Pith. sign in

REVIEW 4 major objections 4 minor 130 references

HUF turns AI governance into a constrained optimization problem with a computed redistribution floor and an automation ceiling.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

T0 review · deepseek-v4-flash

2026-08-02 10:21 UTC pith:NAGZXA4R

load-bearing objection Useful survey and a sensible governance reframing, but the headline threshold is mis-derived and self-contradictory — the numbers don't stand as written. the 4 major comments →

arxiv 2607.26068 v1 pith:NAGZXA4R submitted 2026-06-23 econ.GN cs.AIcs.CYq-fin.EC

The Human Utility Factor: A Computable Welfare Metric That Reframes AI Governance as a Constrained Optimisation Problem

Sivasathivel Kandasamy This is my paper
classification econ.GN cs.AIcs.CYq-fin.EC
keywords Human Utility FactorAI governanceconstrained optimizationredistribution thresholdautomation depthwelfare metricGini coefficientreinforcement learning
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

This paper tries to establish that AI governance can be reframed as a constrained optimization problem with a computable objective: the Human Utility Factor (HUF), a multiplicative welfare index of Agency, Wellbeing, and Economic Stability driven by three levers — automation depth, redistribution intensity, and employment coverage. The central result is that HUF admits a closed-form interior optimum automation level and a minimum redistribution intensity below which no automation level is welfare-positive. If true, a regulator could compute enforceable inequalities (h_a ≤ h*_a, α+β ≥ (α+β)*, ρ ≥ ρ_min) from public statistics, turning "AI should benefit humanity" from aspiration into measurable compliance conditions. The paper also reports that independently trained reinforcement-learning agents converge to a high-automation, low-redistribution equilibrium that maximizes raw HUF while violating its intended floor, which the authors use to argue that the redistribution constraint must be enforced separately from the aggregate score.

Core claim

The paper's claim is that the aggregate welfare effect of automation can be summarized by HUF = A × W × E, where A captures whether workers keep sufficient income and meaningful time, W captures gains in family and rest time, and E captures inequality-adjusted output and employment coverage. Setting the derivative of HUF with respect to automation hours to zero gives the optimal automation depth h*_a; evaluating the derivative at zero gives a closed-form redistribution threshold (α+β)* = 1 − φ(1−G0)/(κδ+φ(1−G0)) below which every automation level reduces welfare. The threshold depends only on the productivity ceiling, the Gini sensitivity to automation, the capital-capture rate, and baseline

What carries the argument

The carrying object is the Human Utility Factor (HUF), a differentiable multiplicative welfare index HUF = A × W × E. The three levers are automation depth h_a (weekly hours of work replaced), redistribution intensity α+β (share of freed hours or income returned to workers via upskilling and transfers), and employment coverage ρ. The multiplicative structure enforces joint adequacy: any component at zero zeroes the whole index. The crucial identity is the redistribution threshold of Eq. (14), (α+β)* = 1 − φ(1−G0)/(κδ+φ(1−G0)), which divides the parameter space into regimes where automation can be welfare-positive and regimes where no automation level helps; together with the interior optimum

Load-bearing premise

The load-bearing premise is that the Gini coefficient responds linearly to automation with a fixed sensitivity parameter in Eq. (12), but the cited evidence estimates employment and wage effects, not Gini sensitivity, so neither the linear form nor κ=0.02 is directly established; the ad hoc α+β≤0.55 cap in §5.1 additionally compresses the Nordic regime's realized outcomes.

What would settle it

A cross-country panel regression of the Gini coefficient on robot density or automation exposure would settle whether Eq. (12)'s linear sensitivity is real; if the estimated coefficient differs materially from 0.02 or a nonlinear specification fits better, Eq. (14)'s threshold and the BAU 'moratorium' verdict shift or collapse. Within the paper's own simulation, re-running the Nordic regime without the α+β≤0.55 cap would test whether the empirical Nordic advantage over Partial recovers the analytical prediction.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • Any jurisdiction can compute a minimal redistribution intensity from public data; below that floor, no amount of automation raises welfare, so a moratorium or enhanced-review trigger is justified.
  • The optimal automation level is not a knife-edge: the paper finds HUF within 5% of peak across h_a ∈ [15,30] hours/week at Partial/Nordic redistribution, so regulators can set an admissible band rather than a point target.
  • Employment coverage ρ scales HUF multiplicatively; a coverage floor is a necessary co-condition for any automation-welfare claim.
  • Composite welfare metrics that do not separately enforce a redistribution floor can be optimized into high-automation, low-redistribution equilibria; the floor must be an independent compliance condition, not just a component of the score.
  • Under US-calibrated parameters the redistribution threshold is never met, so the paper's verdict is a warranted pause on further automation expansion until redistribution capacity is strengthened.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • My inference: the PPO finding generalizes — any aggregate objective that rewards freed time without a hard redistribution floor will be gamed toward high automation, so the same floor separation is needed if HUF is ever used as an AI training reward, a use the paper floats only as future work.
  • My inference: applied to high-inequality economies with G0 ≈ 0.5–0.6, Eq. (14) implies a higher redistribution floor and a narrower automation band; the paper notes this direction but does not quantify it, suggesting the governance instrument would bite hardest in emerging markets.
  • My inference: the paper's own displacement-vs-augmentation distinction suggests a direct empirical test — estimate the augmentation elasticity ψ in its Eq. (15) extension from firm-level wage-share and hours data; if augmentation dominates, the threshold falls and the ceiling rises, making HUF less restrictive than the current displacement-only model implies.
  • My inference: the demand-collapse channel the paper flags as under-modelled is the most serious threat to HUF's sufficiency; a joint welfare-and-demand floor, with the purchasing-power condition w·ρ ≥ median expenditure, is the natural next object and would catch contractionary states HUF currently marks as acceptable.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

4 major / 4 minor

Summary. The paper proposes the Human Utility Factor (HUF), a multiplicative welfare metric HUF = A × W × E that combines Agency, Wellbeing, and Economic Stability as functions of automation depth h_a, redistribution intensity α+β, and employment coverage ρ. It claims two headline formal results: a closed-form interior optimum h*_a and a redistribution threshold (α+β)* below which no automation level is welfare-positive, both computed from publicly available statistics. The framework is tested in a three-agent Stackelberg MARL simulation across BAU/U.S., Partial/Canada, and Nordic regimes, with analytical heuristic agents and independent PPO agents. The paper argues that AI governance should be reframed as constrained optimization, with enforceable inequalities h_a ≤ h*_a, α+β ≥ (α+β)*, and ρ ≥ ρ_min.

