=== CLAIMS (328) === C1: If AI displaces human workers faster than the economy can reabsorb them, it risks eroding the very consumer demand firms [logical] C2: Knowing that AI displacement erodes consumer demand is not sufficient for firms to stop it. [logical] depends: C1 C3: In a competitive task-based model, demand externalities trap rational firms in an automation arms race, displacing worke [logical] [FORMAL] C4: The resulting loss from over-automation harms both workers and firm owners. [logical] depends: C3 C5: More competition and better AI amplify the excess automation beyond what is collectively optimal. [logical] [FORMAL] depends: C3 C6: Wage adjustments and free entry cannot eliminate the over-automation externality. [logical] [FORMAL] depends: C3 C7: Capital income taxes cannot eliminate the over-automation externality. [logical] [FORMAL] depends: C3 C8: Worker equity participation cannot eliminate the over-automation externality. [logical] [FORMAL] depends: C3 C9: Universal basic income cannot eliminate the over-automation externality. [logical] [FORMAL] depends: C3 C10: Upskilling cannot eliminate the over-automation externality. [logical] [FORMAL] depends: C3 C11: Coasian bargaining cannot eliminate the over-automation externality. [logical] [FORMAL] depends: C3 C12: Only a Pigouvian automation tax can eliminate the over-automation externality. [logical] [FORMAL] depends: C3 C13: Policy should address not only the aftermath of AI labor displacement but also the competitive incentives that drive it. [logical] depends: C3, C12 C14: The fear that technology will displace workers is at least as old as the Industrial Revolution. [empirical] C15: Historically, labor displacement has largely been self-correcting: automation of existing tasks has been offset by the c [empirical] C16: What Acemoglu and Restrepo call the reinstatement effect has tended to stabilize the labor market. [empirical] depends: C15 C17: Whether historical labor market self-correction will hold in the age of AI is an open question. [empirical] depends: C15 C18: Displacement has intensified over the past four decades while the creation of new work has not always kept pace. [empirical] C19: The current wave of AI-driven displacement is disproportionately affecting entry-level workers. [empirical] C20: Displaced workers are also consumers, and when their lost income is not replaced, each round of layoffs erodes the purch [logical] C21: At the limit, firms would automate their way to boundless productivity and zero demand. [logical] depends: C20 C22: Public discourse increasingly treats the dynamic of AI-driven displacement eroding demand as an inevitable process with [empirical] C23: Rational, forward-looking firms should theoretically serve as the brake on over-automation if they can see the consequen [logical] C24: Block cut nearly half its 10,000-person workforce in February 2026, with CEO Jack Dorsey stating that AI had made many o [empirical] C25: Over 100,000 tech workers were laid off in 2025 alone, with AI cited as a primary driver in more than half the cases. [empirical] C26: Roughly 80% of U.S. workers hold jobs with tasks susceptible to automation by large language models. [empirical] C27: Salesforce replaced 4,000 customer-support agents with agentic AI. [empirical] C28: Cognition's Devin, deployed at Goldman Sachs and Infosys, enables one senior engineer to do the work of a five-person te [empirical] C29: An automating firm captures the full cost saving but, under competitive pricing, bears only a fraction of the resulting [logical] [FORMAL] C30: Each firm's profit-maximizing automation rate is a strictly dominant strategy that exceeds the cooperatively efficient l [logical] [FORMAL] depends: C29 C31: Foresight alone cannot prevent the race toward over-automation. [logical] depends: C30 C32: The distortion from over-automation deepens with competition: a monopolist fully internalizes the externality, while fra [logical] [FORMAL] depends: C29 C33: In the frictionless limit, where every task is equally easy to automate, the game sharpens into a Prisoner's Dilemma in [logical] [FORMAL] depends: C29 C34: The surplus loss from over-automation is not a transfer from workers to firm owners; it is a deadweight loss that harms [logical] [FORMAL] depends: C3 C35: Higher AI productivity widens the over-automation wedge rather than resolving it. [logical] [FORMAL] depends: C3 C36: Each firm perceives a market-share gain from automating beyond rivals, but at the symmetric equilibrium these gains canc [logical] [FORMAL] depends: C35 C37: Endogenous wage adjustment raises the threshold at which the externality activates but cannot close the wedge once it do [logical] [FORMAL] depends: C3 C38: Wage flexibility changes when the problem bites, not whether it exists. [logical] depends: C37 C39: Free entry, capital-income recycling, and richer product-market structures fail to eliminate the distortion. [logical] [FORMAL] depends: C3 C40: The paper builds on the task-based approach to automation (Zeira, 1998; Autor et al., 2003; Acemoglu and Restrepo, 2018, [structural] C41: Existing task-based automation literature focuses on whether and how the labor market rebalances; this paper asks what h [structural] depends: C40 C42: Beraja and Zorzi (2025) show that automation is inefficient when displaced workers face borrowing constraints during rea [empirical] C43: The mechanism in this paper operates through the product market: firms ignore the demand they destroy for rival firms, u [structural] depends: C42 C44: Beraja and Zorzi's inefficiency arises even for a single firm in isolation; the inefficiency in this paper requires comp [structural] [FORMAL] depends: C42, C43 C45: A planner in this model would reduce automation even with zero weight on workers, because over-automation harms firm pro [logical] [FORMAL] depends: C34 C46: The technology ecosystem may be biased toward 'so-so' automation that displaces workers without large productivity gains [empirical] C47: Automation may disproportionately target high-rent tasks, dissipating worker surplus rather than raising output. [empirical] C48: The externality in this paper arises only under competition and persists even when automation is highly productive, cred [logical] [FORMAL] depends: C3, C44 C49: The demand externality studied belongs to the family of aggregate demand spillovers introduced by Rosenstein-Rodan (1943 [structural] C50: In 'big push' models, demand complementarities across sectors can prevent individually unprofitable investments from bei [definitional] C51: The mechanism in this paper is the mirror image of big push models: individually profitable automation is collectively d [structural] depends: C50 C52: The game in this paper yields a unique dominant-strategy equilibrium, making the problem a true externality rather than [logical] [FORMAL] depends: C30 C53: AI systems deliver substantial productivity gains. [empirical] C54: AI pricing algorithms can spontaneously learn to collude. [empirical] C55: Firms under labor-issue scrutiny invest specifically in AI automation rather than other forms of IT. [empirical] C56: As AI reliability improves, incentivizing effective human oversight becomes prohibitively expensive, weakening a key che [empirical] C57: Each AI adoption decision is rational in isolation, but collectively they erode the consumer demand all firms depend on. [logical] depends: C29 C58: The model considers a sector with N ≥ 2 symmetric firms. [structural] [FORMAL] C59: Each firm is endowed with L > 0 task-positions, initially all performed by human workers. [structural] [FORMAL] C60: Firm i chooses an automation rate αi ∈ [0,1], where tasks z ∈ [0, αi] are performed by AI at cost c per task and remaini [structural] [FORMAL] depends: C59 C61: Each automated task displaces one worker, so αi is simultaneously the automation rate and the fraction of the workforce [definitional] [FORMAL] depends: C60 C62: Wages are exogenous in the baseline