ai-research-survey

Systematic scan of agentic development research. What's signal, what's noise.
git clone https://git.shiptheloop.com/ai-research-survey.git
Log | Files | Refs

scan.json (24001B)


      1 {
      2   "paper": {
      3     "title": "Scenarios for the Transition to AGI",
      4     "authors": ["Anton Korinek", "Donghyun Suh"],
      5     "year": 2024,
      6     "venue": "Social Science Research Network",
      7     "arxiv_id": "2403.12107",
      8     "doi": "10.2139/ssrn.4762962"
      9   },
     10   "scan_version": 2,
     11   "active_modules": [],
     12   "methodology_tags": ["theoretical"],
     13   "key_findings": "The paper develops a compute-centric economic framework showing that the effects of automation on wages depend on the distribution of task complexity and a race between automation and capital accumulation. Under unbounded task distributions with sufficiently thick tails, wages rise forever; under bounded distributions (reflecting finite human cognitive capacity), wages collapse even before full AGI. Fixed factors like minerals worsen wage outlook by bottlenecking growth, while automating R&D can lead to a growth singularity. Slowing automation can maximize wage growth but at rapidly increasing output cost.",
     14   "checklist": {
     15     "artifacts": {
     16       "code_released": {
     17         "applies": true,
     18         "answer": false,
     19         "justification": "The paper includes numerical simulations (Figure 8, etc.) with specific parameters (Table 2) but provides no code repository, GitHub link, or supplementary archive for reproducing these simulations."
     20       },
     21       "data_released": {
     22         "applies": true,
     23         "answer": false,
     24         "justification": "Simulation parameters are listed in Table 2 and the mathematical model is fully specified, but no simulation output data or scripts are released. The model specification is sufficient for reimplementation but no data artifacts are provided."
     25       },
     26       "environment_specified": {
     27         "applies": true,
     28         "answer": false,
     29         "justification": "The paper runs numerical simulations producing figures but provides no information about the computational environment, software, or libraries used."
     30       },
     31       "reproduction_instructions": {
     32         "applies": true,
     33         "answer": false,
     34         "justification": "No reproduction instructions are provided. While the mathematical model is fully specified and parameters are in Table 2, there are no step-by-step instructions for reproducing the numerical results."
     35       }
     36     },
     37     "statistical_methodology": {
     38       "confidence_intervals_or_error_bars": {
     39         "applies": false,
     40         "answer": false,
     41         "justification": "Purely theoretical paper with deterministic mathematical models and simulations. No stochastic elements requiring confidence intervals."
     42       },
     43       "significance_tests": {
     44         "applies": false,
     45         "answer": false,
     46         "justification": "No empirical comparative claims requiring significance tests. All results are derived analytically from the mathematical model."
     47       },
     48       "effect_sizes_reported": {
     49         "applies": false,
     50         "answer": false,
     51         "justification": "Theoretical paper deriving closed-form results and numerical illustrations from a mathematical model. No empirical effect sizes to report."
     52       },
     53       "sample_size_justified": {
     54         "applies": false,
     55         "answer": false,
     56         "justification": "Theoretical paper with no empirical samples."
     57       },
     58       "variance_reported": {
     59         "applies": false,
     60         "answer": false,
     61         "justification": "Deterministic simulations with fixed parameters. No variance across runs to report."
     62       }
     63     },
     64     "evaluation_design": {
     65       "baselines_included": {
     66         "applies": true,
     67         "answer": true,
     68         "justification": "The paper includes a 'business-as-usual' scenario (Section 3.1) as a baseline against which two AGI scenarios and a mixed scenario are compared. The business-as-usual scenario corresponds to traditional automation patterns similar to Aghion et al. (2019)."
     69       },
     70       "baselines_contemporary": {
     71         "applies": true,
     72         "answer": true,
     73         "justification": "The paper builds on and references contemporary related work including Acemoglu and Restrepo (2018, 2022), Aghion et al. (2019), Davidson (2023), and Besiroglu et al. (2022), positioning its contribution clearly against the current literature."
     74       },
     75       "ablation_study": {
     76         "applies": true,
     77         "answer": true,
     78         "justification": "Section 4 provides systematic extensions that function as ablations: adding fixed factors (4.1), automating R&D (4.2), nostalgic jobs (4.3), heterogeneous workers (4.4), and specific capital (4.5). Each isolates the contribution of one component to the model's predictions."
     79       },
     80       "multiple_metrics": {
     81         "applies": true,
     82         "answer": true,
     83         "justification": "The paper tracks multiple outcome variables: output (Y), wages (w), capital returns (R), labor share, fraction of unautomated tasks (1-Φ), and factor price frontiers. Figure 8 shows output split into wage bill and capital returns."
