, and compute infrastructure remain proprietary — and the strategic logic is as much about undermining competitors' moat as about democratizing access. Meanwhile [[DeepSeek]]'s breakthroughs from China have intensified the [[AI Arms Race]], raising the geopolitical stakes of who leads in frontier capability and reinforcing the security framing that pushes AI governance toward national-security institutions and away from democratic accountability. The [[Stargate]] joint venture — a massive compute-infrastructure deal between [[OpenAI]], [[SoftBank]], and the federal government — and the [[Golden Dome]] missile defense initiative represent the apotheosis of this trend: public money underwriting private AI infrastructure justified by national security. ## AI and the right-wing ecosystem Three domains connect AI most directly to the right-wing power structures this vault documents. **Defense technology** is where AI becomes a weapon system. [[Palantir]]'s software runs intelligence analysis for the CIA, immigration enforcement for ICE (see [[Palantir ICE Systems]]), and battlefield decision-making for the Pentagon — its [[Palantir Product Ecosystem]] spans the full spectrum of state coercion. [[Anduril]], founded by [[Palmer Luckey]] with [[Trae Stephens]] from [[Thiel Foundation]], builds autonomous drones, perimeter surveillance, and the [[Anduril Lattice]] command system. Both companies sit at the center of the [[Defense Tech Ecosystem]] and the broader [[Military-Industrial Complex 2.0]] — the new generation of venture-backed defense contractors that replaced Lockheed Martin–style cost-plus contracting with Silicon Valley speed-and-scale logic. [[Google Defense]] and [[Meta Defense]] mark the point at which companies that once refused Pentagon contracts reversed course. The [[Palantir Mafia]] traces how Palantir alumni have seeded the next generation of defense tech startups. The political spending is documented across `/Deals/` and the [[Defense Tech Conflicts of Interest]] page. **Surveillance and social control** is the second domain. [[Computer Vision]] feeds [[Facial Recognition]], which feeds the [[Surveillance State]]. [[Clearview AI]] scraped billions of photos from the internet to build a facial recognition database marketed to law enforcement without the knowledge or consent of the people in those photos. [[Palantir ICE Systems]] used data integration to identify, track, and target undocumented immigrants — a deployment that connected the technical capacity of AI to the enforcement priorities of the state in ways that would have been impossible a decade earlier. The [[Surveillance Capitalism]] framework — [[Shoshana Zuboff]]'s argument that tech companies extract behavioral surplus from users and sell predictions of their behavior — provides the economic logic: surveillance is not a side effect of AI but a business model. The data that feeds AI systems is the same data that feeds advertising targeting, insurance risk-scoring, and law enforcement profiling. These dynamics are explored further in the [[Surveillance Lexicon]]. **Disinformation and algorithmic radicalization** is the third. [[Generative AI]] makes synthetic media cheap, fast, and increasingly indistinguishable from reality — text, images, audio, video. The [[Online Radicalization Pipeline]] documents how recommendation algorithms funnel users from mainstream content toward increasingly extreme material. The intersection of generative AI with [[Disinformation]] and [[Propaganda]] is covered in the Disinformation subject area, but the enabling technology lives here. A fourth connection cuts across the other three: the **[[DOGE-Palantir Nexus]]** and the broader pattern of AI companies gaining direct access to government data and decision-making. [[DOGE]]'s access to federal databases, [[Palantir]]'s contracts with immigration enforcement and military intelligence, and the [[Revolving Door]] between AI companies and government agencies all represent a kind of [[Corporate Capture]] that is specific to the AI era — not just lobbying for favorable regulation but installing private AI systems at the center of state operations. The [[Defense Tech Conflicts of Interest]] page documents specific cases where the people making procurement decisions are the same people who stand to profit from them. ## AI governance and the regulatory debate AI governance is contested terrain where technical uncertainty, economic interest, and ideological commitment collide. The [[EU AI Act]] represents the most comprehensive attempt at regulation: a risk-based framework that categorizes AI systems and imposes requirements scaling with potential harm. The United States, by contrast, has largely