{ "id": "S15", "slug": "superintelligence-takeoff", "title": "Superintelligence takeoff", "version": "0.3.0", "status": "active", "orientation": "upside", "family": "D", "short_title": "Superintelligence takeoff", "thematics": [ "ai-cyber-digital", "economic-security-supply-chains", "demographics-human-capital" ], "likelihood_type": "systemic", "likelihood_note": null, "scores": { "likelihood_2y": { "band": 3, "rationale": "Frontier laboratories have set dates for automated research: an 'intern' in 2026 and a full AI researcher by 2028. Measured task horizons have been doubling every few months. Yet as of April 2026 no laboratory reported a doubling of overall research progress, and sceptics put the feedback loop well short of self-sustaining." }, "likelihood_10y": { "band": 4, "rationale": "Published forecasts cluster the start of an intelligence explosion between 2027 and the early 2030s; expert surveys put it far later. Band 4 splits the difference." }, "impact_systemic": { "level": 5, "rationale": "Level 5 on irreversible structural change rather than loss: output may rise sharply while the distribution of power and income is rewritten." }, "impact_national": { "level": 5, "rationale": "For a nation with no frontier capability, the tax base, labour market and strategic position all shift at once." }, "confidence": { "level": "very low", "reason": "The central mechanism has never been observed and informed forecasters disagree by decades." }, "priority": 15, "tier": "I", "source": "ginc-desk-v0.3" }, "scorecard": { "onset": "rapid (months)", "duration": "3 years", "warning": "months", "scope": "global", "origin": "accidental (technological)", "external_support": "none", "policy_response": "current plans", "recovery_horizon": "structural" }, "summary": "AI systems automate AI research itself. Each generation designs a better successor and, within about two years of the loop closing, systems outperform the best human experts in every field of science and engineering. A century of research progress is compressed into a decade or less: new drugs, materials, energy designs and weapons arrive faster than institutions can absorb them. Output grows at rates with no precedent. The gains go first to whoever owns the models, compute and energy. Many researchers class this as the largest risk of all; the Library carries it as an upside scenario because the central case is abundance, with control, concentration and distribution as the open problems.", "distribution": { "gains": [ "Nations hosting frontier projects, compute and power", "Owners of models and chips", "Patients, as medical research accelerates", "Any state that adopts faster than its peers" ], "loses": [ "Knowledge-work exporters and graduate labour markets", "States that tax payroll more than capital", "Militaries without frontier systems", "Nations with nothing the model owners need" ] }, "narrative": { "dateline": "March 2030", "text": "The announcement is three paragraphs long. The laboratory's research agents have been improving their own training methods for eleven months, and the newest system has not been evaluated by any human who fully understands its results. Two rivals say the same within a quarter. By then the evidence is in the journals: a room-temperature catalyst, a malaria vaccine designed and trialled in simulation before a single dose is made, a reactor design that passes review in weeks. A year of this produces more patents than the previous twenty. Growth in the two economies that host the data centres runs at 9 per cent and their electricity demand at twice that. Elsewhere the picture is stranger. Finance ministries watch corporate profits migrate to three jurisdictions. A mid-sized country's entire professional-services export sector is repriced to the cost of inference. Defence staffs brief that the side with the better models now designs, tests and fields a weapon in months, and that nobody can verify what the other side has. Governments that asked for a seat at the table are offered an API key. The leaders' summit agrees a notification protocol and a compute registry, and neither can be inspected by the people who signed it. Hospitals are the first to feel unambiguous benefit, and the first to ask who is liable. The countries that do well are not the ones with the most engineers. They are the ones with power to spare, a state that can adopt as fast