     84       },
     85       "human_evaluation": {
     86         "applies": false,
     87         "answer": false,
     88         "justification": "Purely theoretical/mathematical paper. Human evaluation is irrelevant to validating mathematical proofs and closed-form solutions."
     89       },
     90       "held_out_test_set": {
     91         "applies": false,
     92         "answer": false,
     93         "justification": "No datasets or test sets involved. This is a theoretical paper with analytical derivations and numerical illustrations."
     94       },
     95       "per_category_breakdown": {
     96         "applies": true,
     97         "answer": true,
     98         "justification": "Results are broken down across four distinct scenarios (business-as-usual, baseline AGI, aggressive AGI, mixed) in Figure 8, and separately for each extension in Section 4. Proposition 7 characterizes three distinct cases based on the rate of automation."
     99       },
    100       "failure_cases_discussed": {
    101         "applies": true,
    102         "answer": true,
    103         "justification": "The paper extensively discusses scenarios where wages collapse (bounded distributions, thin-tailed unbounded distributions, fixed factor bottlenecks). These are explicitly modeled 'failure cases' for labor welfare."
    104       },
    105       "negative_results_reported": {
    106         "applies": true,
    107         "answer": true,
    108         "justification": "The paper reports several negative outcomes: wage collapse under AGI scenarios (Figure 8 panels b,c), fixed factors causing wage decline even under infinite-tail distributions (Proposition 10, Figure 9), and the large output cost of slowing automation to protect wages (Figure 11)."
    109       }
    110     },
    111     "claims_and_evidence": {
    112       "abstract_claims_supported": {
    113         "applies": true,
    114         "answer": true,
    115         "justification": "The abstract's claims are all supported by formal results in the paper: unbounded thick-tailed distributions leading to perpetual wage growth (Proposition 7), wage collapse under bounded distributions (Proposition 5), capital accumulation race (Section 3), fixed factor bottlenecks (Proposition 10), and automating R&D lifting returns (Proposition 11)."
    116       },
    117       "causal_claims_justified": {
    118         "applies": true,
    119         "answer": true,
    120         "justification": "The paper's causal claims ('automation leads to wage changes') are derived formally from a well-specified mathematical model with explicit assumptions. For theoretical work, rigorous mathematical derivation from stated axioms constitutes adequate causal justification. All propositions include proofs."
    121       },
    122       "generalization_bounded": {
    123         "applies": true,
    124         "answer": true,
    125         "justification": "The paper is explicit about presenting 'scenarios' rather than predictions. It consistently qualifies results with the distributional assumptions required (bounded vs unbounded, thick vs thin tails). The conclusion states specific limitations and next steps. The title itself frames results as 'scenarios.'"
    126       },
    127       "alternative_explanations_discussed": {
    128         "applies": true,
    129         "answer": true,
    130         "justification": "The paper systematically explores alternative model specifications that lead to different outcomes: Section 4 analyzes five extensions (fixed factors, R&D automation, nostalgic jobs, heterogeneous skills, specific capital) that modify or reverse baseline conclusions. Proposition 10 vs Proposition 7 show how adding fixed factors reverses the infinite-tail result."
    131       },
    132       "proxy_outcome_distinction": {
    133         "applies": true,
    134         "answer": true,
    135         "justification": "The paper's claims match the granularity of its model: it analyzes wages, output, and factor returns as directly modeled quantities. The paper acknowledges that 'welfare' is broader than wages and discusses this through extensions (Section 4). No proxy gap exists between what is modeled and what is claimed."
    136       }
    137     },
    138     "setup_transparency": {
    139       "model_versions_specified": {
    140         "applies": false,
    141         "answer": false,
    142         "justification": "No AI/ML models are used in this paper. It is a theoretical economics paper using mathematical models, not machine learning systems."
    143       },
    144       "prompts_provided": {
    145         "applies": false,
    146         "answer": false,
    147         "justification": "The paper does not use any AI prompting. It is a theoretical economics paper."
    148       },
    149       "hyperparameters_reported": {
    150         "applies": true,
    151         "answer": true,
    152         "justification": "Table 2 provides all parameter values used in numerical simulations: discount rate ρ=0.04, risk aversion η=2, depreciation δ=0.1, elasticity of substitution σ=0.5, TFP A=0.5, L=1, Φ₀=0.608, K₀=4.6. Additional parameters for each scenario are specified in Section 3.1."