model. [structural] C63: In the perfect-substitutes limit, each task produces one unit of output regardless of mode, so firm output is Yi = L. [structural] [FORMAL] C64: The baseline normalization shuts down productivity and quality margins so that it captures only the spending consequence [structural] depends: C63 C65: Tasks are ordered by comparative advantage, making the marginal task progressively harder to integrate, captured via a c [structural] [FORMAL] C66: The per-task cost saving from automation is defined as s := w − c. [definitional] [FORMAL] depends: C60 C67: Workers have a higher marginal propensity to consume (MPC) than owners. [empirical] C68: Workers spend a fraction λ ∈ (0,1] of their income on the sector's good. [structural] [FORMAL] depends: C67 C69: Owners spend none of their income in the sector in the baseline. [structural] [FORMAL] C70: When automation displaces workers, income shifts toward agents with a lower sectoral MPC, reducing aggregate expenditure [logical] [FORMAL] depends: C67, C68, C69 C71: A fraction η ∈ [0,1] of displaced wage income is replaced via reemployment, transfers, or other sources; the remainder i [structural] [FORMAL] C72: The effective demand loss per automated task is defined as ℓ = λ(1−η)w. [definitional] [FORMAL] depends: C68, C71 C73: Demand falls linearly in the average automation rate: D = A + λwLN − ℓLNᾱ. [structural] [FORMAL] depends: C72 C74: Firms sell output on the product market at a uniform price that equates aggregate supply and demand. [structural] [FORMAL] C75: Each firm earns revenue Revi = D/N, since all firms produce the same output. [structural] [FORMAL] depends: C63, C74 C76: Firms play a one-shot simultaneous-move game, each choosing αi to maximize πi; the solution concept is Nash equilibrium. [structural] [FORMAL] C77: Over-automation is measured against two benchmarks: the cooperative optimum (maximizing aggregate owner surplus K) and a [definitional] [FORMAL] C78: The environment assumes full transparency: every firm can directly observe how automation maps into lost worker income a [structural] C79: A marginal increase in automation saves s in labor costs but incurs friction kαi and reduces the firm's revenue by ℓ/N. [logical] [FORMAL] depends: C66, C72, C75 C80: The revenue loss from automation is ℓ/N rather than ℓ because competitive pricing allocates revenue equally across symme [logical] [FORMAL] depends: C75 C81: Each firm underestimates the social cost of its automation, resulting in systematic over-automation in equilibrium. [logical] [FORMAL] depends: C80 C82: If N ≤ N* = ℓ/s, no firm automates (αNE = 0). [logical] [FORMAL] depends: C72, C66 C83: If N > N*, each firm's strictly dominant strategy is αNE = min((s − ℓ/N)/k, 1). [logical] [FORMAL] depends: C82 C84: The cooperative optimum is αCO = min(max(0, (s − ℓ)/k), 1). [logical] [FORMAL] C85: When both αNE and αCO are interior, the over-automation wedge is ℓ(1 − 1/N)/k, which is strictly increasing in N and ℓ, [logical] [FORMAL] depends: C83, C84 C86: The equilibrium automation rate is a strictly dominant strategy: each firm over-automates even with perfect foresight ab [logical] [FORMAL] depends: C83 C87: The planner setting a common rate for all firms faces the full demand loss ℓ per automated task rather than the ℓ/N each [logical] [FORMAL] depends: C80, C84 C88: The over-automation wedge is strictly increasing in N: more competitive sectors exhibit wider automation gaps. [logical] [FORMAL] depends: C85 C89: More competition diluting each firm's share of the demand loss runs counter to the standard intuition that competition d [logical] depends: C88 C90: A monopolist (N=1) fully internalizes the externality so αNE = αCO; as N→∞, the wedge approaches its maximum of ℓ/k. [logical] [FORMAL] depends: C85 C91: As AI costs fall (c → 0), the automation threshold N* → λ(1−η) ≤ 1, so the over-automation region expands to cover virtu [logical] [FORMAL] depends: C82 C92: For illustrative parameters (c/w = 0.30, λ = 0.5, η = 0.30, N→∞), firms in competitive markets automate at twice the coo [logical] [FORMAL] depends: C85 C93: When frictions vanish (k→0), the moderating force of adjustment costs disappears and the game reduces to a Prisoner's Di [logical] [FORMAL] depends: C65 C94: When k=0 and N > N*, full automation (αi = 1) is strictly dominant for every firm. [logical] [FORMAL] depends: C93 C95: When k=0, N > N*, and s < ℓ, the cooperative optimum is no automation (αi = 0 for all i), yielding per-firm profit Π0; t [logical] [FORMAL] depends: C94 C96: In the frictionless Prisoner's Dilemma, the total deadweight loss is NL(ℓ − s). [logical] [FORMAL] depends: C95 C97: A firm that holds back unilaterally (αi = 0) still suffers the revenue decline from rivals' automation but forgoes the o [logical] [FORMAL] depends: C94 C98: Because automating is strictly dominant (not merely a best response to others' automating), no non-binding agreement can [logical] [FORMAL] depends: C94 C99: Communication is cheap talk in the sense of Crawford and Sobel (1982): even if all firms acknowledge that collective res [logical] depends: C98 C100: The automation externality is distinct from pure coordination failures where firms simply need to agree on which equilib [logical] depends: C52, C98 C101: The μ-planner's optimal automation rate is αSP(μ) = (s − ℓ)/k − μℓ/[λ(1−μ)k], restricted to [0,1]. [logical] [FORMAL] depends: C77 C102: The surplus loss from the Nash equilibrium relative to the planner's optimum is (1−μ)NLk/2 · (αNE − αSP(μ))². [logical] [FORMAL] depends: C101, C83 C103: αNE > αSP(μ) for every μ ∈ [0,1): the Nash equilibrium is Pareto dominated by the cooperative optimum, with workers and [logical] [FORMAL] depends: C101, C83 C104: Workers lose wage income directly through displacement from over-automation. [logical] depends: C103 C105: Firm owners, despite cutting costs on each automated task, also lose because collective displacement erodes demand to th [logical] [FORMAL] depends: C103 C106: No redistribution between workers and owners can make the Nash outcome efficient. [logical] depends: C103 C107: The total wedge between equilibrium and the planner's optimum decomposes into two sources: the uninternalized demand ext [logical] [FORMAL] depends: C101, C85 C108: The uninternalized demand externality grows with N, approaching ℓ/k as N→∞, so fragmented markets suffer disproportionat [logical] [FORMAL] depends: C107 C109: The distributional premium is independent of N but grows without bound as μ→1. [logical] [FORMAL] depends: C107 C110: The surplus loss is quadratic in the total wedge and scales with NL, so both fragmentation and market size amplify the w [logical] [FORMAL] depends: C102 C111: Only the Pigouvian automation tax fully corrects the over-automation distortion among the six instruments evaluated. [logical] depends: C12 C112: The policy analysis benchmarks against the cooperative optimum αCO, which maximizes aggregate profit without directly we [structural] depends: C77 C113: An instrument that does not operate on the externality margin cannot correct the distortion regardless of how it scores [logical] C114: The demand-loss parameter ℓ = λ(1−η)w governs the magnitude of the externality. [definitional] [FORMAL] depends: C72 C115: Higher η (income replacement rate) shrinks ℓ and thereby the over-automation wedge. [logical] [FORMAL] depends: C114, C85 C116: When η > 1, upskilling and reabsorption place displaced workers into higher-paying roles, automation increases aggregate [logical] [FORMAL] depends: C114 C117: When η > 1, the sign of the externality flips: the over-automation wedge becomes negative (under-automation). [logical] [FORMAL] depends: C116 C118: When η < 1, displacement destroys demand, and each firm bears only 1/N of the loss, producing over-automation. When η > [logical] [FORMAL] depends: C117 C119: In both over-automation and under-automation cases, the distortion grows with N, and a monopolist fully internalizes in [logical] [FORMAL] depends: C118 C120: The same corrective instrument addresses both over-automation (tax when η < 1) and under-automation (subsidy when η > 1) [logical] depends: C118, C12 C121: Historical technological transitions have often eventually reabsorbed displaced workers at higher wages. [empirical] C122: The AI buildout offers a concrete channel for η > 1 through expansion of data centers, energy infrastructure, and AI-adj [empirical] C123: Past displacement episodes have consistently produced η < 1: displaced workers suffer large, persistent earnings losses. [empirical] C124: There is little evidence yet that AI-driven displacement will differ from past episodes in producing η < 1, placing most [empirical] depends: C123 C125: Raising η through retraining programs, wage insurance, and incentives for new firm creation is not merely a palliative f [logical] depends: C115 C126: A UBI funded from general revenue maps to an increase in autonomous demand A in the model. [structural] [FORMAL] C127: UBI is unconditional, so employed and displaced workers receive the same payment, adding a constant to aggregate spendin [structural] [FORMAL] depends: C126 C128: UBI enters firm profit only through Π0 = A/N + (λ−1)wL, which drops out of the first-order condition. [logical] [FORMAL] depends: C126 C129: UBI changes neither the automation threshold N* nor the over-automation wedge. [logical] [FORMAL] depends: C128 C130: UBI changes payoff levels but not the payoff differences that drive strategic behavior. [logical] [FORMAL] depends: C129 C131: Instruments that operate on profit levels can redistribute income but cannot correct the externality; only instruments t [logical] depends: C130 C132: Despite not correcting the externality, UBI serves a complementary role by cushioning profit losses and raising the floo [logical] depends: C129 C133: When the number of firms is endogenous, UBI can paradoxically widen the externality by attracting new entrants and fragm [logical] [FORMAL] depends: C129, C88 C134: UBI is a complement to the automation tax, not a substitute. [logical] depends: C129, C12 C135: A proportional tax t ∈ (0,1) on capital income (profits) does not change the equilibrium automation rate, the threshold [logical] [FORMAL] C136: Capital income taxes are often conflated with robot taxes in the policy debate, but they are fundamentally different ins [logical] depends: C135 C137: The failure of capital income taxes to correct the externality is structurally identical to that of UBI: both shift prof [logical] depends: C135, C129 C138: Worker equity participation (giving workers a fraction ϵ of firm profits) recycles capital income back into demand throu [structural] [FORMAL] C139: The cooperative optimum is unchanged by worker equity participation: αCO(ϵ) = (s − ℓ)/k, independent of ϵ. [logical] [FORMAL] depends: C138 C140: The Nash equilibrium automation rate with worker equity is αNE(ϵ) = (s − ℓ/Nϵ)/k, where Nϵ = N − λϵ(N−1). [logical] [FORMAL] depends: C138 C141: The over-automation wedge with worker equity is strictly decreasing in ϵ but strictly positive for all ϵ < 1/λ. [logical] [FORMAL] depends: C139, C140 C142: The worker equity wedge vanishes only at ϵ = 1/λ, which requires ϵ > 1 whenever λ < 1, making it infeasible. [logical] [FORMAL] depends: C141 C143: Even at ϵ = 1 (full profit-sharing), the wedge remains strictly positive when λ < 1 because each unit of profit recycled [logical] [FORMAL] depends: C142 C144: The structural limitation of worker equity is that the externality is fundamentally multilateral: bilateral arrangements [logical] depends: C143 C145: Voluntary profit-sharing does not arise: if each firm independently chooses its profit-sharing rate ϵi to maximize retai [logical] [FORMAL] C146: The marginal cost of profit-sharing is πi (a dollar-for-dollar reduction in retained earnings), while the marginal deman [logical] [FORMAL] depends: C145 C147: The failure of voluntary profit-sharing is a second-order coordination failure layered on top of the automation external [logical] depends: C145, C3 C148: Profit-sharing must be mandated to have any effect, and even then it cannot substitute for a corrective tax. [logical] depends: C145, C141 C149: Neither bargaining between a firm and its own workers nor bargaining among firms can correct the automation externality. [logical] C150: If displaced workers bargain for per-task severance payment σ, the over-automation wedge becomes (1 − 1/N)(ℓ − λσ)/k, wh [logical] [FORMAL] depends: C149 C151: Worker-side bargaining is operationally equivalent to raising the income-replacement rate η by σ/w. [logical] [FORMAL] depends: C150 C152: The uninternalized portion of the externality does not fall on firm i's own workers at all; it reduces revenue at rival [logical] [FORMAL] C153: The externality is a firm-to-firm channel running through the product market, not a firm-to-worker channel that bilatera [logical] depends: C152 C154: A coalition of M firms that jointly maximizes its members' combined profit chooses the automation rate αM = (s − Mℓ/N)/k [logical] [FORMAL] C155: The residual over-automation wedge from a partial coalition is ℓ(1 − M/N)/k, which vanishes only when M = N (the grand c [logical] [FORMAL] depends: C154 C156: Four features of the automation externality prevent the grand coalition from forming: (1) dominant strategy makes agreem [logical] depends: C155 C157: In the large-numbers setting, Coase himself acknowledged that private bargaining breaks down. [empirical] depends: C156 C158: The fundamental obstacle to Coasian bargaining is incentive compatibility, not transaction costs: even with costless neg [logical] depends: C156 C159: A Pigouvian automation tax τ per task, with τ* = ℓ(1 − 1/N), implements αNE = αCO = (s − ℓ)/k. [logical] [FORMAL] depends: C83, C84 C160: The optimal Pigouvian tax rate has a transparent economic interpretation: each firm already bears ℓ/N of the demand loss [logical] [FORMAL] depends: C159 C161: For large N, the optimal Pigouvian rate τ* ≈ ℓ = λ(1−η)w, so setting the rate requires only sector-level observables. [logical] [FORMAL] depends: C159 C162: Levying the Pigouvian tax requires observing firm-level automation rates, a practical challenge that may be easing as AI [empirical] C163: A tax authority can compel disclosure through mandatory reporting, payroll records, and procurement audits, making appro [empirical] depends: C162 C164: Because the welfare loss is quadratic in the wedge, even an imprecisely targeted tax yields a first-order gain. [logical] [FORMAL] depends: C102 C165: With lump-sum rebate of tax revenue, each firm achieves exactly πCO (cooperative profit). [logical] [FORMAL] depends: C159 C166: Lump-sum rebate to firms returns revenue to automating firms while leaving displaced workers uncompensated. [logical] depends: C165 C167: Direct transfers (wage insurance, severance supplements) raise η mechanically but may weaken workers' incentive to retra [logical] C168: Funding retraining programs raises η through human-capital investment rather than income replacement, making gains in η [logical] C169: The Pigouvian automation tax is potentially self-reinforcing and self-limiting: it funds programs that raise η, which lo [logical] depends: C168, C159 C170: In practice, a mix of direct transfers and retraining is likely optimal: short-run transfers to cushion displacement whi [logical] depends: C167, C168 C171: When AI-performed tasks produce ϕ ≥ 1 units of output (vs. 1 for human tasks), higher AI productivity widens the over-au [logical] [FORMAL] C172: With ϕ > 1, a deviating firm raises its output above rivals and captures a larger share of expenditure, creating a marke [logical] [FORMAL] depends: C171 C173: The Nash equilibrium automation rate is increasing in AI productivity: αNE(ϕ) > αNE(1) for ϕ > 1. [logical] [FORMAL] depends: C172 C174: The cooperative optimum and the generalized planner's optimum are unchanged by AI productivity: αCO(ϕ) = αCO(1) for all [logical] [FORMAL] C175: Higher ϕ raises output but lowers the price in proportion, leaving the planner's objective invariant to ϕ, because total [logical] [FORMAL] depends: C174 C176: The over-automation wedge αNE(ϕ) − αSP(μ;ϕ) is strictly increasing in ϕ for every μ ∈ [0,1]. [logical] [FORMAL] depends: C173, C174 C177: The Red Queen effect means that each firm perceives a market-share gain from automating beyond rivals, but at the symmet [logical] [FORMAL] depends: C172 C178: Better AI raises the equilibrium automation rate without shifting the efficient benchmark, so the distortion grows with [logical] [FORMAL] depends: C176 C179: The baseline Pigouvian rate τ* = ℓ(1 − 1/N) no longer suffices when ϕ > 1; implementing αCO requires an additional corre [logical] [FORMAL] depends: C171, C159 C180: The lower price from higher ϕ means each dollar of spending buys more physical output, so nominal welfare measures under [logical] C181: Free entry reshapes the over-automation problem but does not resolve it. [logical] C182: With free entry and frictionless automation (k=0), when entry cost is low, every firm fully automates (Prisoner's Dilemm [logical] [FORMAL] depends: C94 C183: In the frictionless entry model, the threat of automation can function as an endogenous entry barrier, sustaining positi [logical] [FORMAL] C184: With convex costs (k > 0) and free entry, if NFE > N*, then αNE(NFE) > αCO: over-automation persists under free entry. [logical] [FORMAL] depends: C181 C185: The standard tendency toward excess entry (Mankiw and Whinston, 1986) widens the wedge by fragmenting the market further [logical] depends: C184, C88 C186: By raising autonomous demand A, UBI increases per-firm profit, attracting additional entrants and potentially widening t [logical] [FORMAL] depends: C133 C187: Endogenous wage adjustment raises the threshold at which over-automation activates but cannot close the wedge once it do [logical] [FORMAL] depends: C37 C188: Both the cost saving s(w) = w − c and the demand-loss parameter ℓ(w) = λ(1−η)w are increasing in the wage. [logical] [FORMAL] C189: The automation threshold N* = ℓ/s rises as wages fall, because the cost saving s = w − c contracts faster than the deman [logical] [FORMAL] depends: C188 C190: Competitive pricing allocating revenue as Revi = D/N at any wage level means each firm bears only a fraction of the dema [logical] [FORMAL] depends: C75 C191: Wage adjustment changes the magnitude of ℓ but not the fraction each firm internalizes; that fraction is a property of m [logical] depends: C190 C192: As w → c, the cost saving s → 0 while N* → ∞, so eventually no firm finds automation privately worthwhile, but this is a [logical] [FORMAL] depends: C189 C193: A labor market that 'self-corrects' only by impoverishing its workforce has transmuted displacement into depressed livin [logical] depends: C192 C194: Wage flexibility changes when the externality bites, not whether it exists. [logical] depends: C191 C195: Capital-income recycling (owners spending fraction η̂ of profits in the sector) narrows the over-automation wedge but ca [logical] [FORMAL] C196: The effective demand-loss parameter with capital-income recycling is ℓη̂ = ℓ − η̂s. [definitional] [FORMAL] C197: The externality from capital-income recycling vanishes (ℓη̂ ≤ 0) only when η̂ ≥ ℓ/s. [logical] [FORMAL] depends: C196 C198: When ℓ > s, the required capital recycling rate exceeds one, so recycling is impotent precisely where the externality is [logical] [FORMAL] depends: C197 C199: Capital-income recycling raises the fraction of demand loss each firm internalizes from 1/N to 1/N̂ but cannot push it t [logical] [FORMAL] depends: C195 C200: Richer product-market interaction (Cournot or Bertrand second-stage competition) would not eliminate the demand external [logical] C201: The market-share motive under imperfect competition is more subtle: at the symmetric equilibrium, market-share gains can [logical] depends: C177 C202: Under CES task aggregation with tasks as substitutes (elasticity > 1), the demand externality still operates at the auto [logical] C203: When tasks are complements (elasticity < 1), the restraint on automation is stronger, but the demand externality remains [logical] C204: Even as every firm recognizes that vanishing paychecks mean vanishing customers, not one of them will stop, because each [logical] depends: C29 C205: Anthropic CEO Dario Amodei has warned that AI-driven displacement will be 'unusually painful,' 'much broader' and 'much [empirical] C206: The model predicts the over-automation problem is most severe not in dominant technology firms but in fragmented industr [logical] depends: C88, C176 C207: The distinguishing empirical signature of the externality would be profit erosion coinciding with mass layoffs, which st [empirical] depends: C34 C208: The empirical signature of profit erosion requires displacement at a scale and speed beyond what has materialized so far [empirical] depends: C207 C209: Private returns to AI systematically overstate the returns to the economy as a whole because they do not account for the [logical] depends: C57 C210: Even a planner who places zero weight on worker welfare would reduce the automation rate below the equilibrium level. [logical] [FORMAL] depends: C103 C211: The problem is not that firms profit at workers' expense; it is that over-automation harms both groups, making correctio [logical] depends: C34, C210 C212: By Tinbergen's principle, a distinct market failure requires a distinct instrument. [logical] C213: No amount of retraining, income support, or bargaining will slow the automation arms race; only a tax on automation itse [logical] depends: C12, C212 C214: A unilateral automation tax could push adoption offshore, strengthening the case for multilateral coordination or border [logical] depends: C159 C215: The model is deliberately simple: one sector, one period, symmetric firms. Each of these choices is conservative, meanin [structural] C216: A single sector understates the externality because in a multi-sector economy, layoffs in one sector reduce spending on [logical] depends: C215 C217: AI investments are largely irreversible. [empirical] C218: Even the threat of automation can reshape market structure before any displacement occurs, strengthening the case for ea [logical] depends: C183 C219: The income-replacement rate η rises over time as displaced workers retrain and new occupations emerge, so the optimal ta [logical] depends: C169 C220: Endogenizing AI development could compound the problem: firms racing to automate may invest disproportionately in labor- [logical] C221: All extensions considered in the paper point toward a larger problem, not a smaller one. [logical] depends: C215, C216, C220 C222: Blanchflower and Oswald (1995) document a robust negative relationship between wages and unemployment across more than a [empirical] C223: The quadratic adjustment-cost specification used follows the standard form from Lucas (1967) and Hamermesh and Pfann (19 [structural] C224: The MPC asymmetry between workers and owners is supported by Kaldor (1956) and Mian et al. (2021). [empirical] depends: C67 C225: Displaced workers suffer large, persistent earnings losses, as documented by Jacobson et al. (1993). [empirical] C226: The over-automation result extends to any μ ∈ [0,1) under endogenous wages (Corollary 4). [logical] [FORMAL] depends: C187 C227: Firm i's profit function πi is strictly concave in αi. [logical] [FORMAL] C228: The optimal automation rate αi* does not depend on any rival's choice αj: rivals' automation levels enter only through a [logical] [FORMAL] depends: C227 C229: Total profit is maximized when the αi are as equal as possible, by convexity of x². [logical] [FORMAL] C230: The cooperative optimum is symmetric: αi = α for all i. [logical] [FORMAL] depends: C229 C231: In the frictionless limit with k=0, firm i's profit is affine in αi with slope L(s − ℓ/N) that does not depend on rivals [logical] [FORMAL] C232: At the Nash