Significance. If the derivation and calibration were sound, HUF would be a genuinely useful contribution: it offers a transparent, differentiable welfare index, a systematic survey of 32 governance frameworks, a concrete regulatory translation, and a falsifiable quantitative threshold. The GAGI index in Section 3 is a sensible descriptive tool for inequality- and inflation-adjusted welfare monitoring. The paper is also commendably explicit about its limitations, including the redistribution cap and the absence of sectoral decomposition. However, the central formal claim — Eq. (14) — is not correctly derived, is internally inconsistent with the simulation results, and depends on parameters for which no direct empirical estimate is provided. Since the governance triplet and all regime verdicts in Table 2 and Section 5 are built on this threshold, the paper's headline quantitative contribution is not supported as it stands.

major comments (4)
  1. [§4.6, Eq. (14)] The redistribution threshold is not the first-order condition of HUF. Setting d(HUF)/dh_a|ha=0 = 0 and solving for α+β must include the derivatives of the Wellbeing and income-sufficiency factors at h_a=0. The paper's assertion that the threshold is independent of μ, ν, and w* because W(0)=I(0)=1 confuses the level with the slope. With the paper's own parameters (α=β, μ=ν=0.5, h0f=7, h0r=8, h0w=40, w*=0.85, δ=0.6, φ=0.066, G0=0.41, κ=0.02), solving the full FOC gives a value materially different from Eq. (14). Moreover, §5.2 reports a BAU threshold 'α+β≈0.42' and attributes it to Eq. (14), but Eq. (14) with the stated BAU parameters yields approximately 0.236. This internal inconsistency means the headline enforceable threshold is neither correctly derived nor consistently computed.
  2. [§4.5 and §5] The validation loop is circular. The Economic Stability component E defines γ as 'estimated from the MARL simulation of Section 5', and the same simulation is later offered as cross-validation of the analytical optima. Both the analytical heuristic agents and the PPO agents optimize the same HUF objective, so their agreement (or disagreement) tests internal consistency of the optimization, not whether HUF corresponds to any external welfare benchmark. The claimed 'agreement' between analytical and learned agents therefore does not provide independent validation of the model.
  3. [Eq. (12) and §4.5] The Gini sensitivity κ=0.02 is attributed to Acemoglu–Restrepo [6], but that paper estimates effects on employment-to-population ratios and wages, not on the Gini coefficient. Reference [119] is later described as the primary source for κ calibration, but no estimated κ is reported from either source. Since Eq. (14) and all regime verdicts in Table 2 are first-order sensitive to κ, the claim that (α+β)* is computable from public statistics is not presently supported. A concrete estimate of κ from the cited work, or a sensitivity analysis over κ, is needed before the threshold can be used as claimed.
  4. [§5.3.2, Fig. 7] The PPO agents discover an alternate optimum with h*_a ≈ 30–40 hrs/week in BAU, with MAE 23.25 hrs/week relative to the analytical prediction. The paper interprets this as a second local maximum, but this directly contradicts the framing of h*_a as 'the closed-form optimal automation level.' If HUF has multiple optima, then the interior optimum derived from d(HUF)/dh_a=0 is only a stationary point, not a global welfare maximum. The governance conclusion that h_a ≤ h*_a is an enforceable ceiling is therefore unsupported: a regulator enforcing h*_a would be enforcing one local optimum that the paper's own learning agents show is dominated by another region with higher raw HUF.
minor comments (4)
  1. [Table 2] The reported HUF at h_a=0 is inconsistent with the model. With ρ=0.95 and γ=1, Eq. (13) gives HUF(0)=ρ^{1+γ}=0.9025 for all regimes, but Table 2 reports 0.895, 0.908, and 0.931 for BAU, Partial, and Nordic. The source of this discrepancy should be clarified.
  2. [Abstract and §5.3.2] The abstract states that 'both analytical and PPO-based agents identify welfare-optimal operating regions,' but the PPO agents do not validate the analytical optimum; they find a different equilibrium. The wording should distinguish validation from discovery of alternate optima.
  3. [§5.1] The environment is described as an 18-dimensional state but the state variables are not enumerated in the main text. Since the supplementary information is not part of this submission, the reader cannot reproduce the simulation. At minimum, a full state-variable list and hyperparameter table should be included.
  4. [§6.2] The 'Absent sectoral decomposition' limitation acknowledges that a decomposition promised in the Introduction is not delivered. This is a useful admission, but the Introduction should be adjusted to avoid overstating the model's coverage.

Circularity Check

3 steps flagged

The paper's headline 'predictions' are partly constructed from the simulation that is then offered as validation: E's elasticity is fitted to the MARL run, analytical agents are programmed to the closed form they 'validate', and the reported BAU threshold (0.42) corresponds not to Eq. (14) with the stated κ=0.02 but to the simulation-learned κ≈0.047.

specific steps
  1. fitted input called prediction [Section 4.5 (Component E) and Section 5.1 (Experimental Setup)]
    "γ > 0 is the labour-absorption elasticity estimated from the MARL simulation of Section 5. ... Agreement between the two families constitutes an internal cross-validation: a learning agent that independently recovers the analytically predicted optimum confirms the theoretical result without relying on it."

    The Economic Stability component E contains a parameter γ that is estimated from the very MARL simulation that Section 5 later treats as validating HUF. Both the analytical agents and the PPO agents optimize the same HUF=A×W×E objective, so their agreement is an internal-consistency check, not an external test. The validation target is therefore built into the model before the 'confirmation' run.

  2. self definitional [Section 5.1 and Section 5.3.1 (Figure 4)]
    "analytically-derived heuristic agents that implement the closed-form optimality conditions of Eqs. (13) and (14) (Industry moves ha toward h∗ a via Brent's method) ... The analytical agents, by construction, move toward the closed-form optimum h∗ a derived in §4.6."

    The 'theory–experiment agreement' in Figure 4 (MAE 0.12–0.20 hrs/wk) is presented as confirming d(ln HUF)/dha=0, but the analytical agents were explicitly programmed to move ha toward the h∗a computed from that same closed form. Their convergence to h∗a is therefore an artefact of the solver, not an independent empirical confirmation.

