The Digitalist Papers (Stanford, Volumes 1–2)

The Digitalist Papers are a Stanford-housed, multi-disciplinary essay project modeled on the Federalist Papers, asking how AI will reshape democratic institutions and the economy and what institutional renewal it calls for. Volume 1 (democracy focus) was published September 22, 2024; Volume 2 (economy, geopolitics, and future focus) began publishing December 11, 2025. The essays are normative and programmatic — they argue for institutional designs rather than establish empirical facts — and the editors designed them as a bundled set of complementary and opposing perspectives. Each essay below is described as a position ("X argues"), not as evidence, and claims from any single essay should be cited accordingly.

The Volume 1 editors are Erik Brynjolfsson, Alex Pentland, Nathaniel Persily, Condoleezza Rice, and Angela Aristidou. The editors' introduction sets the project as Federalist-style institutional design for the AI era and poses two questions to each contributor: (1) how the world is different because of AI and what that means for democratic institutions, governance, and governing; and (2) what the contributor's vision is and the strategy to reach it. The introduction explicitly does not advocate a single position, framing the essays as complementary and sometimes opposing lenses spanning political theory, law, industry practice, and economics.

Volume 1 essays — AI and democracy

"Protected Democracy" — Lawrence Lessig

Lessig argues that U.S. democracy is a vetocracy (a term he draws from Fukuyama) with two structural vulnerabilities that AI will exacerbate: representatives' corrupting dependence on private wealth (campaign funding plus lobbyist-dependent legislative support, dating to Newt Gingrich's defunding of OTA, CRS, and GAO in the 1990s); and polarization (driven by Fox News-era cable business models, engagement-optimized social media, and foreign influence operations). He argues AI will worsen both because the engagement-optimization objective is constitutionally hard to regulate under the First Amendment. He proposes protected democratic deliberation — citizen-assembly-style sortition processes, shielded from the engagement and funding-distortion loop, to resolve issues the regular process cannot. As historical setup, he describes the "broadcast democracy" era (roughly 1936 to the early 1990s), which produced correlated cross-party opinion shifts (the Nixon resignation graph); today's fragmentation, he argues, makes such shifts impossible.

"A Vision of Democratic AI" — Divya Siddarth, Saffron Huang, Audrey Tang

The authors argue that people are fundamentally educable and can rationally self-govern given time, space, and resources. They describe Alignment Assemblies — deliberative processes run by the Collective Intelligence Project (CIP) — including work with OpenAI (participatory risk prioritization with 1,000 demographically representative Americans via the AllOurIdeas wiki-survey, June 2023); with Anthropic (Collective Constitutional AI, in which a publicly drafted constitution yielded a model less biased but equally capable as the researcher-drafted version, with over 75% consensus on free speech and roughly 90% anti-racism); and with Taiwan's moda (Ideathons, Recursive Public AI-governance deliberation, and a Pol.is-mediated information-integrity assembly in March 2024). Taiwan's TAIDE open model uses a constitution from its 2023 Alignment Assembly for alignment-tuning.

"Democracy 2.0" — Eric Schmidt

Schmidt argues that democracy must evolve rather than be replaced by algocracy. He proposes deploying AI across the executive branch ("AI in the situation room" — war-gaming, Putin-chatbot simulations, sentiment and crisis prediction); the judicial branch (COMPAS-style pattern detection that augments rather than replaces human judgment, with models required to be auditable and transparent); and the legislative branch (an AI-augmented Congressional Research Service, legislative drafting, and constituent services). He cites Canada cutting low-risk visa processing times by 87% via AI triage, and the U.S. SNAP eligibility 45% error rate as an efficiency opportunity. He sets three guardrails: humans remain final arbiters and accountable; bureaucratic decisions must remain explainable; and stringent nondiscrimination and privacy controls apply (warning explicitly against China's social credit system and Trump-era social media disability-fraud surveillance). He cites Pol.is in Taiwan as proof-of-concept for AI-scaled deliberative democracy.