pursued voluntary commitments and executive action rather than binding legislation. The [[AI Action Plan]] and related policy documents reflect the current administration's approach — one that prioritizes American AI dominance over precautionary regulation. The [[National Security Commission on Artificial Intelligence]] recommended treating AI as a strategic asset on par with nuclear weapons, a framing that pulls governance toward national security and away from civil liberties. The deeper fault line is ideological. The safety-versus-acceleration debate splits the AI world into camps that map imperfectly but recognizably onto the vault's ideological taxonomy. [[Effective Altruism]], [[Longtermism]], and the [[AI Safety]] community argue that [[AGI]] poses an [[Existential Risk]] — that poorly [[Alignment]]ed superintelligent systems could cause civilizational catastrophe — and that regulation, safety research, and even development pauses are warranted. [[Nick Bostrom]]'s [[Superintelligence (Book)]] is the canonical statement of this concern; [[Eliezer Yudkowsky]] and the [[Machine Intelligence Research Institute]] represent its most urgent expression; [[AI Doomerism]] is what critics call the more extreme versions. On the other side, [[Effective Accelerationism]] and the [[Techno-Optimism]] championed by [[Marc Andreessen]]'s [[Techno-Optimist Manifesto]] argue that AI development should be accelerated as fast as possible, that safety concerns are exaggerated or actively harmful, and that regulation amounts to incumbent protection. The [[Dark Enlightenment]] and [[Neoreaction]] currents in Silicon Valley provide a more radical ideological substrate — the argument that democratic governance is itself the problem and that technologically empowered elites should operate outside democratic constraints. This is not a fringe position in the AI world; it describes, with varying degrees of explicitness, the operating assumptions of several of the vault's most-documented figures. The [[Open Source AI]] debate cuts through the governance question in a different direction. Proponents argue that open models — where weights, and sometimes training code and data, are publicly released — democratize AI by preventing any single company from monopolizing the technology. Critics argue that releasing powerful models without guardrails makes misuse inevitable and that "open" is itself a contested term when the compute required to train these models remains inaccessible to anyone outside a handful of corporations. The tension between openness and safety, between democratization and control, has no clean resolution — it mirrors the broader tension in the vault between [[Libertarianism]]'s suspicion of regulation and the documented consequences of unregulated corporate power. The [[Regulatory Capture]] dynamic complicates both sides: AI companies lobby for regulatory frameworks that advantage incumbents (who can afford compliance) over newcomers, while simultaneously arguing against regulation in principle. The [[Antitrust]] question looms: should AI companies that control the infrastructure, the models, the data, and increasingly the distribution channels be treated as monopolies? The result, for now, is a governance vacuum in which the most consequential technology of the century is shaped primarily by the private decisions of a few corporations and the ideological commitments of their founders. ## AI and labor The labor question is where AI's economic implications become most tangible for most people, and where the technology intersects most directly with the [[Economics Hub]]'s analysis of inequality. The standard economic argument is familiar: AI automates tasks, not whole jobs; displaced workers retrain into new roles created by the technology; productivity gains raise living standards across the board. This argument has historical precedent — the mechanization of agriculture, the computerization of office work — but the current wave differs in two important respects. First, AI is automating cognitive and creative tasks that were previously assumed to be automation-proof: writing, analysis, code, design, legal reasoning, medical diagnosis. The categories of work that remain unambiguously safe from AI disruption are shrinking with each model generation. Second, the gains from AI-driven productivity are accruing overwhelmingly to capital — to the companies that own the models and the investors who fund them — rather than to the workers whose tasks are being automated. The [[Scaling Laws]] dynamic reinforces this: the bigger the model, the more expensive to train, the fewer the organizations that can compete. The [[Defense Tech as Asset Class]] framing