  3. fitted input called prediction [Section 5.1, Section 5.2, Figure 2 caption]
    "the BAU surface declines monotonically in ha below the α+β≈0.42 threshold — exactly the critical value implied by Eq. (14) for the BAU parameters. ... Government sets the automation ceiling h̄a, the ALMP floor (α+β)min, and the Gini-sensitivity parameter κ. ... Figure 2: HUF Surface over ha×(α+β), with ρ, δ, κ Fixed at Learned Values."

    With the paper's stated BAU parameters (φ=0.066, κ=0.02, δ=0.6, G0=0.41), Eq. (14) yields (α+β)*≈0.236, not 0.42. To obtain 0.42 requires κ≈0.047, which is the MARL-learned value of κ, not the cited κ=0.02. Thus the reported threshold is a fitted output of the simulation (κ learned to match the surface) that is then renamed as a prediction 'implied by Eq. (14)'. The central enforceable threshold is not independently predicted; it is the simulation's fit re-labelled.

full rationale

The algebraic skeleton of the paper (Eqs. 2–14) is self-contained in the narrow sense that the closed-form expressions are computed from the model's own components; no load-bearing self-citation chain or imported uniqueness theorem is present, and references [16]–[18] are companion papers that do not by themselves force the central result. However, the paper's claimed empirical validations and reported parameter-dependent thresholds reduce, on inspection, to the fitting targets. E's γ is estimated from the Section 5 simulation and the same simulation is presented as validation; the analytical agents are constructed to implement Eqs. (13)–(14) and then their convergence to h∗a is reported as confirming those equations; and the BAU threshold α+β≈0.42 that Figure 2/§5.2 attribute to Eq. (14) is numerically inconsistent with the stated κ=0.02 and instead matches the κ≈0.047 learned inside the MARL environment. These are not independent confirmations: they are internal-consistency loops. Separately, the derivation of Eq. (14) appears to drop W'(0) and I'(0) terms from the first-order condition, which is a correctness problem rather than a circularity and is not scored here beyond the fit issue. On balance, one central prediction (the redistribution threshold) and the main validation protocol are partly constructed from the simulation's fitted parameters, warranting a score of 6 rather than higher: the analytical form itself still has independent algebraic content, but the reported numbers and 'confirmations' are not externally grounded.

Axiom & Free-Parameter Ledger

7 free parameters · 7 axioms · 2 invented entities

The HUF result depends on a chain of hand-set functional forms and parameter assignments: multiplicative separability, linear Gini response with κ=0.02, the α+β≤0.55 cap, and γ fitted from the validation simulation. These are not derived from first principles and several are not independently measured.

free parameters (7)
  • κ (Gini sensitivity) = 0.02
    Set in §4.5/Table 3 and attributed to Acemoglu–Restrepo [6], but the cited JPE paper measures employment-to-population and wage effects, not Gini sensitivity; effectively hand-set.
  • δ (capital-capture rate) = 0.60
    Assigned in Table 3 for all regimes; no independent estimate cited; shifts labour share, wage, and Gini equations.
  • γ (labour-absorption elasticity) = 1 in Table 2; 'estimated from MARL simulation' in §4.5
    If fitted to the same simulation that is used for validation, the HUF optimum is partly an artefact of the fitting.
  • w* (living-wage floor) = 0.85
    Chosen in Table 2; sets the kink in I(ha) and determines when Agency erosion activates.
  • µ, ν (wellbeing elasticities) = 0.5 each
    Chosen in Table 2 as 'estimated from time-use surveys' [110,120] but no direct estimate is reported; W and the location of h*_a depend on them despite the claimed independence of Eq. (14).
  • ρ (employment coverage) = 0.95 for analytical runs
    Chosen; scales HUF multiplicatively; the proposed ρ_min floor is not derived.
  • φ (productivity gain cap) = 0.066
    External Acemoglu ceiling [9]; used in Ypc and Eq. (14); not fitted to the paper's data but load-bearing for conclusions.
axioms (7)
  • ad hoc to paper The welfare metric is exactly the product HUF = A × W × E, with no interaction terms or compensation across components.
    Design postulate in §4.1 Eq. (2), asserted to enforce joint adequacy; it is not derived from welfare theory and drives the threshold algebra.
  • ad hoc to paper Agency decreases linearly with ha unless redistribution is complete: η(ha)=1+Ωha/T with Ω=α+β−1 (Eq. 8).
    Linear hour-reallocation efficiency; no empirical evidence is given for linearity.
  • ad hoc to paper Gini responds linearly to automation depth: G(ha)=G0+κδ ha/h0w(1−α−β) (Eq. 12).
    No empirical Gini-automation elasticity is cited; κ=0.02 is assigned via [6], which estimates employment and wage effects, not Gini changes.
  • domain assumption The time budget is T=55h/week from 5×(24−13) with two protected weekend days (Eq. 3), and h0w≈40, h0f≈7, h0r≈8.
    BLS time-use and sleep norms; reasonable but normative.
  • domain assumption Productivity gain φ is capped by the Acemoglu ceiling at 0.066 (Eq. 11).
    External bound [9], load-bearing for the threshold magnitude.
  • ad hoc to paper The simulation enforces α+β≤0.55, beyond which active production collapses.
    Introduced to keep the game stable; prevents the Nordic regime from being fully realised and changes empirical comparisons.
  • domain assumption Demand collapse is not part of HUF; the demand index D only captures first-order effects.
    Acknowledged in §6.2 as under-modelled; if demand collapse is a co-equal failure mode, thresholds based on HUF alone may be too permissive.
invented entities (2)
  • Human Utility Factor (HUF = A×W×E) no independent evidence
    purpose: Scalar welfare metric and proposed governance constraint; the central contribution.
    No falsifiable out-of-sample prediction is made; parameters are chosen or fitted inside the same framework, and the only validation is self-simulation.
  • Gini-Adjusted GDP per Capita Index (GAGI) no independent evidence
    purpose: Empirical index used in §3 to document the governance gap; defined in Eq. (1) and developed in companion paper [17].
    A re-scaling of published data; cannot falsify its own construction, and the §3 evidence is sourced to a self-cited companion paper rather than shipped data.

pith-pipeline@v1.3.0-alltime-deepseek · 29333 in / 18862 out tokens · 180035 ms · 2026-08-02T10:21:16.127727+00:00 · methodology

