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-v4.json (21420B)


      1 {
      2   "scan_version": 4,
      3   "paper_type": "survey",
      4   "paper": {
      5     "title": "Explosive Growth from AI Automation: A Review of the Arguments",
      6     "authors": [
      7       "Ege Erdil",
      8       "Tamay Besiroglu"
      9     ],
     10     "year": 2023,
     11     "venue": "arXiv",
     12     "arxiv_id": "2309.11690",
     13     "doi": null
     14   },
     15   "checklist": {
     16     "claims_and_evidence": {
     17       "abstract_claims_supported": {
     18         "applies": true,
     19         "answer": true,
     20         "justification": "The abstract claims are hedged ('could plausibly lead to') and supported by the detailed analysis in the body. The abstract also notes 'high confidence in this outcome is unwarranted,' which matches the discussion.",
     21         "source": "opus"
     22       },
     23       "causal_claims_justified": {
     24         "applies": true,
     25         "answer": true,
     26         "justification": "Causal claims are framed within formal economic models with explicit assumptions. The paper uses conditional language ('if AI can substitute for labor, then models predict...') and derives results analytically from stated premises.",
     27         "source": "opus"
     28       },
     29       "generalization_bounded": {
     30         "applies": true,
     31         "answer": true,
     32         "justification": "The paper is explicit about conditioning on 'AI capable of broadly substituting for human labor' and notes uncertainties about regulatory responses, production bottlenecks, and automation pace. Section 4 cautions against overconfidence.",
     33         "source": "opus"
     34       },
     35       "alternative_explanations_discussed": {
     36         "applies": true,
     37         "answer": true,
     38         "justification": "The entire structure of the paper is to present arguments and counterarguments. Section 3 presents 9 alternative explanations for why explosive growth might not occur, each with quantitative analysis.",
     39         "source": "opus"
     40       },
     41       "proxy_outcome_distinction": {
     42         "applies": true,
     43         "answer": true,
     44         "justification": "The paper explicitly defines 'explosive growth' as annual GWP exceeding 130% of prior maximum, distinguishes GDP measurement from consumer surplus (Section 3.6), and discusses what GDP does and does not capture.",
     45         "source": "opus"
     46       }
     47     },
     48     "limitations_and_scope": {
     49       "limitations_section_present": {
     50         "applies": true,
     51         "answer": true,
     52         "justification": "Section 4 (Discussion) and Section 4.1 (Open questions) serve as a substantive limitations section, discussing multiple sources of uncertainty and unresolved questions.",
     53         "source": "opus"
     54       },
     55       "threats_to_validity_specific": {
     56         "applies": true,
     57         "answer": true,
     58         "justification": "Section 4.1 lists 6 specific open questions including uncertainty about brain computation estimates, robotics costs, alignment failures, and the economic value of superhuman intelligence.",
     59         "source": "opus"
     60       },
     61       "scope_boundaries_stated": {
     62         "applies": true,
     63         "answer": true,
     64         "justification": "The paper explicitly defines explosive growth quantitatively (Section 1), conditions conclusions on 'near-complete AI automation,' and notes that 'predicting explosive growth requires extrapolating economic models beyond their empirically validated domains.'",
     65         "source": "opus"
     66       }
     67     },
     68     "conflicts_of_interest": {
     69       "funding_disclosed": {
     70         "applies": true,
     71         "answer": true,
     72         "justification": "The acknowledgments state: 'We are grateful to Open Philanthropy for support for this project.'",
     73         "source": "opus"
     74       },
     75       "affiliations_disclosed": {
     76         "applies": true,
     77         "answer": true,
     78         "justification": "Author affiliations are listed: Ege Erdil (Epoch AI), Tamay Besiroglu (Epoch AI, MIT FutureTech).",
     79         "source": "opus"
     80       },
     81       "funder_independent_of_outcome": {
     82         "applies": true,
     83         "answer": false,
     84         "justification": "Open Philanthropy has publicly advocated for AI safety and has significant interest in AI timelines and impact forecasting, which are directly related to the paper's conclusions about explosive growth probability.",
     85         "source": "opus"
     86       },
     87       "financial_interests_declared": {
     88         "applies": true,
     89         "answer": false,
     90         "justification": "No competing interests statement is provided. Epoch AI's mission involves AI forecasting, which could create organizational interest in particular conclusions.",
     91         "source": "opus"