"AI, Society, and Democracy: Just Relax" — John H. Cochrane (the "Grumpy Economist")

Cochrane argues that AI regulation threatens democracy more than AI does. He contends nobody has ever successfully predicted technology's social and political effects, and that preemptive regulation has repeatedly destroyed benefits (the Chinese Treasure Fleet in 1430, European GMO bans, 1970s nuclear power, 2001 embryonic stem cells). Drawing on public-choice economics (Hayek, Stigler, Buchanan), he frames regulation as capture: industries use safety fears to entrench incumbents, and worries about hallucinations and deepfakes are, in his view, the latest millenarian panic. His alternative is competition plus rule of law plus trusting civil society, media, and academia to detect problems as they occur. He explicitly targets Biden's EO 14110 and the AI Bill of Rights for smuggling partisan social goals ("racial equity," "equitable clean energy") into technical regulation, and cites Murthy v. Missouri as evidence that regulatory threat chills speech.

"Misunderstanding AI's Democracy Problem" — Nathaniel Persily

Persily argues that AI-panic is itself the democracy problem. He contends the real harm of deepfakes is not persuasion (their reach is tiny: "fake news" was about 0.15% of 2020 Americans' media diet) but the liar's dividend — broad erosion of trust in true content, enabling politicians to disclaim real footage. He notes platforms already remove roughly 4 billion accounts per year for coordinated inauthentic behavior, and argues the network takedowns, not content moderation, matter most. He distinguishes new concerns from the social-media-era playbook: hallucinations and bias matter more as people replace search with single-answer AI (Google's Gemini black-Nazi and female-pope incident; polling-place misinformation addressed via OpenAI/NASS partnerships and Anthropic/Democracy Works TurboVote); and competition, where open-weight models (Meta Llama, xAI Grok, Mistral) democratize access but enable harms such as the CSAM collapse that followed Stable Diffusion's evolution into Unstable Diffusion. He offers four macroprinciples: (1) society cannot "tech its way out" (watermarking is not enough); (2) enforcement and administration are harder than principles; (3) a civil-society auditing ecosystem is needed; and (4) public compute for independent researchers is required.

"Generative AI and Political Power" — Eugene Volokh

Volokh argues that AI's "Public Safety and Social Justice Model" is a retreat from the User Sovereignty Model of prior technology (word processors, browsers, search — all of which served user intent). By embedding guardrails against outputs the company deems harmful or immoral, AI companies become co-authors and thus de-facto editors of political speech. He cites a Future of Free Speech study finding that Gemini and GPT-3.5 wrote pro-transgender-athlete but not anti, pro-abortion-access but not anti, and in favor of "systemic racism" measures but not against. He gives five reasons for the retreat: authorship responsibility; feasibility (AI understands its own outputs); guardrail creep (legal liability leading to ideological guardrails); the Terminator Problem (existential-risk framing justifying content controls); and the 2010s–20s political zeitgeist (technologists as equity agents, versus the 1970s–90s user-serving norm). His prescription is a return to user sovereignty via competition plus select structural mandates, and he warns that concentrated AI combined with government pressure is an even more tempting censorship vector than the social-media era.

"Rediscovering the Pleasures of Pluralism" — Lily L. Tsai and Alex Pentland

Tsai and Pentland argue that cross-cutting civic infrastructure, not just local community, is needed to counter "partisan mega-identities" (Liliana Mason). They focus on digital civic infrastructure that enables both local and national direct democracy, and borrow Bernardo Zacka's concept of "in-between spaces" (balconies, stoops) that enable "reserved sociability" — civic participation at a modulated distance. They argue online platforms collapsed this dimension, leaving people either "in" or "out." The design implication is platforms where people can observe, lurk, calibrate engagement, and scale up deliberation on shared rather than personal problems, operated by a public actor or tightly regulated to avoid engagement-gaming.