    153       },
    154       "scaffolding_described": {
    155         "applies": false,
    156         "answer": false,
    157         "justification": "No agentic scaffolding is used. This is a theoretical economics paper."
    158       },
    159       "data_preprocessing_documented": {
    160         "applies": false,
    161         "answer": false,
    162         "justification": "No data is collected or preprocessed. This is a theoretical paper with analytical results and numerical simulations."
    163       }
    164     },
    165     "limitations_and_scope": {
    166       "limitations_section_present": {
    167         "applies": true,
    168         "answer": false,
    169         "justification": "There is no dedicated limitations section. The conclusion mentions 'interesting next steps' (adjustment costs, endogenous innovation, distributional impacts) but these are framed as future work rather than substantive discussion of current limitations."
    170       },
    171       "threats_to_validity_specific": {
    172         "applies": true,
    173         "answer": false,
    174         "justification": "No specific threats to validity are discussed. The paper does not address sensitivity of numerical results to parameter choices, the realism of key assumptions (CES aggregation, exogenous automation growth), or the gap between stylized models and real economies."
    175       },
    176       "scope_boundaries_stated": {
    177         "applies": true,
    178         "answer": true,
    179         "justification": "The paper is explicit about what it models and what it does not. The conclusion lists unmodeled features: 'labor and capital adjustment costs, modeling endogenous innovation, analyzing distributional impacts more fully, studying macroeconomic dynamics and policies.' Each extension in Section 4 acknowledges what it excludes (e.g., Section 4.2: 'A model that comprehensively incorporates both effects is beyond the scope of this paper')."
    180       }
    181     },
    182     "data_integrity": {
    183       "raw_data_available": {
    184         "applies": true,
    185         "answer": false,
    186         "justification": "The simulation results shown in figures are not available for independent verification. No data files, simulation outputs, or code are released. However, the fully specified model and parameters in principle allow reimplementation."
    187       },
    188       "data_collection_described": {
    189         "applies": false,
    190         "answer": false,
    191         "justification": "No data is collected. This is a theoretical paper with mathematical derivations and deterministic numerical simulations from specified parameter sets."
    192       },
    193       "recruitment_methods_described": {
    194         "applies": false,
    195         "answer": false,
    196         "justification": "No participants or external data sources are recruited. This is a purely theoretical paper."
    197       },
    198       "data_pipeline_documented": {
    199         "applies": false,
    200         "answer": false,
    201         "justification": "No data pipeline exists. The paper derives results analytically and illustrates them with deterministic simulations."
    202       }
    203     },
    204     "conflicts_of_interest": {
    205       "funding_disclosed": {
    206         "applies": true,
    207         "answer": false,
    208         "justification": "No funding source is disclosed. The acknowledgments thank individuals at Anthropic, Google, GovAI, and OpenAI for 'helpful conversations and comments' but do not state any funding source."
    209       },
    210       "affiliations_disclosed": {
    211         "applies": true,
    212         "answer": true,
    213         "justification": "Author affiliations are listed: 'University of Virginia, Brookings, and GovAI.' The GovAI affiliation (Governance of AI) is relevant given the paper's AGI focus."
    214       },
    215       "funder_independent_of_outcome": {
    216         "applies": true,
    217         "answer": false,
    218         "justification": "No funding source is disclosed, making independence impossible to assess. The paper acknowledges conversations with people at Anthropic, Google, and OpenAI—companies with direct financial interests in AGI narratives—but does not discuss whether these interactions influenced the work."
    219       },
    220       "financial_interests_declared": {
    221         "applies": true,
    222         "answer": false,
    223         "justification": "No competing interests or financial interests statement is included in the paper."
    224       }
    225     },
    226     "contamination": {
    227       "training_cutoff_stated": {
    228         "applies": false,
    229         "answer": false,
    230         "justification": "This paper does not evaluate any pre-trained model on a benchmark. It is a theoretical economics paper."
    231       },
    232       "train_test_overlap_discussed": {
    233         "applies": false,
    234         "answer": false,
    235         "justification": "This paper does not evaluate any pre-trained model on a benchmark. It is a theoretical economics paper."
    236       },
    237       "benchmark_contamination_addressed": {
    238         "applies": false,
    239         "answer": false,
    240         "justification": "This paper does not evaluate any pre-trained model on a benchmark. It is a theoretical economics paper."
    241       }
    242     },
    243     "human_studies": {
    244       "pre_registered": {
    245         "applies": false,
    246         "answer": false,
    247         "justification": "No human participants. This is a theoretical economics paper."
    248       },
    249       "irb_or_ethics_approval": {
    250         "applies": false,
    251         "answer": false,
    252         "justification": "No human participants. This is a theoretical economics paper."