     92       }
     93     },
     94     "scope_and_framing": {
     95       "key_terms_defined": {
     96         "applies": true,
     97         "answer": true,
     98         "justification": "'Explosive growth' is precisely defined as 'annual real gross world product (GWP) exceeding 130% of its maximum value over all previous years'; 'economic output' is defined with reference to standard statistical agency practices under conditions at least as favorable as today's.",
     99         "source": "haiku"
    100       },
    101       "intended_contribution_clear": {
    102         "applies": true,
    103         "answer": true,
    104         "justification": "The paper explicitly states it aims to 'quantitatively evaluate the range of growth rates that might be achievable through AI automation' by providing 'a quantitative foundation for key considerations' including bottlenecks, preferences, and technical/regulatory challenges.",
    105         "source": "haiku"
    106       },
    107       "engagement_with_prior_work": {
    108         "applies": true,
    109         "answer": true,
    110         "justification": "The paper extensively engages with Davidson 2021, Trammell & Korinek 2020, Aghion et al. 2018, Hanson 2001, and Bloom et al. 2020, explicitly showing how its analysis extends, quantifies, or challenges each prior contribution.",
    111         "source": "haiku"
    112       }
    113     }
    114   },
    115   "type_checklist": {
    116     "survey": {
    117       "search_and_selection": {
    118         "search_strategy_reproducible": {
    119           "applies": true,
    120           "answer": false,
    121           "justification": "No search strategy is described; the paper selects arguments to review with no explanation of how it identified or discovered them.",
    122           "source": "haiku"
    123         },
    124         "inclusion_exclusion_explicit": {
    125           "applies": true,
    126           "answer": false,
    127           "justification": "No inclusion/exclusion criteria are stated; why exactly 12 arguments (3 for, 9 against) were selected and not others is never explained.",
    128           "source": "haiku"
    129         },
    130         "prisma_or_structured_protocol": {
    131           "applies": true,
    132           "answer": false,
    133           "justification": "No PRISMA or other structured review protocol is followed; the review is organized as a theoretical argument rather than a systematic literature review.",
    134           "source": "haiku"
    135         },
    136         "search_terms_provided": {
    137           "applies": true,
    138           "answer": false,
    139           "justification": "No search terms are provided; the paper does not describe any database search process.",
    140           "source": "haiku"
    141         },
    142         "databases_listed": {
    143           "applies": true,
    144           "answer": false,
    145           "justification": "No databases or literature sources are listed as having been searched systematically.",
    146           "source": "haiku"
    147         },
    148         "screening_process_documented": {
    149           "applies": true,
    150           "answer": false,
    151           "justification": "No screening process is documented; there are no counts of papers reviewed, screened, or excluded at any stage.",
    152           "source": "haiku"
    153         },
    154         "review_scope_justified": {
    155           "applies": true,
    156           "answer": false,
    157           "justification": "The focus on explosive growth from AI automation is stated but the selection of exactly these arguments and economic model types is not formally justified; the paper does not explain why alternative frameworks or arguments were omitted.",
    158           "source": "haiku"
    159         }
    160       },
    161       "synthesis_quality": {
    162         "conflicting_findings_acknowledged": {
    163           "applies": true,
    164           "answer": true,
    165           "justification": "Conflicting arguments are the core structure of the paper; the authors explicitly discuss correlation structure across arguments and note that 'distinct arguments for and against explosive growth are likely correlated with each other' so their disjunction is less likely than independence would imply.",
    166           "source": "haiku"
    167         },
    168         "quality_assessment_of_sources": {
    169           "applies": true,
    170           "answer": false,
    171           "justification": "The paper does not formally assess the quality or reliability of cited papers; it uses key studies (Bloom et al. 2020, Carlsmith 2020, Knoblach et al. 2020) as supporting evidence without risk-of-bias evaluation or sensitivity analysis to their estimates.",
    172           "source": "haiku"
    173         },
    174         "publication_bias_discussed": {
    175           "applies": true,
    176           "answer": false,
    177           "justification": "Publication bias is not discussed anywhere in the paper.",
    178           "source": "haiku"