"The Potential for AI to Restore Local Community Connectedness" — Sarah Friar and Laura Bisesto

Writing as Nextdoor's CEO and general counsel, Friar and Bisesto argue that local connections are the bedrock of a healthy democracy and that AI-enabled platforms can either fragment or strengthen them. They cite the U.S. Surgeon General's loneliness advisory and Nextdoor/Holt-Lunstad research finding that knowing as few as 6 neighbors cuts loneliness by half and that small acts of kindness compound, along with a Spain loneliness-cost study estimating €14 billion per year (1.17% of GDP) from unwanted loneliness. They frame the prior social-media generation as "slow" and user-driven, whereas the AI-enabled generation is dynamic, personalized, and multimedia, amplifying echo chambers if unmanaged. Their strategy is to build trusted local interfaces (verification, neighborhood-scoped content, weak-tie facilitation) that route AI toward civic engagement rather than engagement-maximization.

"AI Meets the Cascade of Rigidity" — Jennifer Pahlka

Pahlka argues that the root obstacle to effective government AI adoption is not technology risk but diminished state capacity — civic disengagement driven by governments failing to meet public-service needs (citing Joe Soss's research that means-tested benefit programs reduce voting), and Adam Tooze's "polycrisis" framing. She introduces the "cascade of rigidity": how well-intentioned top-level rules (FISMA's 300 security controls, Pendleton-Act merit hiring) become vetocracies as they cascade through risk-averse bureaucracies, summarized as "culture eats policy." She offers the Jack Cable anecdote in which a DDS security-contest winner was rejected by an HR screener for a "gobbledygook" resume. She reframes the question from "how much should we constrain AI in government?" to "how much capacity and competency does government have to deploy AI well?", arguing for building enabling capacity rather than layering mandates.

"Techno-ideologies of the Twenty-first Century" — Mona Hamdy, Johnnie Moore, E. Glen Weyl

The authors critique the two dominant Silicon Valley techno-ideologies: Technocracy (Altman/OpenAI's "Moore's Law for Everything" plus UBI plus centralized AGI control; historical siblings Asimov, Kurzweil, Banks, and Bostrom's Deep Utopia; also called "Fully Automated Luxury Communism") and Libertarianism (crypto, sovereign individuals, the Network State, Thiel/Andreessen funding; Srinivasan, Yarvin). They argue both ideologies are developed by and for "secular wealthy tech-focused white men," a small fraction of the world's population. Their alternative is a plurality framework that centers structurally underrepresented perspectives, particularly religion (they note that by 2050 only 13% of the world's population will be religiously unaffiliated, as the most religious populations are fastest-growing) and ecology. The essay uses mirror-cast authorship (one Muslim, one Christian, one Jewish; one left, one right, one center; three different disciplinary backgrounds) and centers the question of what people actually want.

"Getting AI Right: A 2050 Thought Experiment" — James Manyika

Manyika concludes Volume 1 with a backcast from 2050 — "AI has turned out to be hugely beneficial to society. What happened?" (a thought experiment originally posed by Stuart Russell). The question provokes a Working List of Hard Problems in AI that must be solved: ten problems grouped into five thematic clusters, spanning technical, sociotechnical, economic, social, ethical, and governance domains. He cites the March 2024 UN General Assembly resolution on "Seizing the opportunities of safe, secure and trustworthy artificial intelligence" as a foundation for rights-based, inclusive AI governance anchored in the UN Charter, International Human Rights Law, and the SDGs, and broadens governance beyond regulation to design, development, use, and benefit-distribution in parallel.

"Informational GPS" — Reid Hoffman and Greg Beato

Hoffman and Beato argue that LLMs are to information what GPS is to location — a decentralized, agency-enhancing public utility. They note GPS's post-2000 civilian upgrade produced $1.4 trillion in economic benefits (NIST 2019), while flagging the metaphor's limits: GPS has objective ground truth, whereas LLMs navigate subjective "informational planets" shaped by training data, parameters, and RLHF choices. Their prescription is a multipolar LLM ecosystem with no single "source of truth" and a priority on individual agency, citing the Brockman/Sutskever 2015 OpenAI launch vision that "AI should be an extension of individual human wills." Contra the Packard/Orwell panopticon fear, they argue networked PCs produced bottom-up individualism rather than top-down conformity, and that LLMs extend this trajectory if designed well.