extends this logic: AI-driven defense is not just a military strategy but an investment thesis, one that channels public expenditure into private returns. Robotic Process Automation automates the white-collar back office; [[Generative AI]] automates the creative front office; autonomous systems automate the factory floor and the battlefield. The distributional question is not abstract. When a company deploys an AI system that does the work previously done by fifty employees, the productivity gains go to the company's shareholders and the model's developers — not to the fifty workers who now need new jobs. At scale, across every industry simultaneously, this dynamic produces a version of the inequality story the [[Economics Hub]] documents: gains flowing upward, costs flowing downward, with the political system unable or unwilling to redistribute. The question is not whether these transitions will happen but who will capture the value — and what political arrangements will govern the distribution. ## AI ethics, bias, and accountability AI systems encode the biases present in their training data, the assumptions of their designers, and the incentive structures of the organizations that deploy them. This is not a bug to be fixed but a structural feature of how machine learning works: a system trained on historical data will reproduce historical patterns of discrimination unless actively designed not to — and the design choices themselves are political. [[Algorithmic Bias]] is the umbrella concept, but the research is more specific. [[Facial Recognition]] systems have been shown to perform dramatically worse on darker-skinned faces, a finding with direct consequences when those systems are deployed in policing. Language models absorb and reproduce cultural stereotypes. Hiring algorithms trained on historical data discriminate against women and minorities. Credit scoring, sentencing recommendations, child welfare risk assessments — every domain where AI replaces or augments human judgment is a domain where encoded bias can scale. [[Timnit Gebru]] and [[Emily Bender]]'s work on the risks of large language models — the [[Stochastic Parrot]] critique — argued that massive models trained on internet text amplify existing inequalities and that the AI industry's race to scale was outrunning its capacity to understand what it was building. Gebru's firing from Google (see [[Google AI Ethics Firing]]) became a defining event in AI ethics: a demonstration that the companies building these systems were structurally unwilling to tolerate internal criticism of their core business model. The accountability gap is the deeper problem. When an AI system makes a consequential decision — denying a loan, flagging a person for surveillance, recommending a prison sentence — the question of who is responsible has no clear answer. The developer? The deployer? The regulator who approved it? The dataset that trained it? [[Interpretability]] research aims to make model decisions legible, but frontier models remain substantially opaque. [[Constitutional AI]] and [[RLHF]] are alignment techniques that attempt to build values into model behavior, but they reflect the values of the organizations that implement them — which brings the analysis back to the political economy of who controls AI. The material costs of AI are also unevenly distributed. Training frontier models consumes enormous quantities of electricity and water for cooling. The data that feeds them is scraped from the internet — from the creative work of writers, artists, and musicians who were never asked and never compensated. The hardware depends on rare earth minerals extracted under conditions that are, in many supply chains, exploitative. The human labor required to label training data and filter harmful content is disproportionately outsourced to low-wage workers in the Global South. AI is not weightless or free; its costs are simply displaced onto people and places with less power to object. ## Key figures **AI pioneers:** [[Alan Turing]] (the founding question of machine intelligence), [[John McCarthy]] (coined "artificial intelligence," organized the Dartmouth Workshop), [[Marvin Minsky]] (symbolic AI, the perceptron critique), [[Frank Rosenblatt]] (invented the Perceptron), [[Geoffrey Hinton]], [[Yann LeCun]], and [[Yoshua Bengio]] (the "godfathers" of deep learning), [[Fei-Fei Li]] (ImageNet and the computer vision revolution), [[Judea Pearl]] (causality and probabilistic reasoning), [[Ilya Sutskever]], [[Andrew Ng]], and [[Andrej Karpathy]] (modern deep learning leaders), [[Demis Hassabis]] ([[Google DeepMind]]). **Industry power players:** [[Sam Altman]] ([[OpenAI]]), [[Elon Musk]] ([[xAI]], [[DOGE]]), [[Mark