as its firms, and something the owners of the models need." }, "anchors": [ { "shock_id": "i-j-good-speculations-concerning-the-first-ultraintelligent-machine-1965", "date": "1965", "event": "I. J. Good, 'Speculations Concerning the First Ultraintelligent Machine'", "what_happened": "First statement of the intelligence explosion: a machine that can design better machines leaves human intelligence far behind", "calibrates": "the core mechanism" }, { "shock_id": "bostrom-superintelligence-2014", "date": "2014", "event": "Bostrom, Superintelligence", "what_happened": "Set out slow, moderate and fast takeoff and the idea of a decisive strategic advantage for the first project to cross the threshold", "calibrates": "takeoff speed as the key parameter" }, { "shock_id": "aschenbrenner-situational-awareness-2024", "date": "June 2024", "event": "Aschenbrenner, Situational Awareness", "what_happened": "Argued that automated AI research could compress a decade of algorithmic progress into about a year, and framed the race as a national security project", "calibrates": "speed; state involvement" }, { "shock_id": "amodei-machines-of-loving-grace-2024", "date": "October 2024", "event": "Amodei, Machines of Loving Grace", "what_happened": "Described 'a country of geniuses in a datacenter' compressing 50 to 100 years of biological progress into five to ten", "calibrates": "the upside case" }, { "shock_id": "forethought-preparing-for-the-intelligence-explosion-2025", "date": "March 2025", "event": "Forethought, Preparing for the Intelligence Explosion", "what_happened": "MacAskill and Moorhouse argued that AI could drive a century of technological progress in a decade, raising many grand challenges at once", "calibrates": "breadth of consequences" }, { "shock_id": "ai-2027-2025", "date": "April 2025", "event": "AI 2027", "what_happened": "Kokotajlo and co-authors published a month-by-month scenario in which a superhuman coder leads to superintelligence within about a year, with race and slowdown endings", "calibrates": "a concrete takeoff path" }, { "shock_id": "metr-time-horizon-measurements-2025", "date": "March 2025 to April 2026", "event": "METR time-horizon measurements", "what_happened": "The length of tasks AI agents complete doubled about every seven months, then faster; in April 2026 laboratories reported no dramatic speed-up of their own research", "calibrates": "leading indicator and counter-evidence" }, { "shock_id": "narayanan-and-kapoor-ai-as-normal-technology-2025", "date": "April 2025", "event": "Narayanan and Kapoor, AI as Normal Technology", "what_happened": "Argued that diffusion is limited by institutions and that transformation will take decades", "calibrates": "counter-anchor on speed" } ], "parameters": { "common": { "severity": { "major": "major", "severe": "severe", "extreme": "extreme" }, "duration": { "major": null, "severe": "3 years", "extreme": null }, "onset": { "major": null, "severe": "rapid (months)", "extreme": null }, "warning": { "major": null, "severe": "months", "extreme": null }, "scope": { "major": null, "severe": "global", "extreme": null }, "origin": { "major": null, "severe": "accidental (technological)", "extreme": null }, "external_support": { "major": null, "severe": "none", "extreme": null }, "concurrency": { "major": null, "severe": "standalone", "extreme": null }, "policy_response": { "major": null, "severe": "current plans", "extreme": null }, "recovery_horizon": { "major": null, "severe": "structural", "extreme": null } }, "specific": [ { "name": "Research speed-up at peak", "default": "×10", "range": "×3 to ×100", "unit": null, "note": null }, { "name": "Time from automated researcher to superintelligence", "default": "12 months", "range": "3–60", "unit": "months", "note": null }, { "name": "Leading projects", "default": "3", "range": "1–10", "unit": null, "note": null }, { "name": "Control of the leading systems", "default": "adequate", "range": "options: contested / lost", "unit": null, "note": "lost at Extreme" }, { "name": "Access for nations that host no project", "default": "paid API", "range": "options: open weights / restricted", "unit": null, "note": null }, { "name": "Binding constraint", "default": "compute and power", "range": "options: data / regulation / none", "unit": null, "note": null } ] }, "transmission": [ "AI systems automate AI research; each generation shortens the