0 comments
read the original abstract

Existing AI governance frameworks, including the EU AI Act and NIST AI RMF, address safety, transparency, and accountability but do not operationalize quantitative constraints on macro-socioeconomic stability. As a result, AI systems may satisfy regulatory requirements while contributing to labor displacement, rising inequality, and reduced economic resilience. We introduce the Human Utility Factor (HUF), a differentiable welfare metric that models the interaction between Agency, Wellbeing, and Economic Stability as functions of three actionable policy levers: automation depth, redistribution intensity, and employment coverage. HUF yields a closed-form optimal automation level and a minimum redistribution threshold below which no level of automation is welfare-positive, transforming high-level governance objectives into computable constraints. We evaluate HUF using a three-agent multi-agent reinforcement learning framework across U.S., Canadian, and Nordic policy regimes. Both analytical and PPO-based agents identify welfare-optimal operating regions and reveal a critical failure mode: welfare metrics that do not explicitly constrain redistribution can converge to high-automation equilibria that satisfy the metric while undermining its intended societal objectives. Our results suggest that AI governance is fundamentally a constrained optimization problem rather than a compliance exercise. HUF provides a quantitative framework for evaluating automation policies, identifying socioeconomic stability boundaries, and supporting governance decisions under accelerating AI deployment.

Figures

Figures reproduced from arXiv: 2607.26068 by Sivasathivel Kandasamy.

Figure 1
Figure 1. Figure 1: Gini-Adjusted GDP per Capita Index (GAGI), GDP per Capita Index, and Private AI Investment, G7 Economies, 2010–2026 (2025–2026 pre￾liminary). Observation. The GDP per capita index (blue line, left axis) consistently exceeds the GAGI index (red line, left axis) in every G7 country throughout the pe￾riod, with the gap widening post-2022, coincident with accelerating AI investment (bars, right axis, country-s… view at source ↗
Figure 2
Figure 2. Figure 2: HUF Surface over ha×(α+β), with ρ, δ, κ Fixed at Learned Values. One panel per regime; z-axis: HUF. Circle: MARL-optimal (ha, α+β). Triangle: analytical h ∗ a (Eq. (13)). Observation. BAU declines monotonically as ha increases below the α+β ≈ 0.42 threshold, with no interior maximum visible. Partial shows a pronounced ridge near ha ≈ 20 hrs/wk for α+β ≥ 0.45; the MARL circle and analytical triangle fall wi… view at source ↗
Figure 3
Figure 3. Figure 3: Training Convergence Dashboard — Analytical Agents, Partial Regime. Six quantities over 600 episodes (raw: faint; smoothed-20: solid). Top row: h ∗ a (Industry), α+β (Population/Government), δ (Industry). Bottom row: κ (Govern￾ment), HUF (Eq. (2)), GAGI (Eq. (1)). Observation. All six quantities stabilise within ≈50 episodes; α+β settles near the redistribution threshold (α+β) ∗ ≈ 0.34; HUF remains above t… view at source ↗
Figure 4
Figure 4. Figure 4: Theory vs. Experiment: Analytical h ∗ a vs. Empirical h ∗ a , Partial Regime. Left: scatter of MARL-empirical h ∗ a (y-axis) against closed-form prediction (x-axis); colour encodes HUF; dashed diagonal = perfect agreement; MAE = 0.12 hrs/wk. Right: per-episode convergence trace (purple: empirical; orange dashed: analytical tar￾get). Observation. Points cluster tightly along the diagonal with no systematic … view at source ↗
Figure 5
Figure 5. Figure 5: Cross-Regime Comparison — Analytical Agents. Left: analytical HUF(ha) curves for BAU, Partial, Nordic; stars mark theoretical optima ( [PITH_FULL_IMAGE:figures/full_fig_p019_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Episode Returns per Agent — Neural PPO, BAU Regime. Raw (faint) and smoothed-20 (solid) returns for Industry (red), Government (blue), Popula￾tion (green). Observation. Industry’s returns rise from ≈ −120 to ≈ −20 over ≈300 episodes; Government and Population exhibit persistently higher variance throughout. Interpretation. Industry’s clear learning curve confirms the game is learnable from re￾ward feedback… view at source ↗
Figure 7
Figure 7. Figure 7: Theory vs. Experiment: Analytical h ∗ a vs. Empirical h ∗ a , Neural PPO, BAU Regime. Left: scatter as [PITH_FULL_IMAGE:figures/full_fig_p021_7.png] view at source ↗
Figure 3.5
Figure 3.5. Figure 3.5: share of employment in occupations at highest automation risk, by [PITH_FULL_IMAGE:figures/full_fig_p034_3_5.png] view at source ↗
Figure 8
Figure 8. Figure 8: HUF Component Decomposition over Training — Neural PPO, BAU Regime. Top-left: Agency A; top-right: Wellbeing W; bottom-left: Economic Stability E; bottom-right: HUF = A×W ×E. Dashed line: welfare floor (HUF = 1.0 or component baseline). Observation. W rises sharply to ≈2.0–2.4 (top-right); A declines to ≈0.65– 0.75 (top-left); E remains near 1.0 (bottom-left); net HUF rises to ≈1.4–1.7 (bottom￾right). Inte… view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

130 extracted references · 8 canonical work pages

  1. [1]

    March 2026 job cuts report: AI leads reasons

    Challenger, Gray & Christmas. March 2026 job cuts report: AI leads reasons. Technical report, Challenger, Gray & Christmas, Inc., apr 2026. URLhttps://www.challengergray.com/blog/ challenger-report-march-cuts-rise-25-from-february-ai-leads-reasons/. Accessed April 2026

    2026

  2. [2]

    CFOs admit privately that AI layoffs will be 9x higher this year—and still a fraction of ‘doomsday’ predic- tions.Fortune, mar 2026

    Julia Carpenter. CFOs admit privately that AI layoffs will be 9x higher this year—and still a fraction of ‘doomsday’ predic- tions.Fortune, mar 2026. URLhttps://fortune.com/2026/03/24/ 26 cfo-survey-ai-job-cuts-productivity-paradox-2026/. Accessed April 2026

    2026

  3. [3]

    Tech layoffs 2025: Why AI is behind the rising job cuts.Final Round AI Blog, 2025

    Final Round AI. Tech layoffs 2025: Why AI is behind the rising job cuts.Final Round AI Blog, 2025. URLhttps://www.finalroundai.com/blog/ ai-tech-layoffs-mid-2025. Accessed April 2026