Zuckerberg]] ([[Meta]]), [[Sundar Pichai]] ([[Google]]), [[Jensen Huang]] ([[NVIDIA]]), [[Jeff Bezos]] ([[Amazon]]), [[Dario Amodei]] and [[Daniela Amodei]] ([[Anthropic]]), [[Alex Karp]] ([[Palantir]]), [[Palmer Luckey]] ([[Anduril]]). **Political-economy connectors:** [[Peter Thiel]] ([[Palantir]], [[Founders Fund]], [[PayPal Mafia]]), [[Marc Andreessen]] ([[Andreessen Horowitz]], [[Techno-Optimist Manifesto]]), [[David Sacks]] (AI investor, political operator), [[JD Vance]] (Thiel protégé, vice president), [[Trae Stephens]] ([[Anduril]], [[Thiel Foundation]]). **Safety and alignment:** [[Nick Bostrom]] ([[Superintelligence (Book)]]), [[Eliezer Yudkowsky]] ([[Machine Intelligence Research Institute]]), [[Ilya Sutskever]] ([[Safe Superintelligence]]). **Ethics and critics:** [[Timnit Gebru]] ([[Google AI Ethics Firing]]), [[Emily Bender]] ([[Stochastic Parrot]]), [[Shoshana Zuboff]] ([[Surveillance Capitalism]]). **Venture and institutional connectors:** [[In-Q-Tel]] (CIA venture arm, early Palantir investor), [[Founders Fund]] (Thiel's VC firm), [[Andreessen Horowitz]] (major AI investor), [[SoftBank]] ([[Stargate]] partner), [[Scale AI]] (data labeling, defense contracts). ## Essential works [[Superintelligence (Book)]] (Bostrom, 2014) — the canonical statement of the existential risk argument. Laid out the [[Alignment]] problem and the instrumental convergence thesis: a sufficiently intelligent agent will resist being shut off regardless of its original goals. Shaped a generation of safety research and Silicon Valley anxiety. Essential context for understanding the [[AI Safety]] movement and the [[Effective Altruism]] community's influence on AI governance. The other foundational texts in the critical tradition do not yet have their own entries here but are essential reading for this subject area. [[Shoshana Zuboff]]'s *The Age of Surveillance Capitalism* (2019) provides the economic framework: the extraction of behavioral surplus and its sale as prediction products. The definitive analysis of how the tech industry's business model converts human experience into data and data into power — the theoretical backbone of the vault's [[Surveillance Capitalism]] concept page. Kate Crawford's *Atlas of AI* (2021) maps the material infrastructure — the mines, the labor, the water, the energy — that makes AI possible. Against the industry's rhetoric of weightless intelligence, Crawford documents what AI actually requires from the physical world and from the workers who build, train, and maintain it. Safiya Noble's *Algorithms of Oppression* (2018) documented how search algorithms reproduce and amplify racial and gender bias. Cathy O'Neil's *Weapons of Math Destruction* (2016) showed how opaque algorithmic systems in hiring, lending, and policing create feedback loops that punish the already disadvantaged. Both are essential for understanding [[Algorithmic Bias]] as a systemic rather than accidental phenomenon. Kai-Fu Lee's *AI Superpowers* (2018) framed the US-China AI competition as the defining geopolitical contest of the century — context for the [[AI Arms Race]], [[Chip Export Controls]], and [[Compute Sovereignty]] pages. The [[Palantir 22-Point Manifesto]] (2024) belongs in a different category: not a critical analysis but a primary source — a defense-tech company's declaration of its own worldview. Read alongside the critical texts above, it illustrates the ideological gap between the people building AI-powered state infrastructure and the scholars questioning whether they should. ## See also [[Machine Learning]] | [[Deep Learning]] | [[Natural Language Processing]] | [[Computer Vision]] | Reinforcement Learning | [[Generative AI]] | [[Symbolic AI]] | [[Neurosymbolic AI]] | [[Robotics]] [[Transformer]] | [[Neural Network]] | [[Foundation Model]] | [[Large Language Model]] | [[Scaling Laws]] | [[Alignment]] | [[Interpretability]] [[AI Safety]] | [[AI Arms Race]] | [[Compute Sovereignty]] | [[AI Doomerism]] | [[AI Action Plan]] | [[EU AI Act]] | [[Autonomous Weapons]] [[Surveillance Capitalism]] | [[Surveillance State]] | [[Facial Recognition]] | [[Algorithmic Bias]] | [[Online Radicalization Pipeline]] [[Effective Altruism]] | [[Effective Accelerationism]] | [[Longtermism]] | [[Techno-Optimism]] | [[Dark Enlightenment]] [[Silicon Valley AI Ecosystem]] | [[Defense Tech Ecosystem]] | [[DOGE Ecosystem]] | [[AI Sector]] | [[AI Lexicon]] | [[Artificial Intelligence Timeline]] [[PayPal Mafia]] | [[PayPal Mafia to Pentagon Pipeline]] | [[Palantir Mafia]] | [[Musk-Thiel Funding Map]] | [[DOGE-Palantir Nexus]] [[Oligarchy]] | [[Corporate Capture]] | [[Regulatory Capture]] | [[Military-Industrial Complex 2.0]] | [[Neoliberalism]]