    253       },
    254       "demographics_reported": {
    255         "applies": false,
    256         "answer": false,
    257         "justification": "No human participants. This is a theoretical economics paper."
    258       },
    259       "inclusion_exclusion_criteria": {
    260         "applies": false,
    261         "answer": false,
    262         "justification": "No human participants. This is a theoretical economics paper."
    263       },
    264       "randomization_described": {
    265         "applies": false,
    266         "answer": false,
    267         "justification": "No human participants. This is a theoretical economics paper."
    268       },
    269       "blinding_described": {
    270         "applies": false,
    271         "answer": false,
    272         "justification": "No human participants. This is a theoretical economics paper."
    273       },
    274       "attrition_reported": {
    275         "applies": false,
    276         "answer": false,
    277         "justification": "No human participants. This is a theoretical economics paper."
    278       }
    279     },
    280     "cost_and_practicality": {
    281       "inference_cost_reported": {
    282         "applies": false,
    283         "answer": false,
    284         "justification": "Purely theoretical paper. No inference or computational system is proposed whose cost would be relevant."
    285       },
    286       "compute_budget_stated": {
    287         "applies": false,
    288         "answer": false,
    289         "justification": "Purely theoretical paper. The numerical simulations are simple enough that compute budget is trivial and irrelevant."
    290       }
    291     }
    292   },
    293   "claims": [
    294     {
    295       "claim": "Under bounded task complexity distributions, wages collapse even before full automation (AGI) is reached, regardless of savings behavior.",
    296       "evidence": "Proposition 5 proves upper and lower bounds for wages, showing that when Φ(I)=1 is reached in finite time, wages collapse to A. Figure 8 panels (b) and (c) illustrate wage collapse in baseline and aggressive AGI scenarios (20 and 5 year timelines).",
    297       "supported": "strong"
    298     },
    299     {
    300       "claim": "Under unbounded task distributions with sufficiently thick Pareto tails, wages can rise forever if capital accumulation keeps pace with automation.",
    301       "evidence": "Proposition 7 case 3 proves that if λg ≤ (A-ρ-δ)/η · (1-σ), wages grow exponentially at asymptotic rate λg/(1-σ). Figure 8 panel (a) shows wages rising at ~2% per year in the business-as-usual scenario.",
    302       "supported": "strong"
    303     },
    304     {
    305       "claim": "There exists a threshold automation index separating a labor-scarce region (w > R) from a region where labor and capital are perfect substitutes at the margin (w = R).",
    306       "evidence": "Lemma 1 derives the threshold Φ(Î) = (K/L)/(1 + K/L) and characterizes the two regions analytically, with Region 2 exhibiting AK-model behavior (equation 5).",
    307       "supported": "strong"
    308     },
    309     {
    310       "claim": "Fixed factors (minerals, matter) cause wages to eventually decline even under infinite-tailed task distributions, unlike the baseline model.",
    311       "evidence": "Proposition 10 proves the economy enters Region 2 in finite time when fixed factors are present. Figure 9 shows wages peaking after ~10 years and collapsing by year 25 in the business-as-usual scenario with fixed factors (α=0.9).",
    312       "supported": "strong"
    313     },
    314     {
    315       "claim": "Automating technological progress can lead to a Type II growth singularity (infinite output in finite time) that lifts wages.",
    316       "evidence": "Proposition 11 derives the condition Γ(I♡)(1-Φ(I♡))(1-θ) > 1 for super-exponential growth. Figure 10 shows the explosive path. The proof builds on Trammell and Korinek (2023).",
    317       "supported": "strong"
    318     },
    319     {
    320       "claim": "Slowing automation to maximize wages achieves exponential wage growth of (A-ρ-δ)/η but at rapidly increasing output cost approaching 100%.",
    321       "evidence": "Proposition 12 derives the optimal rate of automation λg = (1-σ)·(A-ρ-δ)/η. Figure 11 right panel shows the output cost grows toward 100% as foregone automation accumulates.",
    322       "supported": "strong"
    323     }
    324   ],
    325   "red_flags": [
    326     {
    327       "flag": "No sensitivity analysis on parameters",
    328       "detail": "Table 2 lists specific parameter values described as 'standard in the literature,' but the paper provides no sensitivity analysis showing how results change under alternative parameter values. The qualitative results hold generally (proven analytically), but the quantitative simulation results (e.g., timing of wage collapse in Figure 8) may be sensitive to parameter choices."