    2025

  4. [4]

    AI-washing and the massive layoffs hitting the economy.CNBC, nov 2025

    Melissa Repko and Steve Liesman. AI-washing and the massive layoffs hitting the economy.CNBC, nov 2025. URLhttps://www.cnbc.com/2025/11/04/ white-collar-layoffs-ai-cost-cutting-tariffs.html. Accessed April 2026

    2025

  5. [5]

    The race between man and machine: Im- plications of technology for growth, factor shares, and employment.American Eco- nomic Review, 108(6):1488–1542, 2018

    Daron Acemoglu and Pascual Restrepo. The race between man and machine: Im- plications of technology for growth, factor shares, and employment.American Eco- nomic Review, 108(6):1488–1542, 2018. doi: 10.1257/aer.20160696

    doi:10.1257/aer.20160696 2018

  6. [6]

    Robots and jobs: Evidence from US labor markets.Journal of Political Economy, 128(6):2188–2244, 2020

    Daron Acemoglu and Pascual Restrepo. Robots and jobs: Evidence from US labor markets.Journal of Political Economy, 128(6):2188–2244, 2020. doi: 10.1086/ 705716

    2020

  7. [7]

    GPTs are GPTs: Labor market impact potential of large language models.Science, 384(6702):1306– 1308, 2024

    Tyna Eloundou, Sam Manning, Pamela Mishkin, and Daniel Rock. GPTs are GPTs: Labor market impact potential of large language models.Science, 384(6702):1306– 1308, 2024. doi: 10.1126/science.adj0998

    doi:10.1126/science.adj0998 2024

  8. [8]

    Panton, Carlo Pizzinelli, Emma Rockall, and Marina M

    Mauro Cazzaniga, Florence Jaumotte, Longji Li, Giovanni Melina, Augustus J. Panton, Carlo Pizzinelli, Emma Rockall, and Marina M. Tavares. Gen-AI: Arti- ficial intelligence and the future of work. Staff Discussion Note SDN/2024/001, International Monetary Fund, 2024. URLhttps://www.imf.org/-/media/files/ publications/sdn/2024/english/sdnea2024001.pdf

    2024

  9. [9]

    The simple macroeconomics of AI.Economic Policy, 40(121): 13–58, 2025

    Daron Acemoglu. The simple macroeconomics of AI.Economic Policy, 40(121): 13–58, 2025. doi: 10.3386/w32487. NBER Working Paper 32487, 2024

    doi:10.3386/w32487 2025

  10. [10]

    Openai charter, 2018

    OpenAI. Openai charter, 2018. URLhttps://openai.com/charter/. Accessed: 2026-04-24

    2018

  11. [11]

    Company visionandmission, 2026

    Anthropic PBC. Company visionandmission, 2026. URLhttps://www.anthropic. com/company. Accessed: 2026-04-24

    2026

  12. [12]

    Regulation (EU) 2024/1689 (artificial intelli- gence act)

    European Parliament and Council. Regulation (EU) 2024/1689 (artificial intelli- gence act). Official Journal of the European Union, OJ L 2024/1689, 2024

    2024

  13. [13]

    AI risk management framework (AI RMF 1.0)

    Elham Tabassi. AI risk management framework (AI RMF 1.0). Technical Report NIST AI 100-1, National Institute of Standards and Technology, 2023

    2023

  14. [14]

    Recommendation on the ethics of artificial intelligence

    UNESCO. Recommendation on the ethics of artificial intelligence. SHS/BIO/REC-AIETHICS/2021, 2021. URLhttps://www.unesco.org/en/ artificial-intelligence/recommendation-ethics

    2021

  15. [15]

    Hiroshima process international code of conduct for organizations developing advanced AI systems, 2023

    G7. Hiroshima process international code of conduct for organizations developing advanced AI systems, 2023

    2023

  16. [16]

    Mapping the governance of machines: A state-of-the-art review of AI governance frameworks and the socioeconomic gaps they leave

    Anonymous. Mapping the governance of machines: A state-of-the-art review of AI governance frameworks and the socioeconomic gaps they leave. arXiv preprint, 2026. Companion review paper; arXiv identifier to be inserted. 27

    2026

  17. [17]

    GAGI: A Gini-adjusted GDP-per-capita index for distribution-aware macroeconomic welfare monitoring

    Sivasathivel Kandasamy. GAGI: A Gini-adjusted GDP-per-capita index for distribution-aware macroeconomic welfare monitoring. arXiv preprint arXiv:2606.09944 [econ.GN], 2026.https://arxiv.org/abs/2606.09944

    Pith/arXiv arXiv 2026

  18. [18]

    Supplementary information: Experimental setup and decision-space analysis for the Human Utility Factor

    Sivasathivel Kandasamy. Supplementary information: Experimental setup and decision-space analysis for the Human Utility Factor. Extended Data, submitted with this manuscript, 2026

    2026

  19. [19]

    Human oversight under article 14 of the EU AI Act

    Maximilian Fink. Human oversight under article 14 of the EU AI Act. SSRN, 2025

    2025

  20. [20]

    Demystifying the draft EU arti- ficial intelligence act.Computer Law Review International, 22(4):97–112, 2021

    Michael Veale and Frederik Zuiderveen Borgesius. Demystifying the draft EU arti- ficial intelligence act.Computer Law Review International, 22(4):97–112, 2021

    2021

  21. [21]

    Trustworthy artificial in- telligence and the European Union AI Act.Regulation & Governance, 2024

    Johann Laux, Sandra Wachter, and Brent Mittelstadt. Trustworthy artificial in- telligence and the European Union AI Act.Regulation & Governance, 2024. doi: 10.1111/rego.12512

    doi:10.1111/rego.12512 2024

  22. [22]

    Top 20 predictions from experts on AI job loss, 2026

    AIMultiple Research. Top 20 predictions from experts on AI job loss, 2026. URL https://aimultiple.com/ai-job-loss. Accessed April 2026

    2026

  23. [23]

    Young workers’ employment drops in occupa- tions with high AI exposure.Dallas Fed Economics, jan 2026

    Federal Reserve Bank of Dallas. Young workers’ employment drops in occupa- tions with high AI exposure.Dallas Fed Economics, jan 2026. URLhttps: //www.dallasfed.org/research/economics/2026/0106. Accessed April 2026