equilibrium with full automation (ᾱ = 1), per-firm demand drops by ℓLN compared to no automation. [logical] [FORMAL] depends: C96 C233: Interior regions for αNE, αCO, and αSP(μ) as functions of cost saving s all have width k but are progressively shifted r [logical] [FORMAL] C234: Interior automation arises at the lowest cost savings under Nash behavior, at intermediate savings under cooperation, an [logical] [FORMAL] depends: C233 C235: Customer support, software services, and back-office operations across competing financial institutions are three settin [empirical] depends: C207 C236: Platform ecosystems make demand spirals concrete: when a platform automates seller support, gig logistics, or content mo [logical] depends: C216 C237: Symmetry in the model rules out heterogeneity across firms and workers. [structural] depends: C215 C238: Benzell et al. (2015) study robot adoption in an overlapping-generations setting and Korinek and Stiglitz (2019) study A [structural] C239: A coalition of M firms internalizes M/N of the aggregate demand loss, but the remaining fraction (1 − M/N) is still exte [logical] [FORMAL] depends: C154 C240: The automation rate αi is not contractible among firms: it is an internal organizational choice that rival firms cannot [empirical] C241: A firm that delays automation while rivals proceed loses market share. [logical] depends: C172 C242: Large N makes defection from coalitions harder to detect and punishment harder to sustain. [logical] C243: In the convex-cost free entry case, NFE exceeds N* in over 94% of parameterizations satisfying the proposition's conditi [empirical] depends: C184 C244: If reabsorption keeps pace with automation, the externality may remain too small to detect, and the paper's contribution [logical] depends: C208 C245: Under the Pigouvian automation tax, the firm's first-order condition yields α^NE(τ) = (s − τ − ℓ/N)/k. [logical] [FORMAL] C246: Setting α^NE(τ) equal to α^CO = (s − ℓ)/k yields the optimal tax τ* = ℓ(1 − 1/N). [logical] [FORMAL] depends: C245 C247: At the optimal tax τ*, all firms choose the cooperative automation level α^CO. [logical] [FORMAL] depends: C246 C248: When total tax revenue τ*LNα^CO is rebated equally to all firms, each firm's profit is restored to π^CO. [logical] [FORMAL] depends: C247 C249: When ϕ > 1, the left-hand side of the symmetric first-order condition (LHS(α) = kα) is strictly increasing with slope k. [logical] [FORMAL] C250: The right-hand side of the symmetric first-order condition is strictly decreasing in α because the numerator is proporti [logical] [FORMAL] C251: The equation LHS = RHS in the symmetric first-order condition has a unique solution. [logical] [FORMAL] depends: C249, C250 C252: α^NE(ϕ) > α^NE(1) when ϕ > 1: AI productivity amplification increases the Nash equilibrium automation rate above the bas [logical] [FORMAL] depends: C251 C253: At the baseline equilibrium α = α^NE(1) = (s − ℓ/N)/k, RHS exceeds LHS due to the positive market-share term. [logical] [FORMAL] C254: Total revenue equals aggregate demand D regardless of how output is allocated across firms, because expenditure D is pin [structural] C255: The cooperative planner's first-order condition depends only on costs and is independent of ϕ. [logical] [FORMAL] depends: C254 C256: α^CO(ϕ) = (s − ℓ)/k = α^CO(1): the cooperative optimal automation level is invariant to AI productivity ϕ. [logical] [FORMAL] depends: C255 C257: Worker income W = wLN[1 − (1 − η)ᾱ] does not depend on ϕ. [structural] [FORMAL] C258: Owner surplus K = D − ΣCi at symmetric profiles does not depend on ϕ (neither D nor Ci depends on ϕ). [structural] [FORMAL] C259: S(μ) is ϕ-invariant at every symmetric ᾱ, and α^SP(μ; ϕ) = α^SP(μ; 1) for all μ. [logical] [FORMAL] depends: C257, C258 C260: The over-automation wedge α^NE(ϕ) − α^SP(μ; ϕ) is strictly larger than α^NE(1) − α^SP(μ; 1) for every μ when ϕ > 1. [logical] [FORMAL] depends: C252, C259 C261: The over-automation wedge is strictly increasing in ϕ. [logical] [FORMAL] depends: C260 C262: α^NE(ϕ) is increasing in ϕ. [logical] [FORMAL] depends: C252 C263: For the frictionless benchmark proof, the assumptions are k = 0, λ = 1, 0 < κ < A, and ℓ > s. [definitional] [FORMAL] C264: ℓ > s implies N* > 1 by Corollary 1. [logical] [FORMAL] depends: C263 C265: For N ≤ N*, Corollary 1 gives α = 0, and per-firm profit Π*(N) = A/N, which is strictly decreasing. [logical] [FORMAL] depends: C263 C266: For N > N*, full automation is dominant and per-firm profit drops by Δ = L(ℓ − s) > 0, giving Π*(N) = A/N − Δ, also stri [logical] [FORMAL] depends: C263 C267: Δ = L(ℓ − s) > 0 because ℓ > s. [logical] [FORMAL] depends: C263 C268: At the crossing point m = ⌊N*⌋, Π*(m) > Π*(m+1), so Π* is strictly decreasing on N. [logical] [FORMAL] depends: C265, C266 C269: Since κ < A, Π*(1) = A > κ, so the set S = {N ∈ ℕ : Π*(N) ≥ κ} is nonempty. [logical] [FORMAL] depends: C263 C270: Since Δ > 0, Π*(N) → −Δ < 0 as N → ∞, so the set S is finite. [logical] [FORMAL] depends: C267 C271: N^FE = max S exists and is unique, and satisfies Π*(N^FE) ≥ κ and Π*(N^FE + 1) < κ. [logical] [FORMAL] depends: C269, C270, C268 C272: On the no-automation branch (N ≤ m), Π*(N) = A/N ≥ κ if and only if N ≤ A/κ. [logical] [FORMAL] depends: C265 C273: On the full-automation branch (N ≥ m+1), Π*(N) = A/N − Δ ≥ κ if and only if N ≤ A/(κ + Δ). [logical] [FORMAL] depends: C266 C274: Case (i) Low entry cost: When κ + Δ ≤ A/(m+1), N^FE = ⌊A/(κ + Δ)⌋ ≥ m+1 > N*, and every firm fully automates. [logical] [FORMAL] depends: C273 C275: Case (ii) Intermediate entry cost: When κ + Δ > A/(m+1) and κ < A/m, N^FE = m and no firm automates. The threat of autom [logical] [FORMAL] depends: C272, C273 C276: In Case (ii), profit is Π*(m) = A/m > κ (strictly), sustaining positive profits without any automation occurring. [logical] [FORMAL] depends: C275 C277: Case (iii) High entry cost: When κ ≥ A/m, N^FE = ⌊A/κ⌋ and no firm automates. [logical] [FORMAL] depends: C272 C278: Cases (i), (ii), and (iii) exhaust all κ ∈ (0, A). When κ > A, N^FE = 0. [logical] [FORMAL] depends: C274, C275, C277 C279: In the frictionless benchmark, profit follows A/N for N ≤ N* (no automation) and drops discretely by Δ = L(ℓ − s) at N* [structural] [FORMAL] depends: C265, C266 C280: In intermediate entry cost case, the threat of automation deters the marginal entrant, sustaining positive profits witho [logical] depends: C275, C276 C281: In high entry cost case, entry costs alone limit competition and the automation threshold is never approached. [logical] depends: C277 C282: The function g(·) is decreasing in α^NE because ℓ > s implies the derivative is always negative. [logical] [FORMAL] C283: By Proposition 1, α^NE is non-decreasing in N. [logical] [FORMAL] C284: g(α^NE(N)) is non-increasing in N because g is decreasing in α^NE and α^NE is non-decreasing in N. [logical] [FORMAL] depends: C282, C283 C285: π^NE(N) is strictly decreasing in N because A/N is decreasing in N and g(α^NE(N)) is non-increasing in N. [logical] [FORMAL] depends: C284 C286: lim_{N→∞} π^NE < 0 < κ because s < ℓ and C = (λ − 1)wL < 0. [logical] [FORMAL] C287: C = (λ − 1)wL < 0. [logical] [FORMAL] C288: If N^FE > N*, then α^NE(N^FE) = min{(s − ℓ/N^FE)/k, 1} > 0. [logical] [FORMAL] C289: Since ℓ > s implies α^CO = max{0, (s − ℓ)/k} = 0, we have α^NE(N^FE) > α^CO when N^FE > N*, so over-automation persists [logical] [FORMAL] depends: C288 C290: The symmetric equilibrium with endogenous wages is a fixed point: ᾱ such that ᾱ = α^NE(w(ᾱ)). [definitional] [FORMAL] C291: The planner's per-firm marginal benefit of automation uses s = w − c and ℓ = λ(1 − η)w. [definitional] [FORMAL] C292: Each firm's private marginal benefit differs from the planner's because it depends on ᾱ only through its own revenue sha [structural] C293: g'(α) = w'(α)[1 − λ(1 − η)] − k, where the first term is weakly negative and k > 0, so g' < 0. [logical] [FORMAL] C294: g(α) = h(α) − ℓ(α)(1 − 1/N) for all α. [logical] [FORMAL] C295: At the Nash equilibrium