next.", "Research output in science and engineering multiplies; discoveries outrun regulators and standards.", "Output and profits concentrate in the firms and jurisdictions that own models, compute and power.", "Knowledge work is repriced to the cost of inference; tax bases built on employment erode.", "Military and intelligence advantage shifts to the leading projects; verification fails.", "States without capability bargain for access with energy, data, minerals or alignment.", "Abundance in health, energy and materials arrives unevenly, by ability to adopt.", "Governance lags: liability, control and distribution are settled after the fact." ], "loading": [ { "dimension": "Hard", "domain": "defence-security", "load": "High", "channel": "decisive advantage for the leading projects; verification gap" }, { "dimension": "Hard", "domain": "strategic-infrastructure", "load": "High", "channel": "data centres, power and grid become the binding constraint" }, { "dimension": "Hard", "domain": "critical-technology", "load": "High", "channel": "frontier models, chips and compute decide who hosts the takeoff" }, { "dimension": "Soft", "domain": "government-effectiveness", "load": "High", "channel": "state adoption speed, liability and control rules, tax redesign" }, { "dimension": "Soft", "domain": "human-capital", "load": "High", "channel": "knowledge work repriced; education and retraining rebuilt" }, { "dimension": "Soft", "domain": "influence-cohesion", "load": "High", "channel": "distributional conflict; influence over standards and treaties" }, { "dimension": "Economic", "domain": "macro-financial", "load": "High", "channel": "growth surge, profit migration, tax base and asset repricing" }, { "dimension": "Economic", "domain": "industry-trade-supply", "load": "High", "channel": "services exports repriced; research-led industries leap" }, { "dimension": "Economic", "domain": "energy-resources", "load": "High", "channel": "electricity demand surge; minerals for compute" } ], "stakeholders": { "government": { "exposure": "Strategic position, the tax base and the state's own capacity to adopt", "actions": [ "Secure access to frontier models on terms that survive a crisis", "Build power and compute ahead of demand", "Redesign taxation to follow value, not payroll", "Join verification and notification regimes early" ], "watch": [ "Laboratory statements on automated research", "Frontier compute and power build-out", "Export controls on chips and model weights" ], "relevance": 5 }, "technology": { "exposure": "The shock is the sector: firms either ride the leading models or are repriced by them", "actions": [ "Hold more than one frontier provider", "Invest in evaluation and control of autonomous systems", "Secure power contracts" ], "watch": [ "Task-horizon and research-automation benchmarks", "Model pricing and access terms" ], "relevance": 5 }, "investors": { "exposure": "Extreme concentration of returns; repricing of labour-intensive services and of long-dated assumptions", "actions": [ "Test portfolios on a growth surge with profit concentrated in a few firms", "Map holdings by exposure to inference-priced work" ], "watch": [ "Hyperscaler capital expenditure", "Share of research output attributed to AI systems", "Electricity demand in host regions" ], "relevance": 5 }, "public": { "exposure": "Jobs and incomes in knowledge work; earlier access to medical advances", "actions": [ "Build skills in directing and checking AI work" ], "watch": [ "Entry-level hiring", "Public service adoption" ], "relevance": 5 } }, "regions": [ { "region": "north-america", "exposure": "High", "rationale": "Host of most leading projects; concentration of gains and of risk" }, { "region": "europe", "exposure": "High", "rationale": "Dependence without a frontier project; regulation as its main lever" }, { "region": "china", "exposure": "High", "rationale": "The other host; domestic chips and power decide its pace" }, { "region": "indo-pacific", "exposure": "High", "rationale": "Chip manufacturing leverage in Taiwan, Korea and Japan; services exporters repriced" }, { "region": "south-asia", "exposure": "High", "rationale": "IT services and outsourcing repriced; large gains from cheap expertise" }, { "region": "gulf-middle-east", "exposure": "Medium", "rationale": "Energy and capital to host compute; a seat bought, not built" }, { "region": "africa", "exposure": "Medium", "rationale": "Largest