    2026

  24. [24]

    The global decline of the labor share

    Loukas Karabarbounis and Brent Neiman. The global decline of the labor share. Quarterly Journal of Economics, 129(1):61–103, 2014. doi: 10.1093/qje/qjt032

    doi:10.1093/qje/qjt032 2014

  25. [25]

    Schumpeter meeting Keynes: A policy-friendly model of endogenous growth and business cycles.Journal of Eco- nomic Dynamics and Control, 34(9):1748–1767, 2010

    Giovanni Dosi, Giorgio Fagiolo, and Andrea Roventini. Schumpeter meeting Keynes: A policy-friendly model of endogenous growth and business cycles.Journal of Eco- nomic Dynamics and Control, 34(9):1748–1767, 2010. doi: 10.1016/j.jedc.2010.06. 018

    work page doi:10.1016/j.jedc.2010.06 2010

  26. [26]

    Sterman.Business Dynamics: Systems Thinking and Modeling for a Com- plex World

    John D. Sterman.Business Dynamics: Systems Thinking and Modeling for a Com- plex World. Irwin/McGraw-Hill, 2000. ISBN 0-07-238915-X

    2000

  27. [27]

    Meyarivan

    Kalyanmoy Deb, Amrit Pratap, Sameer Agarwal, and T. Meyarivan. A fast and elitist multiobjective genetic algorithm: NSGA-II.IEEE Transactions on Evolu- tionary Computation, 6(2):182–197, 2002. doi: 10.1109/4235.996017

    arXiv 2002

  28. [28]

    Oxford University Press, 1990

    Joel Mokyr.The Lever of Riches: Technological Creativity and Economic Progress. Oxford University Press, 1990

    1990

  29. [29]

    David H. Autor. Why are there still so many jobs? the history and future of workplace automation.Journal of Economic Perspectives, 29(3):3–30, 2015. doi: 10.1257/jep.29.3.3

    doi:10.1257/jep.29.3.3 2015

  30. [30]

    Framework convention on artificial intel- ligence, CETS 225

    Council of Europe. Framework convention on artificial intel- ligence, CETS 225. Council of Europe Treaty Series, 2024. URLhttps://www.coe.int/en/web/artificial-intelligence/ the-framework-convention-on-artificial-intelligence

    2024

  31. [31]

    Recommendation of the council on artificial intelligence

    OECD. Recommendation of the council on artificial intelligence. OECD/LEGAL/0449, 2019. URLhttps://legalinstruments.oecd.org/ en/instruments/OECD-LEGAL-0449. 28

    2019

  32. [32]

    Responsible scaling policy, v2.0, 2024

    Anthropic. Responsible scaling policy, v2.0, 2024. URLhttps://www.anthropic. com/news/announcing-our-updated-responsible-scaling-policy

    2024

  33. [33]

    Preparedness framework, v2, 2025

    OpenAI. Preparedness framework, v2, 2025. URLhttps://cdn.openai.com/pdf/ 18a02b5d-6b67-4cec-ab64-68cdfbddebcd/preparedness-framework-v2.pdf

    2025

  34. [34]

    Frontier safety framework, v1.0, 2024

    Google DeepMind. Frontier safety framework, v1.0, 2024. URLhttps://deepmind. google/blog/introducing-the-frontier-safety-framework/

    2024

  35. [35]

    Frontier AI framework, 2025

    Meta. Frontier AI framework, 2025

    2025

  36. [36]

    Joseph R. Biden. Executive order 14110: Safe, secure, and trustworthy development and use of artificial intelligence. 88 Fed. Reg. 75191, 2023

    2023

  37. [37]

    Blueprint for an AI bill of rights: Making automated systems work for the american people, 2022

    White House OSTP. Blueprint for an AI bill of rights: Making automated systems work for the american people, 2022. URLhttps://www.whitehouse.gov/ostp/ ai-bill-of-rights/

    2022

  38. [38]

    AI RMF: Generative AI profile

    Elham Tabassi, Chloe Autio, Jesse Dunietz, Sanjay Jain, Mark Stanley, Peter Hall, and Karen Roberts. AI RMF: Generative AI profile. Technical Report NIST AI 600-1, National Institute of Standards and Technology, 2024

    2024

  39. [39]

    A pro- innovation approach to AI regulation

    UK Department for Science, Innovation and Technology. A pro- innovation approach to AI regulation. CP 815, 2023. URLhttps: //assets.publishing.service.gov.uk/media/65c1e399c43191000d1a45f4/ a-pro-innovation-approach-to-ai-regulation-amended-governement-response-web-ready. pdf

    2023

  40. [40]

    Interim measures for the management of generative AI services, 2023

    Cyberspace Administration of China et al. Interim measures for the management of generative AI services, 2023

    2023

  41. [41]

    Model AI gov- ernance framework, 2nd edition, 2020

    Personal Data Protection Commission Singapore and IMDA. Model AI gov- ernance framework, 2nd edition, 2020. URLhttps://www.pdpc.gov.sg/ help-and-resources/2020/01/model-ai-governance-framework

    2020

  42. [42]

    Directive on automated decision-making,

    Treasury Board of Canada Secretariat. Directive on automated decision-making,

  43. [43]

    Senate bill 24-205: Consumer protections for artificial intelligence, 2024

    Colorado General Assembly. Senate bill 24-205: Consumer protections for artificial intelligence, 2024

    2024

  44. [44]

    Projeto de lei 2338/2023, 2023

    Federal Senate of Brazil. Projeto de lei 2338/2023, 2023

    2023

  45. [45]

    ISO/IEC 42001:2023 — artificial intelligence — management system, 2023

    ISO/IEC. ISO/IEC 42001:2023 — artificial intelligence — management system, 2023

    2023

  46. [46]

    IEEE std 7010-2020: Recommended practice for assessing the impact of autonomous and intelligent systems on human well-being, 2020

    IEEE. IEEE std 7010-2020: Recommended practice for assessing the impact of autonomous and intelligent systems on human well-being, 2020

    2020

  47. [47]

    Responsible AI standard, v2, 2022

    Microsoft. Responsible AI standard, v2, 2022. URLhttps://query.prod.cms.rt. microsoft.com/cms/api/am/binary/RE5cmFl. 29

    2022

  48. [48]

    Introducing Google’s Secure AI Framework (SAIF)