Volume 2 essays — AI and the economy

Volume 2 began publishing December 11, 2025 and pivots from democratic institutions to the economy. Its essays return repeatedly to a labor-to-capital shift (addressed through tax and dividend systems), market power (competition policy and global spillovers), and transition design (insurance, sandboxes, education). Stiglitz and Ventura-Bolet's essay, published December 11, 2025, sits at the intersection of the democracy and economy threads through its focus on the information ecosystem and is treated alongside the Volume 1 democracy material below.

"Information in the Age of AI: Challenges and Solutions" — Joseph Stiglitz, Màxim Ventura-Bolet (2025-12-11)

Stiglitz and Ventura-Bolet argue that information is a public good that markets already under-produce while over-producing mis- and disinformation, and that AI sharpens three failures: (1) undersupply of good information (AI intermediaries undercut traditional producers such as news and research by answering without routing traffic or compensation, gutting their business models); (2) oversupply of mis- and disinformation ("engagement through enragement," with lies cheaper to produce than verified truth); and (3) undersupply of correction (fact-checking is itself a public good). They describe four AI shocks: (1) improved production and dissemination; (2) erosion of the producer business model; (3) the falling relative cost of lies; and (4) a "drone-war effect" in which AI simultaneously detects and evades falsehood detection, producing a verification/deception arms race. Against Cochrane-style deregulation, they argue not all innovation is welfare-enhancing (market-power-exploiting innovation is welfare-decreasing) and that with appropriate regulation AI information-ecosystem effects can be net positive, while without it they expect collapse.

"Beyond Job Displacement: How AI Could Reshape the Value of Human Expertise" — David Autor, Neil Thompson

Autor and Thompson apply the expertise framework from their 2025 "Expertise" paper. Automation that removes an occupation's inexpert tasks makes remaining work more complex, so wages rise and employment contracts; automation that removes expert tasks makes work more accessible, so employment rises and wages fall. They illustrate with accounting clerks (1980–2018), whose wages rose 39% while employment fell 32% (expert tasks preserved), and inventory clerks, whose wages fell 13% while employment rose 175% (expert tasks automated). A four-decade analysis of 303 occupations finds that a +1σ expertise requirement correlates with 16–31% higher wages, and that changes in requirements predict wage changes (+1σ corresponds to +18%). They lay out two scenarios: a transition-management problem, in which society fails to imagine the future of expertise; and labor obsolescence — Herbert Simon's "intolerable abundance" — with three surpassing challenges of social organization, income distribution, and democratic stability, noting affluent countries can hedge while many cannot.

"Economic Possibilities for Artificial Intelligence" — Gabriel Unger

Unger (in a title riffing on Keynes's 1930 essay) argues that transformative AI's payoff depends on answering three questions: a compelling shared vision (Silicon Valley's utopia reads as dystopia to most — "every sci-fi movie is now a horror movie" — contrasted with the cultural optimism of the 1851 World's Fair, Apollo, the "Carousel of Progress," and the Jetsons that persisted even during the nuclear arms race); a theory of economic growth and AI that takes its specificity seriously rather than treating it as generic capital; and a redesign of education and strengthening of social connections, which sits outside pure economics. He frames the essay as hope rather than optimism-versus-pessimism ("both are forms of fatalism") and calls the economics profession to engage with questions Silicon Valley asks but cannot answer alone.

"Resilient by Design: Dual Safety Nets for Workers in the AI Economy" — Ioana Marinescu

Marinescu proposes a two-tier architecture designed to be scenario-robust across a short transition versus permanent labor scarcity. The first tier, AI Adjustment Insurance (AI-AI), combines extended UI duration, wage insurance (70% of the pay gap for two years), and training, modeled on Trade Adjustment Assistance (TAA), which evidence shows boosted earnings and accelerated reemployment without locking workers into low wages; it is earmarked for AI-displaced workers as compensation for the policy choice to encourage AI deployment (America's AI Action Plan, America's AI Action Plan). The second tier, the Digital Dividend (DD), is a small universal cash benefit financed by a tax on the digital sector and is scalable toward UBI if persistent joblessness materializes. The combination hedges uncertainty: AI-AI dominates in a transition scenario, while DD dominates in an obsolescence scenario.