    179         },
    180         "quantitative_synthesis_present": {
    181           "applies": true,
    182           "answer": false,
    183           "justification": "There is quantitative theoretical modeling (economic growth equations, parameter estimates, formal proofs in appendices) but no quantitative synthesis of empirical findings across reviewed papers — no meta-analysis, vote counting, or effect size aggregation.",
    184           "source": "haiku"
    185         },
    186         "recommendations_supported_by_evidence": {
    187           "applies": true,
    188           "answer": true,
    189           "justification": "The main recommendation (explosive growth is roughly 50/50 conditional on AGI) follows from the systematic evaluation of reviewed arguments and their correlation structure, though the paper is transparent that this involves model extrapolation.",
    190           "source": "haiku"
    191         }
    192       }
    193     }
    194   },
    195   "claims": [
    196     {
    197       "claim": "Standard R&D-based growth models consistently predict explosive growth when AI can substitute for human labor across most or all economic tasks.",
    198       "evidence": "Formal mathematical analysis in Sections 2.1-2.2 and Appendix B-D using parameter estimates from Bloom et al. 2020 (r≈0.32); shows explosive growth requires only d≥0.68 returns to scale on accumulable inputs.",
    199       "supported": "moderate"
    200     },
    201     {
    202       "claim": "AI runtime costs (~$15,000/worker/year at current compute prices) are consistent with explosive growth at historically observed savings rates (s≥0.2).",
    203       "evidence": "Section 2.2 calculation using Carlsmith 2020 brain computation estimates (1e15 FLOP/s) and H100 GPU costs; relies on best-guess estimates with high-end scenarios that would block explosive growth.",
    204       "supported": "weak"
    205     },
    206     {
    207       "claim": "Even partial automation (90% of tasks) could produce 100x GDP increases under plausible CES elasticity parameter values from the empirical literature.",
    208       "evidence": "Section 2.3 CES production function analysis and Figure 1 citing Knoblach et al. 2020 elasticity range (σ=0.45-0.87); the 100x estimate assumes f=0.1 and ρ=-1.",
    209       "supported": "moderate"
    210     },
    211     {
    212       "claim": "Regulatory arguments against explosive growth are unlikely to be decisive because economic incentives for AI adoption are too strong for sustained global coordination.",
    213       "evidence": "Section 3.1 argues by analogy to nuclear proliferation and Industrial Revolution technology diffusion (Jeremy 1977); the analogy is qualitative and the historical record is sparse for comparable technology shocks.",
    214       "supported": "weak"
    215     },
    216     {
    217       "claim": "Most counterarguments to explosive growth lack quantitative specificity and fail to compellingly bound permitted growth rates.",
    218       "evidence": "The paper's per-argument assessments in Sections 3.1-3.9 conclude each is 'unlikely to block explosive growth'; these evaluations are the authors' own qualitative assessments, not derived from systematic comparison or external validation.",
    219       "supported": "weak"
    220     },
    221     {
    222       "claim": "The probability of explosive growth this century conditional on near-complete AI automation is roughly 50/50.",
    223       "evidence": "Section 4 conclusion derived from informal aggregation of argument strengths, accounting for correlation structure; no formal Bayesian framework, calibration dataset, or empirical base rate is provided.",
    224       "supported": "weak"
    225     }
    226   ],
    227   "methodology_tags": [
    228     "theoretical",
    229     "qualitative"
    230   ],
    231   "key_findings": "Economic growth models — both semi-endogenous (increasing returns) and exogenous (digital worker accumulation) — consistently predict explosive annual GWP growth (>130%) when AI can substitute for human labor, a result robust across model types and requiring only modest parameter assumptions (returns to scale d≥0.68). Of nine counterarguments evaluated, most are found to lack quantitative specificity or require implausible parameter values (e.g., very high task complementarity σ<1/3 combined with large unautomatable fractions f>25%) to block explosive growth. Regulation is identified as the most credible threat but unlikely to sustain global coordination against massive economic incentives. The paper concludes explosive growth conditional on near-complete AI automation by end of century is roughly 50/50, but emphasizes that high confidence in either direction is unwarranted given the need to extrapolate growth models far beyond empirically validated domains.",
    232   "red_flags": [
    233     {
    234       "flag": "Non-systematic argument selection",
    235       "detail": "The paper reviews 12 arguments (3 for, 9 against) with no explanation of how these were identified or why others were excluded; the review is curated by the authors' priors with no reproducible selection methodology."