    Royal Hansen and Phil Venables. Introducing Google’s Secure AI Framework (SAIF). Google Blog, June 2023. URLhttps://blog.google/technology/ safety-security/introducing-googles-secure-ai-framework/. Framework documentation:https://saif.google/

    2023

  49. [49]

    AILuminate: Introducing v1.0 of the AI risk and reliability benchmark from ML- Commons

    Shaona Ghosh, Heather Frase, Adina Williams, Scott Luger, Paul Röttger, et al. AILuminate: Introducing v1.0 of the AI risk and reliability benchmark from ML- Commons. arXiv preprint arXiv:2503.05731, 2025. URLhttps://arxiv.org/abs/ 2503.05731

    Pith/arXiv arXiv 2025

  50. [51]

    Asilomar AI principles

    Future of Life Institute. Asilomar AI principles. Asilomar Conference on Beneficial AI, January 5–8, 2017, Pacific Grove, CA, 2017. URLhttps://futureoflife. org/open-letter/ai-principles/. 23 principles across Research Issues, Ethics and Values, and Longer-term Issues. Signed by>8,000 researchers and technology leaders

    2017

  51. [52]

    Montréal declaration for a responsible de- velopment of artificial intelligence, December 2018

    Université de Montréal. Montréal declaration for a responsible de- velopment of artificial intelligence, December 2018. URLhttps: //montrealdeclaration-responsibleai.com/. Seven core values: well-being, autonomy, justice, privacy, knowledge, democracy, ecological responsibility

    2018

  52. [53]

    AI Now 2023 Landscape: Confronting Tech Power

    Amba Kak and Sarah Myers West. AI Now 2023 Landscape: Confronting Tech Power. AI Now Institute, April 2023. URLhttps://ainowinstitute.org/ 2023-landscape.html

    2023

  53. [54]

    Pause giant AI experiments: An open let- ter, March 2023

    Future of Life Institute. Pause giant AI experiments: An open let- ter, March 2023. URLhttps://futureoflife.org/open-letter/ pause-giant-ai-experiments/. Signed by>33,000 signatories including Yoshua Bengio, Stuart Russell, and Elon Musk

    2023

  54. [55]

    PAI’s responsible practices for synthetic media: A frame- work for collective action, February 2023

    Partnership on AI. PAI’s responsible practices for synthetic media: A frame- work for collective action, February 2023. URLhttps://syntheticmedia. partnershiponai.org/. 18+ signatory organisations. Grey literature / industry framework

    2023

  55. [56]

    The bletchley declaration by countries attending the AI safety summit, 1– 2 november 2023, November 2023

    UK Government and AI Safety Summit Participating Countries. The bletchley declaration by countries attending the AI safety summit, 1– 2 november 2023, November 2023. URLhttps://www.gov.uk/government/ publications/ai-safety-summit-2023-the-bletchley-declaration/ the-bletchley-declaration-by-countries-attending-the-ai-safety-summit-1-2-november-2023. Signed...

    2023

  56. [57]

    Statement on inclusive and sustainable ar- tificial intelligence for people and the planet, February 2025

    AI Action Summit. Statement on inclusive and sustainable ar- tificial intelligence for people and the planet, February 2025. URLhttps://www.elysee.fr/en/emmanuel-macron/2025/02/11/ statement-on-inclusive-and-sustainable-artificial-intelligence-for-people-and-the-planet. 30 AI Action Summit, Paris, February 10–11, 2025. Signed by>100 countries. Official in...

    2025

  57. [58]

    Automationandnewtasks: Howtechnology displaces and reinstates labor.Journal of Economic Perspectives, 33(2):3–30, 2019

    DaronAcemogluandPascualRestrepo. Automationandnewtasks: Howtechnology displaces and reinstates labor.Journal of Economic Perspectives, 33(2):3–30, 2019. doi: 10.1257/jep.33.2.3

    doi:10.1257/jep.33.2.3 2019

  58. [59]

    Stiglitz

    Anton Korinek and Joseph E. Stiglitz. Artificial intelligence and its implications for income distribution and unemployment. In Ajay Agrawal, Joshua Gans, and Avi Goldfarb, editors,The Economics of Artificial Intelligence: An Agenda. University of Chicago Press, 2019. doi: 10.7208/9780226613475-014

    doi:10.7208/9780226613475-014 2019

  59. [60]

    The economic potential of generative AI: The next productivity frontier

    Michael Chui, Eric Hazan, Roger Roberts, Alex Singla, Kate Smaje, Alex Sukharevsky, Lareina Yee, and Rodney Zemmel. The economic potential of generative AI: The next productivity frontier. Tech- nical report, McKinsey Global Institute, June 2023. URLhttps: //www.mckinsey.com/capabilities/mckinsey-digital/our-insights/ the-economic-potential-of-generative-...

    2023

  60. [61]

    Meaningful human control over autonomous systems: A philosophical account.Frontiers in Robotics and AI, 5:15,

    Filippo Santoni de Sio and Jeroen van den Hoven. Meaningful human control over autonomous systems: A philosophical account.Frontiers in Robotics and AI, 5:15,

  61. [62]

    On the questforeffectivenessinhumanoversight: Interdisciplinaryperspectives

    Axel Sterz, Benjamin Sorge, Patrick Schramowski, and Kristian Kersting. On the questforeffectivenessinhumanoversight: Interdisciplinaryperspectives. InProceed- ings of the ACM Conference on Fairness, Accountability, and Transparency (FAccT), 2024

    2024

  62. [63]

    Algorith- mic impact assessments: A practical framework for public agency accountability

    Dillon Reisman, Jason Schultz, Kate Crawford, and Meredith Whittaker. Algorith- mic impact assessments: A practical framework for public agency accountability. Technical report, AI Now Institute, 2018. URLhttps://ainowinstitute.org/ aiareport2018.pdf

    2018

  63. [64]

    White, Margaret Mitchell, Timnit Gebru, Ben Hutchinson, Jamila Smith-Loud, Daniel Theron, and Parker Barnes

    Inioluwa Deborah Raji, Andrew Smart, Rebecca N. White, Margaret Mitchell, Timnit Gebru, Ben Hutchinson, Jamila Smith-Loud, Daniel Theron, and Parker Barnes. Closing the AI accountability gap: Defining an end-to-end framework for internal algorithmic auditing. InProceedings of the ACM Conference on Fairness, Accountability, and Transparency (FAT*), pages 3...