"Universal Basic Capital: An Idea Whose Time Has Come" — Nicolas Berggruen, Nathan Gardels

Berggruen and Gardels argue that if transformative AI shifts value to capital, sharing prosperity requires ownership stakes in AI rather than wage augmentation. Universal Basic Capital (UBC) gives every citizen a small equity stake in AI firms or AI-generated productivity. They note Piketty's r > g implies wealth concentration absent redistribution, that by 2024 the richest 10% owned 93% of US equity, and that from 1979 to the present productivity rose 86% while hourly pay rose 27%. They distinguish UBC from UBI: UBI maintains consumer-class status, while UBC makes every citizen an owner. They cite precedents in the Alaska Permanent Fund, Singapore's CPF, and sovereign-wealth models, and argue the "time has come" because AI intensifies r > g in ways that can be directly traced to capital.

"The Missing Institution: A Global Dividend System for the Age of Transformative AI" — Anna Yelizarova

Yelizarova argues that UBI debates are national while AI's labor displacement is global: "European software developers, Indian call centers, Mexican factory operators could all be displaced by AI built in Silicon Valley without their countries capturing the gains," and absent a global redistribution institution demand collapses and supply chains fracture. She proposes a global dividend system on the scale of a new Bretton Woods institution, interrogating rather than advocating it: what scale of resources, what governance structure, and what implementation path. She cites precedents including sovereign wealth fund models and climate-finance mechanisms (the Loss & Damage Fund), while noting none are at the required scale.

"Preserving Fiscal Stability in the Age of Transformative AI" — Anton Korinek, Lee Lockwood

Korinek and Lockwood argue that when $1 of value shifts from labor to capital, tax revenue falls roughly 10–15¢ (the US effective labor tax is about 30%, the effective capital tax about 15–20%), so as AI substitutes for labor this creates a structural deficit that compounds, with labor income taxes exceeding 50% of US federal revenue. They lay out a two-phase path: in Phase 1 (the twilight of labor), labor income taxes decline as a fraction of GDP and the system must gradually rebalance toward capital and consumption; in Phase 2 (a labor-light economy), taxation is redesigned around capital income, corporate profits, and possibly AI-generated value directly. They draw the historical analogy of the Industrial Revolution's century-long transition from land-based to income and consumption taxation, arguing AI-era fiscal adaptation must happen faster.

"Transformative AI and the Increase in Returns to Experimentation: Policy Implications" — Ajay Agrawal, Joshua Gans

Agrawal and Gans frame transformative AI as a "genius supply shock" ("country of geniuses in a datacenter," after Amodei). They argue demand adjusts slowly because firms need complementary organizational capital (data infrastructure, evaluation frameworks, liability regimes) before deploying genius-level agents, so the bottleneck shifts from solving to deciding what to solve and testing. Their policy levers to accelerate safe adaptation are regulatory sandboxes (time-limited supervised testbeds) and regulatory holidays (time-boxed exemptions allowing experimentation before rules ossify). As evidence they cite Humanity's Last Exam, on which Grok 4 scored about 44.4% (July 2025); IMO gold-medal performance from GPT and Gemini models; and Meta Superintelligence Labs offering nine-figure researcher packages.

"Transformative AI in Financial Systems" — Alex Pentland, Alex Lipton

Pentland and Lipton argue that AI combined with tokenization of real-world assets will reshape finance, with both promises (frictionless trade, real-time modeling, inclusive ownership) and dangers (market manipulation, instant-coalition crashes, and the cornering of essential-resource markets such as medicines, food, housing, data, and compute). They note SWIFT's AI-enabled "smart contracts" on distributed ledgers are now a core part of its $4 trillion-per-day infrastructure, and that DeFi examples show AI-driven coalition-building already produces explosive instability. They argue transformative-AI markets will need audit and regulation to avoid a 2008-style crisis at much higher speed.