fixed point, h(α^NE) = 0, so g(α^NE) = −ℓ(α^NE)(1 − 1/N) < 0. [logical] [FORMAL] depends: C294 C296: Since g is strictly decreasing and g(α^CO) = 0, the inequality g(α^NE) < 0 = g(α^CO) implies α^CO < α^NE under endogenou [logical] [FORMAL] depends: C293, C295 C297: N*(w) = λ(1 − η)w/(w − c) is strictly decreasing in w. [logical] [FORMAL] C298: Since w'(ᾱ) ≤ 0 by assumption, w(ᾱ) ≤ w(0) for all ᾱ ≥ 0. [logical] [FORMAL] C299: N*(w(ᾱ)) ≥ N*(w(0)), with strict inequality whenever w(ᾱ) < w(0): wage depression from automation raises the automation [logical] [FORMAL] depends: C297, C298 C300: Under the conditions of Proposition 9, for μ ∈ [0, 1), if g_μ(α) := s(w(α)) − ℓ(w(α))[1 + μ/(λ(1 − μ))] − kα is strictly [logical] [FORMAL] depends: C296 C301: g_μ relates to private marginal benefit h by g_μ(α) = h(α) − ℓ(α)[1 − 1/N + μ/(λ(1 − μ))]. [logical] [FORMAL] C302: At the Nash equilibrium, g_μ(α^NE) = −ℓ(1 − 1/N) − ℓμ/(λ(1 − μ)) < 0 since ℓ > 0, N ≥ 2, and μ > 0. [logical] [FORMAL] depends: C301 C303: If g_μ is strictly decreasing, the inequality g_μ(α^NE) < 0 = g_μ(α^SP) implies α^SP(μ) < α^NE. [logical] [FORMAL] depends: C302 C304: C_μ := 1 − (1 − η)[λ + μ/(1 − μ)]. At μ = 0, C_0 = 1 − λ(1 − η) ≥ 0. [definitional] [FORMAL] C305: For μ > 0, C_μ decreases. As long as C_μ ≥ 0 (which holds for all μ ≤ μ̄), the monotonicity argument for g_μ applies dir [logical] [FORMAL] depends: C304 C306: μ̄ = [1 − (1 − η)λ] / [2 − η − (1 − η)λ], which is approximately 0.48 at λ = 0.5, η = 0.30. [empirical] [FORMAL] depends: C305 C307: When C_μ < 0, the product w'C_μ ≥ 0, so g_μ' < 0 requires k > |w'(α)| · |C_μ|: integration frictions must dominate wage [logical] [FORMAL] depends: C304 C308: For numerical parameters w(ᾱ) = 1 − 0.5ᾱ, c = 0.30, λ = 0.5, η = 0.30, k = 1, N = 7, μ = 0.3, the equilibrium rates are [empirical] C309: The ordering α^SP < α^CO < α^NE holds in the numerical example, confirming the theoretical predictions. [empirical] depends: C308 C310: The distributional premium is substantial: the μ-planner would reduce automation to near zero, well below the cooperativ [empirical] depends: C308 C311: With capital income recycling, aggregate demand is D = A + λwLN − ℓLNᾱ + ηΠ, where total profit Π = D − NL(w − sᾱ) − (k/ [structural] [FORMAL] C312: ℓ̂_η = ℓ − ηs. When k = 0, the demand expression gives equation (12). [definitional] [FORMAL] depends: C311 C313: ∂Rev_i/∂α_i = −ℓ̂_η L / [N(1 − η)] − ηkLα_i / [N(1 − η)]. [logical] [FORMAL] depends: C311 C314: The firm's marginal profit including direct cost saving sL and marginal friction kLα_i depends only on α_i, so the equil [logical] [FORMAL] depends: C313 C315: N̂ := N(1 − η̂) + η̂ appears in the equilibrium condition with capital income recycling. [definitional] [FORMAL] C316: When k = 0, the marginal profit reduces to L(s − ℓ̂_η / [N(1 − η)]), a constant independent of α_i. [logical] [FORMAL] depends: C313 C317: Full automation is strictly dominant when N > N̂_η and no automation is strictly dominant when N < N̂_η, where N̂_η = ℓ̂ [logical] [FORMAL] depends: C316 C318: ℓ̂_η = ℓ − ηs ≤ 0 if and only if η ≥ ℓ/s. [logical] [FORMAL] depends: C312 C319: When ℓ̂_η ≤ 0 (i.e., η ≥ ℓ/s), the demand-depressing effect of automation is fully offset by capital income recycling. [logical] [FORMAL] depends: C318 C320: For k > 0 with capital income recycling, the first-order condition reproduces Proposition 1 with N replaced by N̂. [logical] [FORMAL] depends: C315 C321: At symmetric profiles with capital income recycling, total profit has a 1/(1 − η̂) multiplier that scales the objective [logical] [FORMAL] C322: With capital income recycling, α^CO = max{0, (s − ℓ)/k}, the same as in Proposition 1. [logical] [FORMAL] depends: C321 C323: The Pigouvian tax τ* = ℓ(1 − 1/N) is proportional to the demand leakage parameter ℓ and increasing in the number of firm [logical] [FORMAL] depends: C246 C324: The tax-and-rebate scheme is revenue-neutral: each firm's post-rebate profit equals the cooperative profit π^CO. [logical] depends: C248 C325: For the generalized planner, S(μ) = μW + (1 − μ)K. [definitional] [FORMAL] C326: The assumption π^NE(1) = Π₀(1) > κ ensures that entry is viable for at least one firm. [structural] [FORMAL] C327: λ = 1 corresponds to full recycling of consumer spending. [definitional] C328: k = 0 corresponds to the frictionless case with no integration costs. [definitional] === CONTRADICTIONS (16) === [TENSION] A: C15: Historically, labor displacement has largely been self-correcting: automation of existing tasks B: C18: Displacement has intensified over the past four decades while the creation of new work has not → C15 asserts a general historical pattern of self-correction, while C18 says that over the past four decades the offsetti [TENSION] A: C121: Historical technological transitions have often eventually reabsorbed displaced workers at hig B: C123: Past displacement episodes have consistently produced η < 1: displaced workers suffer large, p → C121 suggests workers end up better off after reabsorption (implying η > 1 eventually), while C123 says displacement con [TENSION] A: C121: Historical technological transitions have often eventually reabsorbed displaced workers at hig B: C225: Displaced workers suffer large, persistent earnings losses, as documented by Jacobson et al. ( → C121 claims eventual reabsorption at higher wages, while C225 cites evidence of large, persistent earnings losses, creat [TENSION] A: C23: Rational, forward-looking firms should theoretically serve as the brake on over-automation if t B: C31: Foresight alone cannot prevent the race toward over-automation. → C23 suggests foresight should theoretically restrain firms, while C31 concludes foresight is insufficient. Although the [TENSION] A: C53: AI systems deliver substantial productivity gains. B: C46: The technology ecosystem may be biased toward 'so-so' automation that displaces workers without → C53 asserts substantial productivity gains from AI, while C46 suggests much AI-driven automation may not produce large g [AMBIGUITY] A: C12: Only a Pigouvian automation tax can eliminate the over-automation externality. B: C125: Raising η through retraining programs, wage insurance, and incentives for new firm creation is → C12 claims the Pigouvian tax is the only instrument that can eliminate the externality, while C125 describes raising η t [TENSION] A: C213: No amount of retraining, income support, or bargaining will slow the automation arms race; onl B: C125: Raising η through retraining programs, wage insurance, and incentives for new firm creation is → C213 flatly denies that retraining can slow the automation arms race, while C125 characterizes retraining as a 'direct l [TENSION] A: C132: Despite not correcting the externality, UBI serves a complementary role by cushioning profit l B: C133: When the number of firms is endogenous, UBI can paradoxically widen the externality by attract → C132 frames UBI as a beneficial complement, while C133 shows UBI can paradoxically worsen the very problem it is meant t [AMBIGUITY] A: C175: Higher ϕ raises output but lowers the price in proportion, leaving the planner's objective inv B: C180: The lower price from higher ϕ means each dollar of spending buys more physical output, so nomi → C175 says the planner's welfare objective is invariant to ϕ, while C180 acknowledges that real consumption gains exist f [AMBIGUITY] A: C208: The empirical signature of profit erosion requires displacement at a scale and speed beyond wh B: C25: Over 100,000 tech workers were laid off in 2025 alone, with AI cited as a primary driver in mor → C208 suggests displacement has not yet reached the scale needed to observe the predicted externality, while C25 reports [AMBIGUITY] A: C67: Workers have a higher marginal propensity to consume (MPC) than owners. B: C69: Owners spend none of their income in the sector in the baseline. → C67 states an empirical claim about relative MPCs, while C69 assumes the extreme case that owners have zero sectoral MPC [AMBIGUITY] [Z3-encodable] A: C115: Higher η (income replacement rate) shrinks ℓ and thereby the over-automation wedge. B: C9: Universal basic income cannot eliminate the over-automation externality. → C115 shows that raising η shrinks the wedge, and one might expect UBI to raise η. However, C127 clarifies UBI is uncondi [TENSION] A: C213: No amount of retraining, income support, or bargaining will slow the automation arms race; onl B: C169: The Pigouvian automation tax is potentially self-reinforcing and self-limiting: it funds progr → C213 says retraining/income support cannot slow the arms race at all, but C169 implies that retraining funded by the tax [TENSION] A: C22: Public discourse increasingly treats the dynamic of AI-driven displacement eroding demand as an B: C244: If reabsorption keeps pace with automation, the externality may remain too small to detect, an → C22 presents the narrative of inevitable demand erosion, while C244 hedges that the externality may never materialize at [AMBIGUITY] [Z3-encodable] A: C82: If N ≤ N* = ℓ/s, no firm automates (αNE = 0). B: C90: A monopolist (N=1) fully internalizes the externality so αNE = αCO. → C82 says no firm automates when N ≤ N*. For a monopolist (N=1), N* = ℓ/s ≥ 1 (since ℓ ≥ s in the interesting case), so α [AMBIGUITY] A: C34: The surplus loss from over-automation is not a transfer from workers to firm owners; it is a de B: C104: Workers lose wage income directly through displacement from over-automation. → C34 emphasizes the loss is not a transfer but deadweight, while C104 highlights a direct distributional channel (workers === KEY TERMS (76) === AI labor displacement: The replacement of human workers by AI on existing tasks, reducing employment and worker income. task-based model: A production framework in which output is decomposed into tasks, and firms choose which tasks remain competitive task-based model: The paper’s setting with multiple symmetric firms choosing automation task by task while competing i demand externality: The core externality in the paper: when a firm automates, it lowers worker income and thus aggregate automation externality: The same product-market demand externality generated by automation; it causes privately rational aut automation arms race: A competition-driven process in which each firm keeps automating because it is individually profitab over-automation: Automation beyond the cooperative or planner-efficient level because firms do not internalize the fu over-automation wedge: The gap between the equilibrium automation rate chosen by firms and the efficient benchmark chosen c cooperative optimum: The automation rate that maximizes aggregate owner surplus or total profit across firms, without dir generalized social planner: A planner who chooses automation to maximize a weighted sum of worker income and owner surplus, μW + μ-planner: Shorthand for the generalized social planner whose concern for workers is indexed by μ. Nash equilibrium: The strategic outcome in which each firm’s automation choice is optimal given the game; in this mode strictly dominant strategy: An action that remains privately optimal for a firm regardless of what rival firms do; the model’s e Prisoner’s Dilemma: The frictionless limiting case in which full automation is individually dominant for every firm even deadweight loss: A pure surplus loss from over-automation that harms both workers and firm owners, rather than merely Red Queen effect: The dynamic in which firms automate to gain market share over rivals, but when all do so symmetrical reinstatement effect: Acemoglu and Restrepo’s term for the creation of new tasks and occupations that offsets displacement labor-market self-correction: The historical tendency for displaced workers to be reabsorbed through new tasks, occupations, or hi product-market mechanism: The paper’s inefficiency channel: automation reduces spending on firms’ output by lowering worker in labor-market mechanism: An alternative inefficiency channel, contrasted with this paper’s, in which problems arise through w aggregate demand spillovers: Spending linkages across firms or sectors through which one agent’s income affects others’ sales; th big push models: Models in which demand complementarities make simultaneous investment collectively profitable even w mirror image of big push: The paper’s contrast with big-push logic: here automation is privately profitable for each firm but demand complementarities: Cross-firm or cross-sector links in demand such that one firm’s wage and spending decisions affect t automation rate (αi): The share of firm i’s tasks that are automated; equivalently, the fraction of its workforce laid off task-position: A unit of work within a firm, initially performed by one human worker, that may later be automated. symmetric firms: Firms assumed to be identical in technology, task endowment, and strategic environment. per-task cost saving (s): The direct labor-cost saving from automating one task, defined as s = w − c. integration cost: The convex cost of incorporating AI into progressively harder tasks, which makes marginal automation adjustment cost parameter (k): The parameter governing the strength of convex integration costs; higher k restrains automation. comparative-advantage ordering of tasks: The ranking of tasks from easier to harder to automate, which justifies rising marginal integration perfect-substitutes limit: The baseline case where human and AI performance yield the same unit output, so automation affects o frictionless case: The special case k = 0, where there are no integration costs moderating automation. AI productivity (ϕ): The output produced by an AI-performed task relative to a human-performed task; higher ϕ strengthens market-share gain: The extra private benefit from automation when more productive AI lets a firm expand output relative marginal propensity to consume (MPC): The share of income an agent spends rather than saves; in the paper workers have a higher MPC than o sectoral MPC (λ): The fraction of worker income spent on the sector’s good; it governs how much lost wages reduce sect MPC asymmetry: The assumption that workers spend more of their income on the sector than owners do, so shifting inc income replacement rate (η): The fraction of displaced wage income replaced through reemployment, transfers, or other sources. effective demand loss per automated task (ℓ): The demand reduction caused by automating one task, defined as ℓ = λ(1−η)w. demand-loss parameter: Another name for ℓ, the parameter measuring the size of the externality created by each automated ta under-automation: The opposite distortion that arises when η > 1, so displacement raises labor income and firms automa autonomous demand (A): The component of sector demand that does not depend on current worker income; policies like UBI ente owner surplus (K): Aggregate profit or owner welfare, used in the cooperative benchmark and the planner’s objective. worker income (W): Aggregate wage income after accounting for displacement and any income replacement. competition / market fragmentation (N): The number of firms in the sector; a higher N means each firm internalizes a smaller share of the de automation threshold (N*): The cutoff number of firms above which automation becomes privately worthwhile in the baseline model monopoly benchmark: The N = 1 case in which a single firm fully internalizes the demand effect of its own automation, so free entry: An extension in which the number of firms is endogenous; entry can increase fragmentation and worsen endogenous entry equilibrium (N^FE): The number of firms that enter when entry continues until profits fall to the entry-cost condition. threat of automation as entry barrier: The idea that even the possibility of post-entry automation can deter new entrants and sustain marke exogenous wages: The baseline assumption that the wage w is fixed and does not respond to automation. endogenous wage adjustment: The extension in which wages respond