    329     },
    330     {
    331       "flag": "Potential undisclosed influence from AGI labs",
    332       "detail": "The acknowledgments thank individuals at Anthropic, Google, GovAI, and OpenAI for conversations. These organizations have direct commercial or mission-driven interests in AGI narratives. No funding disclosure or competing interests statement addresses whether these interactions influenced the framing or conclusions."
    333     },
    334     {
    335       "flag": "Calibration details sparse",
    336       "detail": "The initial conditions Φ₀=0.608 and K₀=4.6 are stated to 'match a 66% initial labor share with capital at its steady state,' but the mapping from real economic data to these model parameters is not detailed. The calibration of the Hinton-inspired AGI timelines (5 and 20 years) is taken directly from one individual's predictions without critical assessment."
    337     }
    338   ],
    339   "cited_papers": [
    340     {
    341       "title": "The race between machine and man: Implications of technology for growth, factor shares and employment",
    342       "authors": ["Daron Acemoglu", "Pascual Restrepo"],
    343       "year": 2018,
    344       "relevance": "Foundational task-based automation model that this paper extends toward the AGI limit case."
    345     },
    346     {
    347       "title": "Tasks, automation, and the rise in US wage inequality",
    348       "authors": ["Daron Acemoglu", "Pascual Restrepo"],
    349       "year": 2022,
    350       "relevance": "Empirical analysis of how task automation reshapes labor markets and wage inequality."
    351     },
    352     {
    353       "title": "Artificial intelligence and economic growth",
    354       "authors": ["Philippe Aghion", "Benjamin Jones", "Charles Jones"],
    355       "year": 2019,
    356       "relevance": "Analyzes AI-driven economic growth including the 'Type II singularity' concept that this paper builds on."
    357     },
    358     {
    359       "title": "GPTs are GPTs: An early look at the labor market impact potential of large language models",
    360       "authors": ["Tyna Eloundou", "Sam Manning", "Pamela Mishkin", "Daniel Rock"],
    361       "year": 2023,
    362       "relevance": "Empirical analysis of which tasks are amenable to LLM automation, directly relevant to the task distribution assumptions."
    363     },
    364     {
    365       "title": "How will language modelers like ChatGPT affect occupations and industries?",
    366       "authors": ["Edward Felten", "Manav Raj", "Robert Seamans"],
    367       "year": 2023,
    368       "relevance": "Empirical measurement of AI exposure across occupations, relevant to modeling which tasks get automated."
    369     },
    370     {
    371       "title": "Economic impacts of AI-augmented R&D",
    372       "authors": ["Tamay Besiroglu", "Nicholas Emery-Xu", "Neil Thompson"],
    373       "year": 2022,
    374       "doi": "2212.08198",
    375       "relevance": "Analyzes how AI accelerates R&D, directly relevant to the automated technological progress extension in Section 4.2."
    376     },
    377     {
    378       "title": "What a compute-centric framework says about AI takeoff speeds",
    379       "authors": ["Tom Davidson"],
    380       "year": 2023,
    381       "relevance": "Compute-centric model of AI capability growth that shares the 'tasks in compute space' framing with this paper."
    382     },
    383     {
    384       "title": "Explosive growth from AI automation: A review of the arguments",
    385       "authors": ["Ege Erdil", "Tamay Besiroglu"],
    386       "year": 2023,
    387       "relevance": "Reviews arguments for whether AGI could produce explosive economic growth, directly relevant to the singularity analysis."
    388     },
    389     {
    390       "title": "Economic growth under transformative AI",
    391       "authors": ["Philip Trammell", "Anton Korinek"],
    392       "year": 2023,
    393       "relevance": "Companion paper by one of the authors analyzing broader implications of transformative AI on economic growth."
    394     },
    395     {
    396       "title": "Compute trends across three eras of machine learning",
    397       "authors": ["Jaime Sevilla", "Lennart Heim", "Anson Ho", "Tamay Besiroglu", "Marius Hobbhahn", "Pablo Villalobos"],
    398       "year": 2022,
    399       "relevance": "Empirical data on compute scaling trends that motivate the exponential automation index assumption in the model."
    400     },
    401     {
    402       "title": "Language models and cognitive automation for economic research",
    403       "authors": ["Anton Korinek"],
    404       "year": 2023,
    405       "relevance": "Analyzes how LLMs automate cognitive tasks in research, directly relevant to the task complexity framework."
    406     },
    407     {
    408       "title": "Work of the past, work of the future",
    409       "authors": ["David Autor"],
    410       "year": 2019,
    411       "relevance": "Analyzes how technology reshapes labor markets through task creation and destruction, foundational to the task-based approach."
    412     }
    413   ]
    414 }

Impressum · Datenschutz