"Titans, Swarms, or Human Renaissance?" — Ramin Toloui

Toloui offers a historical typology of technological revolutions along two axes: winner-take-all versus fast-follow, and labor-displacing versus labor-augmenting. He uses the Manchester-versus-Bengal textile example, in which the same technology produced opposite outcomes driven by who controlled deployment and how government responded. He maps four AI-era scenarios: Titan's Dominion (winner-take-all and labor-displacing), Copilot Empire (winner-take-all and labor-augmenting), Disruption Swarm (fast-follow and labor-displacing), and Promethean Fire (fast-follow and labor-augmenting). He identifies five priorities for steering outcomes: tax, innovation, competition, AI governance, and infrastructure.

"The Universal Innervation of the Economy" — Steve Jurvetson

Jurvetson argues that "iterative algorithms" (AI, ML, directed evolution, generative design) are the biggest engineering advance since the scientific method, and that deep learning is domain-independent, making brain-building expertise fungible across industries. He describes a labor-market consequence of hyperbolic competition for "brain builders" ($6–10 million per engineer in acqui-hires; $500k first-job offers from certain Stanford labs; billion-dollar LLM acqui-hires). He frames specialized silicon (GPUs to FPGAs to AI ASICs) as the enabling substrate and argues AI will innervate every industry — not just tech but product positioning, supply chain, and materials science. Writing from a VC perspective, he argues this is already the dominant pattern at Google, Amazon, Baidu, and Microsoft.

"Cheap Goods for Everyone?" — Susan Athey, Fiona Scott Morton

Athey and Scott Morton argue that whether transformative-AI productivity gains reach consumers depends on competition: without it, firms retain cost savings as profit and stifle downstream innovation, and workers displaced by AI are protected by the welfare gain only if lower prices materialize. Their policy recommendations are to make competition a core consideration in AI regulation; to pay attention to each layer of the AI value chain (chips, compute, data, models, apps, distribution, devices), since bottlenecks anywhere enable dominance; to watch AI-specific concerns such as data flywheels, network effects, exclusive supply deals, and vertical integration; and to weigh trade policy, since when AI is an imported factor of production, market power can mean the difference between increases and decreases in national income. They warn that AI safety rules can reduce competition as a side effect.

"Advanced AI as a Global Public Good and a Global Risk" — Yoshua Bengio

Bengio argues that AI is a global public good produced under competitive conditions that systematically underprovide safety. He identifies three catastrophic risk categories: destructive chaos from weak actors (bioweapons, cyberattacks); concentration of power among strong actors; and loss of control to rogue AIs. He cites task-duration doubling every roughly 7 months (METR), which he argues implies human-level planning around 2030, aligning with the International AI Safety Report and AI 2027 scenarios. He applies the precautionary principle (severe risks warrant dedicated safety resources commensurate with existential stakes) and a tragedy-of-the-commons framing (corporations internalize benefits while risks are borne collectively), arguing for cooperative global governance as the only structural fix, consistent with his LawZero initiative.

"Beyond Rivalry: A US–China Policy Framework" — George Graylin

Graylin argues that a zero-sum US-China AGI race is neither inevitable nor necessary, and calls for a Marshall-Plan-scale US initiative to establish cooperative AGI governance with China. His seven-step framework is to define AI categories (GOFAI, ANI, Agentic, AGI, ASI) with policy-relevant distinctions; reject the "AGI finish line" framing; decompose risk (misuse dominates misalignment today); establish cooperation prerequisites; build US domestic readiness (UBII, reskilling, open access); develop globally representative data; and pursue a dual-track approach of cooperating on safety and the public good while competing responsibly elsewhere. He critiques America's AI Action Plan for lacking a holistic perspective on deployment and geopolitics.

"Open Global Investment as a Governance Model for TAI" — Nick Bostrom

Bostrom argues that Open Global Investment (OGI) — private ventures open to international shareholding within a government-defined regulatory framework — outperforms alternatives (the Manhattan Project model, CERN, Intelsat, nonprofit stewardship) on short timelines. He evaluates it against criteria of democratic oversight, national security, international cooperation, legitimacy, economic efficiency, equitable benefit-sharing, AI safety, responsible deployment, and operational security. He describes variants: OGI-1 (a single lead firm) versus OGI-N (multiple firms), and US-OGI (US-domiciled) versus geographically neutral. His key claim is that the status quo already approximates OGI, so moving toward its ideal is the feasible path, whereas proposals for nationalization (a US "Manhattan Project") push away from it. He adds a robustness constraint: any governance structure must survive intelligence-explosion-scale strain rather than merely being optimal under normal operation.