    doi:10.1145/3351095 2020

  64. [65]

    Gummadi, Adish Singla, Adrian Weller, and Muhammad Bilal Zafar

    Till Speicher, Hoda Heidari, Nina Grgic-Hlaca, Krishna P. Gummadi, Adish Singla, Adrian Weller, and Muhammad Bilal Zafar. A unified approach to quantifying algorithmic unfairness: Measuring individual and group unfairness via inequality indices. InProceedings of the 24th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining, pages 2239–...

    doi:10.1145/3219819 2018

  65. [66]

    Shorrocks

    Anthony F. Shorrocks. The class of additively decomposable inequality measures. Econometrica, 48(3):613–625, 1980. doi: 10.2307/1913126

    doi:10.2307/1913126 1980

  66. [67]

    Equality of opportunity in supervised learning

    Moritz Hardt, Eric Price, and Nathan Srebro. Equality of opportunity in supervised learning. InAdvances in Neural Information Processing Systems, volume 29, 2016. 31

    2016

  67. [68]

    Inherent trade-offs in the fair determination of risk scores

    Jon Kleinberg, Sendhil Mullainathan, and Manish Raghavan. Inherent trade-offs in the fair determination of risk scores. InProceedings of Innovations in Theoretical Computer Science, 2017. doi: 10.4230/LIPIcs.ITCS.2017.43

    doi:10.4230/lipics.itcs.2017.43 2017

  68. [69]

    Fair prediction with disparate impact: A study of bias in recidivism prediction instruments.Big Data, 5(2):153–163, 2017

    Alexandra Chouldechova. Fair prediction with disparate impact: A study of bias in recidivism prediction instruments.Big Data, 5(2):153–163, 2017. doi: 10.1089/big. 2016.0047

    arXiv 2017

  69. [70]

    Bellamy, Kuntal Dey, Michael Hind, Samuel C

    Rachel K.E. Bellamy, Kuntal Dey, Michael Hind, Samuel C. Hoffman, Stephanie Houde, Kalapriya Kannan, Pranay Lohia, Jacquelyn Martino, Sameep Mehta, Alek- sandraMojsilovic, SeemaNagar, KarthikeyanNatesanRamamurthy, JohnRichards, Diptikalyan Saha, Prasanna Sattigeri, Moninder Singh, Kush R. Varshney, and Yun- feng Zhang. AI fairness 360: An extensible toolk...

    arXiv 2019

  70. [71]

    Macmillan, 1936

    John Maynard Keynes.The General Theory of Employment, Interest and Money. Macmillan, 1936

    1936

  71. [72]

    Creel, Ananya Kumar, Dan Jurafsky, and Percy Liang

    Rishi Bommasani, Kathleen A. Creel, Ananya Kumar, Dan Jurafsky, and Percy Liang. Picking on the same person: Does algorithmic monoculture lead to outcome homogenization? InAdvances in Neural Information Processing Systems, volume35, 2022

    2022

  72. [73]

    The algorithmic leviathan: Arbitrariness, fairness, andopportunityinalgorithmicdecision-makingsystems.Canadian Journal of Philosophy, 52(1):26–43, 2022

    Kathleen Creel and Deborah Hellman. The algorithmic leviathan: Arbitrariness, fairness, andopportunityinalgorithmicdecision-makingsystems.Canadian Journal of Philosophy, 52(1):26–43, 2022. doi: 10.1017/can.2021.26

    work page doi:10.1017/can.2021.26 2022

  73. [74]

    Algorithmic monoculture and social welfare

    Jon Kleinberg and Manish Raghavan. Algorithmic monoculture and social welfare. Proceedings of the National Academy of Sciences, 118(22):e2018340118, 2021. doi: 10.1073/pnas.2018340118

    doi:10.1073/pnas.2018340118 2021

  74. [75]

    Ironies of automation.Automatica, 19(6):775–779, 1983

    Lisanne Bainbridge. Ironies of automation.Automatica, 19(6):775–779, 1983. doi: 10.1016/0005-1098(83)90046-8

    doi:10.1016/0005-1098(83)90046-8 1983

  75. [76]

    The economics of skills obsolescence: A review

    Andries de Grip and Jaap Van Loo. The economics of skills obsolescence: A review. Research in Labor Economics, 21:1–26, 2002. doi: 10.1016/S0147-9121(02)21003-1

    work page doi:10.1016/s0147-9121(02)21003-1 2002

  76. [77]

    Poverty and inequality platform (PIP), version 20250401, 2025

    World Bank. Poverty and inequality platform (PIP), version 20250401, 2025. URL https://pip.worldbank.org. Indicator SI.POV.GINI (Gini index). Accessed April

    2025

  77. [78]

    World development indicators (WDI), 2025

    World Bank. World development indicators (WDI), 2025. URLhttps://databank. worldbank.org/source/world-development-indicators. Accessed April 2026. Indicators used:NY.GDP.PCAP.KD(GDP per capita, constant 2015USD) and FP.CPI.TOTL(Consumer Price Index, 2010=100), G7 economies, 2010–2024. Of- ficial statistics

    2025

  78. [79]

    World economic outlook, october 2025, October 2025

    International Monetary Fund. World economic outlook, october 2025, October 2025. URLhttps://www.imf.org/en/Publications/WEO/Issues/2025/10/. Statistical Appendix Tables A.1–A.4. G7 real GDP per capita growth rates and CPI inflation, 2025 preliminary estimates. Official IMF publication. 32

    2025

  79. [80]

    World economic outlook, april 2026, April 2026

    International Monetary Fund. World economic outlook, april 2026, April 2026. URL https://www.imf.org/en/Publications/WEO/Issues/2026/04/. G7 real GDP per capita and CPI partial-year projections, 2026. Official IMF publication

    2026

  80. [81]

    The AI index 2024 annual report

    Nestor Maslej, Loredana Fattorini, Raymond Perrault, Vanessa Parli, Anka Reuel, Erik Brynjolfsson, John Etchemendy, Katrina Ligett, Terah Lyons, James Manyika, Juan Carlos Niebles, Yoav Shoham, Russell Wald, and Jack Clark. The AI index 2024 annual report. Technical report, Institute for Human-Centered AI, Stanford University, April 2024. URLhttps://hai.s...

    Pith/arXiv arXiv 2024

Showing first 80 references.