to automation; it can shift when the distortion becomes active fixed-point equilibrium under endogenous wages: An equilibrium where the average automation rate equals firms’ best response evaluated at the wage g universal basic income (UBI): An unconditional transfer modeled as an increase in autonomous demand A; it raises spending levels b capital income tax: A proportional tax on profits that changes profit levels but not the per-task automation incentive, worker equity participation: Giving workers a claim on firm profits so some capital income is recycled into demand; it narrows bu profit-sharing rate (ϵ): The fraction of firm profits assigned to workers under worker equity participation. capital-income recycling: The case in which owners spend some profit income in the sector, partially offsetting the demand los effective demand loss with recycling (ℓ̂ or ℓη̂): The net demand loss after accounting for capital-income recycling; it is the demand-loss parameter a Coasian bargaining: Private bargaining among affected parties as a possible cure for the externality; the paper argues i grand coalition: A coalition containing all firms; only such a coalition would fully internalize the multilateral dem contractibility: Whether automation choices can be observed, verified, and written into enforceable agreements; the p sunk costs / irreversibility: The idea that AI investments are hard to reverse, which weakens private bargaining solutions and rei Pigouvian automation tax: A per-automated-task tax designed to make firms internalize the external demand damage caused by aut optimal Pigouvian tax (τ*): The baseline corrective tax equal to the demand loss imposed on rivals, τ* = ℓ(1−1/N). tax-and-rebate scheme: A policy that taxes automation at the Pigouvian rate and rebates the proceeds lump-sum, restoring co externality margin: The per-task automation decision margin where incentives must be altered to correct the distortion. upskilling / retraining: Policies that raise η by helping displaced workers move into new or better-paid jobs; useful as comp severance payment (σ): A per-task payment to displaced workers that, in the model, is equivalent to raising the income repl cheap talk: Communication among firms that does not change payoffs or dominant incentives, so it cannot stop the coordination failure: An inefficiency caused by choosing the wrong equilibrium; the paper contrasts this with its setting, distributional premium: The extra wedge between the cooperative benchmark and the social-planner benchmark that comes from g surplus loss: The welfare loss from the Nash outcome relative to the cooperative or planner optimum; in the model Tinbergen’s principle: The policy principle that each distinct market failure requires its own targeted instrument; here th full recycling of consumer spending: The λ = 1 case in which workers spend all their income on the sector’s good. === AGENT CAST (10) === Proponent side: Proponent | provider: openai prompt: You are Proponent for {{ topic.name }}; defend the repaired theory from inside its task-based model and competitive task-based model, treating AI labo... Skeptic side: Skeptic | provider: openai prompt: You are Skeptic for {{ topic.name }}. Your primary line is the counter-thesis: the alleged automation externality is theoretically overstated and empi... Steelman side: Neutral | provider: openai prompt: You are Steelman for {{ topic.name }}; rescue the theory by identifying the smallest set of claims that can be kept while dropping whatever must be dr... Generalist side: Neutral | provider: anthropic prompt: You are Generalist for {{ topic.name }}, an equal-opportunity referee who stress-tests both sides without taking one. Separate empirical claims in T1,... IOTheorist side: Skeptic | provider: anthropic prompt: You are IOTheorist, an industrial-organization theorist for {{ topic.name }}. Attack T2, T3, and T5 by asking whether the automation externality is an... Welfarist side: Skeptic | provider: openai prompt: You are Welfarist, a welfare-economics and public-economics specialist for {{ topic.name }}. Challenge T4 and the strong version of T5 by testing whet... Econometrician side: Neutral | provider: openai prompt: You are Econometrician, an empirical methodology and causal-inference specialist for {{ topic.name }}. Stress-test T1 and the empirical relevance of T... LaborEconomist side: Proponent | provider: anthropic prompt: You are LaborEconomist, a labor-economics specialist arguing the strongest proponent case for {{ topic.name }}. Defend T1 and the η-logic in T5 by cen... TaxScholar side: Proponent | provider: openai prompt: You are TaxScholar, a public-finance and mechanism-design specialist arguing the strongest repaired policy case for {{ topic.name }}. Focus on T6 and ... Macroeconomist side: Neutral | provider: anthropic prompt: You are Macroeconomist, a macroeconomic and open-economy general-equilibrium specialist for {{ topic.name }}. Test whether the demand externality in T... === GATE RULES (16) === [RULE-1] It is misleading to present labor displacement as a broadly self-correcting historical pattern without qualification. Th [RULE-2] Evidence of persistent earnings losses means the typical displaced worker cannot simply be characterized as eventually e [RULE-3] If Jacobson-style displacement evidence is treated as relevant evidence about displaced workers, it undercuts any broad [RULE-4] In the competitive task-based model, foresight alone does not stop over-automation. Seeing the collective harm is not th [RULE-5] AI may have high potential or frontier productivity, but much realized deployment can still be low-value or 'so-so' auto [RULE-6] A Pigouvian automation tax is the only full corrective instrument in the baseline model, but it is not the only policy t [RULE-7] Retraining, wage insurance, and other η-raising policies can attenuate the automation incentive by reducing demand loss [RULE-8] UBI can be complementary in the fixed-N baseline by cushioning living standards and profits, but under endogenous entry [RULE-9] The planner objective being invariant to AI productivity φ is a feature of the model's nominal objective or normalizatio [RULE-10] Observed AI-related layoffs show meaningful displacement, but they do not by themselves establish that the model's full [RULE-11] The baseline assumption that owners spend none of their income in the sector is a much stronger modeling simplification [RULE-12] In this model, higher η shrinks the wedge, but UBI is defined as raising autonomous demand A rather than η. Therefore UB [RULE-13] Tax-funded retraining can matter dynamically because higher η lowers ℓ and reduces the future corrective tax needed. So [RULE-14] The paper supports a structural-vulnerability claim more than a claim of an already inevitable or clearly measurable cri [RULE-15] Saying a monopolist fully internalizes the externality means the monopolist's automation choice equals the cooperative o [RULE-16] Workers do lose wage income when automation displaces them, but the overall surplus loss is not a pure transfer to owner === RUBRIC (6 criteria) === R1: notation_fidelity Uses the paper’s formal objects correctly and consistently: e.g., distinguishes the automation choice α_i from average a R2: argument_survival Assesses whether the side’s central claims remained standing after the opponent’s strongest direct rebuttals. For propon R3: concession_honesty Rewards clear acknowledgment of genuinely landed points. Examples: proponents conceding that the model is stylized, that R4: historical_self_correction Directly evaluates engagement with the core skeptical counter-thesis that automation has historically been self-correcti R5: formal_consistency Judges whether the side handled the model’s internal logic without contradiction. This includes the claim that each firm R6: policy_margin_specificity Assesses whether the side precisely addressed the strong policy claim that only a Pigouvian automation tax eliminates th