"Strategic Dynamics in the Race to AGI: A Time to Race Versus a Time to Restrain" — Abraham, Kavner, Moon, Matheny (CSET)

The CSET authors use game theory to model the US-China AGI race and argue it is not universally a Prisoner's Dilemma; the correct model depends on actor diversity (private firms now dominate R&D), capability asymmetries, and uncertainty about what constitutes "winning." Their conclusion is that cooperation is not inevitable, but neither is competition. They review prior game-theoretic analyses (Dung and Hellrigel-Holderbaum, who prefer slower, risk-averse development; Katze and Futerman, who find international coordination strategically better; Kreps, who argues accelerated development resolves uncertainty). The policy implication is that policymakers must identify which game is actually being played before choosing a strategy, since different games yield different equilibria. They note Europe's ambiguous role (the AI Act versus exclusion from the frontier) complicates simple two-player framings, and identify Saudi Arabia, the UAE, and emerging markets as potential third-party actors.

"Private Physical AI for the Edge" — Daniela Rus

Rus offers a counter-narrative to the "ever-larger models" trajectory. Private Physical Edge AI is compact, energy-efficient, on-device intelligence computed locally where data is generated and actions occur. Her examples include liquid neural networks (Liquid AI's commercialization of MIT research; compact, causal, and able to generalize zero-shot across environments — trained on a drone hiking in summer woods and transferred to winter and urban contexts) and state-space models (LinOSS, Mamba) that compete with transformers on benchmarks at a fraction of the compute. She proposes tokens per watt, not parameters, as the measure of intelligence, and describes an ARM-analogous shift in which value moves from centralized compute providers to builders of applications and human-AI collaborations. She gives applications including AI glasses for blind and visually impaired users without cloud dependency, edge-powered medical triage in rural clinics, and farmers using solar-powered local AI. Against the concentration story, she argues edge AI could "redistribute the benefits of intelligence more widely than any prior innovation in history," while also producing a new AI literacy divide centered on knowing how to use AI effectively rather than on access.

"The Democratization of Intelligence" — Sarah Friar

Friar (Nextdoor CEO, writing in 2025) argues that AI adoption is democratizing faster than prior waves, drawing on an OpenAI usage study. At the November 2022 launch, 80% of users had masculine names and use was concentrated in rich countries (a coding use-case skew). By May 2025, low- and middle-income-country growth outpaced rich countries 4-to-1; at the 25th percentile of GDP per capita, internet-adjusted adoption matched high-income economies; and feminine-name users slightly outnumbered masculine. The usage mix by mid-2025 was 60% non-work, about 80% practical guidance, information, or writing; education accounted for 10% of all messages (40% of practical guidance), and writing tasks leaned toward editing 2:1 over new creation. Her field applications include Kenya's Penda Health AI Consult (a 16% reduction in diagnostic errors), UNICEF Uruguay accessible digital textbooks (produced in days rather than months), and Nigeria's ADVISER vaccination system (raising child vaccination uptake from 43.6% to 73.9%). She argues the new intelligence divide is in effective use rather than access, requiring investment in AI literacy, institutions, and trust infrastructure.

"The San Francisco Consensus" — Eric Schmidt

Schmidt argues that beneath Silicon Valley's visible divides (doomers versus accelerationists, open versus closed, regulation versus deregulation) lies a three-part consensus among leading AI builders: that scaling laws will drive continued rapid progress; that timelines are short, with superintelligence within 2–5 years; and that benefits will be transformative, depicted as hockey-stick graphs of scientific, financial, and human progress. He draws parallels to the postwar Keynesian and Washington consensuses, noting that "the fact that such a consensus is widely held does not, of course, make it true," and acknowledges dissent from Yann LeCun on LLM limits, Fei-Fei Li's alternative approaches, Peter Thiel-style progress skepticism, and distinct European and Chinese perspectives. He describes a three-axis architecture: the language revolution (done), the agentic revolution (underway, functional rather than creative), and the reasoning revolution (most consequential and most speculative, with AGI emerging "as a property of scale"). He identifies material constraints in energy (about 92 nuclear plants' worth), data (the internet absorbed, requiring synthetic and embodied sources), and algorithms. The essay complements his Volume 1 "Democracy 2.0" essay; Schmidt is the only author to contribute to both volumes.