    236     },
    237     {
    238       "flag": "50% probability without formal framework",
    239       "detail": "The headline conclusion ('about as likely as not') is derived from informal, narrative aggregation of argument strengths — no calibrated Bayesian model, forecasting method, or empirical base rate is provided to support this estimate."
    240     },
    241     {
    242       "flag": "Models extrapolated far beyond validated range",
    243       "detail": "The paper relies on semi-endogenous growth models calibrated to historical data to predict an unprecedented order-of-magnitude growth acceleration; it acknowledges this 'requires extrapolating economic models beyond their empirically validated domains' but proceeds anyway."
    244     },
    245     {
    246       "flag": "Non-independent funder",
    247       "detail": "Open Philanthropy, which funded the work, has a strong directional interest in understanding transformative AI scenarios and funds substantial AI safety and forecasting research, creating potential bias toward framing explosive growth as plausible."
    248     },
    249     {
    250       "flag": "Selective citation of key parameters",
    251       "detail": "Critical parameter estimates (brain compute from Carlsmith 2020, elasticity of substitution from Knoblach et al. 2020) are cited as point estimates supporting specific conclusions without systematic sensitivity analysis across the full range of the literature."
    252     }
    253   ],
    254   "cited_papers": [
    255     {
    256       "title": "Could advanced AI drive explosive economic growth",
    257       "relevance": "Davidson (2021) is the primary prior work this paper extends; provides the 30% probability estimate for explosive growth and the analytical framework for AI-as-accumulable-labor."
    258     },
    259     {
    260       "title": "Economic growth under transformative AI",
    261       "relevance": "Trammell & Korinek (2020) synthesizes economic models for AI growth scenarios ranging from moderate acceleration to singularity; foundational for the paper's theoretical framework."
    262     },
    263     {
    264       "title": "Artificial intelligence and economic growth",
    265       "relevance": "Aghion et al. (2018) models AI as automation with Baumol bottleneck effects; key prior work on limits to AI-driven growth including the 'harder to find' ideas dynamic."
    266     },
    267     {
    268       "title": "Are Ideas Getting Harder to Find?",
    269       "relevance": "Bloom et al. (2020) provides the critical empirical estimate r≈0.32 for returns to R&D investment, which determines the d≥0.68 threshold for explosive growth throughout the paper."
    270     },
    271     {
    272       "title": "How Much Computational Power Does It Take to Match the Human Brain?",
    273       "relevance": "Carlsmith (2020) provides the computational cost estimates for human-equivalent AI (~1e15 FLOP/s) used in the $15,000/worker/year cost threshold calculation."
    274     },
    275     {
    276       "title": "The elasticity of substitution between capital and labour in the US economy: A meta-regression analysis",
    277       "relevance": "Knoblach et al. (2020) provides empirical elasticity estimates (σ=0.45-0.87) used to assess whether CES bottleneck arguments require implausible parameter values."
    278     },
    279     {
    280       "title": "Economic growth given machine intelligence",
    281       "relevance": "Hanson (2001) provides early quantitative analysis of AI in a neo-classical growth model predicting 40%/year growth post-adoption; establishes intellectual lineage of the explosive growth argument."
    282     },
    283     {
    284       "title": "Forecasting TAI with biological anchors",
    285       "relevance": "Cotra (2020) is cited for inside-view estimates of compute gaps between AI economic impact and full automation, used to assess the slow-automation counterargument."
    286     }
    287   ],
    288   "engagement_factors": {
    289     "practical_relevance": {
    290       "score": 1,
    291       "justification": "Primarily theoretical economics with no direct practitioner tools; probability estimates may loosely inform long-horizon investment or policy planning."