gains from health and agricultural science; least bargaining power" }, { "region": "latin-america", "exposure": "Medium", "rationale": "Minerals and power as bargaining chips; services exposed" }, { "region": "russia-eurasia", "exposure": "Medium", "rationale": "Cut off from frontier chips; military implications dominate" } ], "indicators": [ { "name": "Length of tasks AI agents complete autonomously", "source": "METR", "threshold": null, "cadence": null }, { "name": "Share of frontier laboratory research and code produced by AI systems", "source": "company disclosures", "threshold": null, "cadence": null }, { "name": "Frontier compute and data-centre power under construction", "source": "Epoch AI, IEA", "threshold": null, "cadence": null }, { "name": "Price of inference at fixed capability", "source": null, "threshold": null, "cadence": null }, { "name": "Forecasts for automated AI research", "source": "Metaculus", "threshold": null, "cadence": null }, { "name": "Export controls on chips and model weights", "source": "BIS", "threshold": null, "cadence": null }, { "name": "AI safety notification and verification agreements", "source": null, "threshold": null, "cadence": null } ], "compounds": { "triggers": [ { "id": "S13", "note": "displacement at speed" }, { "id": "S03", "note": "asset repricing" }, { "id": "S07", "note": "distributional conflict" }, { "id": "S16", "note": "energy research accelerates" }, { "id": "S17", "note": "biomedical research accelerates" } ], "triggered_by": [], "amplified_by": [], "triggered_by_other": [], "amplifying_trends": [ "compute scaling", "algorithmic progress", "capital concentration in AI", "great-power competition" ] }, "open_questions": [ "Whether a scenario many researchers rank as the gravest risk belongs on the upside list, or needs a paired loss-of-control scenario.", "Whether takeoff is one event or a decade of fast diffusion, in which case it is S13 at higher intensity.", "How to rate impact when output rises but most nations' share of it falls." ], "commentary": null, "overview": { "source": "ginc-desk-v0.3", "lenses": { "political": "Governments that asked for a seat at the table are **offered an API key**. A summit agrees a compute registry nobody can inspect.", "economic": "Growth in the host economies runs at **9 per cent**. Profits migrate to three jurisdictions and services exports are repriced to the cost of inference.", "social": "Hospitals feel the benefit first. **Knowledge work is repriced** and the question becomes who shares in the gain.", "technological": "Research agents improve their own training for eleven months. **A year produces more patents than the previous twenty**.", "legal": "**Liability and control** are settled after the fact; no human fully understands the newest system's results.", "environmental": "Host-economy **electricity demand** grows at twice the rate of output; new reactor designs pass review in weeks." }, "regions": { "north-america": "**Host** of most leading projects: the gains, the power demand and the control problem all land here first.", "europe": "**Dependence** without a frontier project. Regulation is the main lever and adoption speed the main test.", "china": "**The other host**. Domestic chips and power decide whether it keeps pace.", "indo-pacific": "**Chip manufacturing** gives Taiwan, Korea and Japan leverage; services exporters are repriced.", "south-asia": "IT services and outsourcing are **repriced to the cost of inference**, while cheap expertise reaches hundreds of millions.", "gulf-middle-east": "Energy and capital **buy a seat**: compute is hosted where power is cheap.", "africa": "The largest gains from health and agricultural science, with the **least bargaining power** over access.", "latin-america": "**Minerals and power** are bargaining chips; services employment is exposed.", "russia-eurasia": "Cut off from frontier chips; the **military implications** dominate." } }, "history": [ { "edition": 2027, "rank": null, "L2y": 3, "L10y": 4, "impact_systemic": 5, "impact_national": 5, "confidence": "very low" } ], "changelog": [ { "version": "0.3.0", "date": "2026-10-03", "change": "Entered the Library at v0.3 as an upside scenario, with GINC desk scores." } ], "citation": "GINC (2027). Scenario S15 Superintelligence takeoff, Scenario Library v0.3. scenarios.ginc.org/library/superintelligence-takeoff" }