"What's There to Fear in a World with Transformative AI? With the Right Policy, Nothing." — Betsey Stevenson

Stevenson argues there is no iron law that societal well-being rises with AI-driven prosperity, identifying three problems to solve simultaneously: how ordinary people can seek a better life if human work is devalued; how resources will be distributed; and where people will find meaning and purpose. Her empirical anchor is that the income-to-life-satisfaction relationship is roughly log-linear, so a more equal distribution at any given average income yields higher aggregate well-being, making the distributional question not secondary to the growth question. She sets three policy priorities: reshape the jobs conversation from "will we lose jobs" to "how do we manage the speed and process of transformation"; recognize humans as data producers on whom AI depends, designing institutions that ensure data-generators share in the value created; and shore up reciprocity and trust so AI prosperity is experienced as a shared project rather than something done to people. Her argument resonates with Marinescu's safety-net architecture, Korinek and Lockwood's fiscal rebalancing, and Yelizarova's global dividend, but is centered on meaning and the social contract rather than income alone.

"'Career' Advice from the AI Frontier" — Avital Balwit (Anthropic)

Balwit, an Anthropic staffer, offers a biographical essay arguing that the economically relevant benchmark for AI is not "better than the best human" but "better than the human who would otherwise do the task," and that young people need new mindsets, not just skills. On timelines, she views transformative AI as "likely to arrive in the next 2–4 years," in the sense that it "exists in the lab and works well enough for customer use," while diffusion is slower. Her advice themes are to focus on judgment, taste, and agency over execution speed; to cultivate inter-AI collaboration skills (orchestration, verification, prompt-craft); to build resilient identity structures (meaning, purpose, community) that do not depend on labor-market status; and to pursue education reform, since traditional reskilling is insufficient and mindset-level preparation is needed for a world where AI surpasses humans cognitively.

Cross-essay tensions and consensus

The essays disagree most sharply along several axes:

The essays largely converge on several points: that AI combined with engagement-optimized social media is the problem rather than AI alone; that concentrated AI corporate power is a serious risk; that the 2016–2024 "deepfake flood decides elections" panic was wrong in its particulars; that digital civic infrastructure and deliberative mechanisms are worth experimenting with; and that public compute plus an outside-auditor ecosystem is needed for accountability.

Relationships

• supports: (Source: anthropic.com) — Tsai/Pentland on "reserved sociability" aligns with the wellbeing-over-engagement framing.
• contradicts: the essays contradict each other across the regulatory-stance axis; see the tensions table.
• related: Claude's Constitution — referenced by Siddarth/Huang/Tang as the mechanism enabling Collective Constitutional AI.
• related: ChatGPT, Can You Solve the Content Moderation Dilemma? — shared territory on platform-level moderation and user sovereignty.
• related: AI as Normal Technology — Cochrane's "relax" and Narayanan/Kapoor's "normal technology" are closely aligned regulatory moods.
• related: FLI — Pause Giant AI Experiments: An Open Letter / Statement on AI Risk (CAIS) — Cochrane explicitly targets the existential-risk framing.
• related: Techno-Federalism: How Regulatory Fragmentation Shapes the U.S.-China AI Race / America's AI Action Plan / EU AI Act (Regulation 2024/1689) — object-level regulatory debates the Digitalist Papers address at the meta level.
• related: Synthetic Content: Exploring the Risks, Technical Approaches, and Regulatory Responses — Persily's position on watermarking maps onto the FPF synthetic-content findings.
• related: Stanford GSB — Measuring Perceived Slant in LLMs (Westwood, Grimmer, Hall) / Manhattan Institute — Measuring Political Preferences in AI Systems (Rozado) — empirical complements to Volokh's political-content analysis.
• depends-on: Collective Intelligence Project (CIP); Deliberative Alignment.