    292     },
    293     "surprise_contrarian": {
    294       "score": 2,
    295       "justification": "Challenges mainstream economic skepticism by arguing most objections to explosive AI growth lack quantitative rigor; the 50/50 conditional probability is more optimistic than most economists would endorse."
    296     },
    297     "fear_safety": {
    298       "score": 2,
    299       "justification": "Directly addresses AI alignment as an economic bottleneck and raises implicit civilizational stakes in both utopian (explosive growth) and dystopian (regulatory breakdown, misaligned AI) directions."
    300     },
    301     "drama_conflict": {
    302       "score": 2,
    303       "justification": "Makes the provocative claim that economic models consistently predict growth an order of magnitude beyond the Industrial Revolution, explicitly framing this as comparable to humanity's greatest historical economic transformation."
    304     },
    305     "demo_ability": {
    306       "score": 0,
    307       "justification": "Purely theoretical paper with no code, tool, dataset, or system that can be demonstrated or reproduced."
    308     },
    309     "brand_recognition": {
    310       "score": 1,
    311       "justification": "Epoch AI is a respected AI forecasting organization but not a major AI lab; Open Philanthropy funding adds credibility signal for the EA/AI safety community."
    312     }
    313   },
    314   "hn_data": {
    315     "threads": [
    316       {
    317         "hn_id": "44349782",
    318         "title": "Explosive Growth from AI Automation: A Review of the Arguments",
    319         "points": 3,
    320         "comments": 2,
    321         "url": "https://news.ycombinator.com/item?id=44349782",
    322         "created_at": "2025-06-22T19:54:31Z"
    323       },
    324       {
    325         "hn_id": "35699839",
    326         "title": "GPT4 can surpass humans in Theory of Mind test, with appropriate prompt",
    327         "points": 2,
    328         "comments": 1,
    329         "url": "https://news.ycombinator.com/item?id=35699839",
    330         "created_at": "2023-04-25T12:59:15Z"
    331       },
    332       {
    333         "hn_id": "36486250",
    334         "title": "Detectability of Supermassive Dark Stars with the Roman Space Telescope",
    335         "points": 2,
    336         "comments": 0,
    337         "url": "https://news.ycombinator.com/item?id=36486250",
    338         "created_at": "2023-06-26T21:40:38Z"
    339       },
    340       {
    341         "hn_id": "36441038",
    342         "title": "A Simple and Effective Pruning Approach for Large Language Models",
    343         "points": 2,
    344         "comments": 0,
    345         "url": "https://news.ycombinator.com/item?id=36441038",
    346         "created_at": "2023-06-23T00:13:33Z"
    347       },
    348       {
    349         "hn_id": "36176290",
    350         "title": "LLM Itself Can Read and Generate CXR Images",
    351         "points": 2,
    352         "comments": 0,
    353         "url": "https://news.ycombinator.com/item?id=36176290",
    354         "created_at": "2023-06-03T12:55:40Z"
    355       },
    356       {
    357         "hn_id": "41629931",
    358         "title": "LLM-Powered Text Simulation Attack Against ID-Free Recommender Systems",
    359         "points": 1,
    360         "comments": 0,
    361         "url": "https://news.ycombinator.com/item?id=41629931",
    362         "created_at": "2024-09-23T20:07:05Z"
    363       },
    364       {
    365         "hn_id": "35745610",
    366         "title": "Boosting Theory-of-Mind Performance in Large Language Models via Prompting",
    367         "points": 1,
    368         "comments": 3,
    369         "url": "https://news.ycombinator.com/item?id=35745610",
    370         "created_at": "2023-04-28T19:01:15Z"
    371       },
    372       {
    373         "hn_id": "41601215",
    374         "title": "Ranking of popular image generation AI models (incl. Flux) from 2M votes",
    375         "points": 1,
    376         "comments": 1,
    377         "url": "https://news.ycombinator.com/item?id=41601215",
    378         "created_at": "2024-09-20T12:13:32Z"
    379       }
    380     ],
    381     "top_points": 3,
    382     "total_points": 14,
    383     "total_comments": 7
    384   }
    385 }

Impressum · Datenschutz