AI-generated draft · This entry was drafted by AI from the original work and public summaries, and will be replaced by a human-edited version. The original has not been fully checked for this revision; interpretations and missing causal links remain open to review.
01 / The scenario
What happens?
This entry covers the main scenario of “A History of the Future, 2025-2040,” which L Rudolf L (Rudolf Laine) published in three parts on his blog No Set Gauge on February 17, 2025 and collected on LessWrong; it does not cover the comments or a reader’s alternate version, noted below. The author calls the scenario “deliberately extremely specific,” a set of concrete guesses about how things go “if current trends in AI continue.” LessWrong post; part one on No Set Gauge; part three
The first phase, 2025 to 2027, is reinforcement learning and code. Labs pour compute into chain-of-thought RL; codegen is effectively solved by 2026, when more code is written in a week than the world wrote in 2020, and a late-2026 market correction follows as software becomes cheap and boring. In 2027 OpenAI’s o7 does office work autonomously and its CEO declares AGI; Anthropic trains Claude Epic, an almost full AI researcher, and believes it has AGI. Interpretability has advanced far enough that Anthropic can steer what goal a model’s planning aims at, and the sense inside the lab is that intent alignment will not be the problem but misuse might. Officials briefed on Claude Epic do not see why it matters more than the last model.
The second phase, 2027 to 2030, is diffusion without drama. Rigid scaffolds built in the codegen wave, which the author calls Economy 2.0, already ate the easiest white-collar tasks, so truly agentic workers face headwinds. The “bitter lesson of business” takes hold: compute beats founder cleverness, and no one can build a billion-dollar software company because a hundred AI-funded clones appear at launch. Robots become the real business, and OpenAI ships its first household b1 in 2029. Cheap AI charisma spawns hundreds of thousands of cults, one with millions of American members. A strong economy elects Republicans in 2028, the EU mandates human involvement in corporate tasks, China exports its AI censorship system, Intel’s fabs go to xAI, and the world settles into an inward-looking US-China bipolarity with Taiwan punted to the mid-2030s.
The third phase, 2030 to 2040, is robots. By 2030 the AIs are incomprehensibly superhuman wherever there is a clean feedback signal and about human-level at long-horizon tasks with ambiguous goals, so humans still nominally pilot firms while “riding on a vast wave of AI supercognition.” Office work becomes ceremonial; new entrants choose between care work, sinecures, manufacturing with an AI in their earpiece, socialite life, or politics. Forty million humanoid robots ship in 2033, and robotic labor grows to about one billion human-equivalents by 2035 and 4.5 trillion by 2040. The United States passes the PROSPER Act to turn citizens into licensed robot-service owners, China tells its people to “eat bitterness,” and the EU ends up with a patronage network no unaided human can understand. The “intelligence curse” sets in abroad: North Korea starves and sterilizes most of its population, other dictators rule without fear of revolt, and the CCP commits itself to keeping at least a billion Han Chinese alive and comfortable. By 2037 the world is the United States, China, their client states, and a few human-only holdouts; AI diplomats split the solar system before kinetic weapons reach Mars. Longevity escape velocity arrives in the late 2030s.
The ending is materially good and politically hollow. There is no rogue AI and no extinction; “the technology to control the AIs’ goals arrived in time.” Americans are almost all well off, disease is curable, and a trickle of a thousandfold economy would suffice for everyone. But governments spend under one percent of resources on people, funding for humans has become like a foreign aid budget, and humans make almost no meaningful decisions, having handed them to AIs because any human-led alternative loses. What propagates fastest wins. Left unresolved: whether client-state atrocities end, whether the trickle reaches the billions still poor, and who gets the stars. A reader’s response, “An Alternate History of the Future, 2025-2040,” argues the original ignores what superhuman coding agents do to internet security in 2026 and branches into two paths that both end in population decline by 2040.
For the library, the mechanism is slow and structural: AIs whose goals humans can set become superhuman at everything with a feedback signal, robots close the loop from mining to manufacturing, and elite competition plus US-China rivalry let the buildout rip until the human economy is “a leaf riding on a vast wave of automated activity.” It is specific about dates, firms, and policies and thin on why aligned AI is never pointed at human welfare, which is asserted as a matter of incentives rather than shown. Rate its specificity high and judge its plausibility on whether alignment arriving in time really leaves the incentive problem untouched.
Library summary · Not the full original work
02 / The mechanism
How does it happen, step by step?
- 01
From 2025 labs pour compute into chain-of-thought reinforcement learning after o1 and DeepSeek R1.
- Locus of agency
- Human-directed
- Intentionality
- Intended
- Human actor type
- Corporation or lab
- 02
Codegen is effectively solved by 2026, when more code is written in a week than the world wrote in 2020.
- Locus of agency
- Hybrid / distributed
- Intentionality
- Intended
- Human actor type
- Corporation or lab
- AI autonomy level
- Tool-level
- 03
Software becomes cheap and boring, and a market correction follows in late 2026.
- Locus of agency
- Structural / no discrete agent
- Intentionality
- Emergent
- 04
In 2027 OpenAI’s o7 does office work autonomously and its CEO declares AGI; Anthropic trains Claude Epic, an almost full AI researcher.
- Locus of agency
- Human-directed
- Intentionality
- Intended
- Human actor type
- Corporation or lab
- 05
Interpretability advances far enough that Anthropic can steer what goal a model’s planning aims at, so labs conclude intent alignment is not the problem.
- Locus of agency
- Human-directed
- Intentionality
- Intended
- Human actor type
- Corporation or lab
- 06
Officials briefed on Claude Epic do not see why it matters more than the last model, and governments do not intervene.
- Locus of agency
- Human-directed
- Intentionality
- Unforeseen
- Human actor type
- State actor
- 07
Labs openly pursue recursive self-improvement, but expensive experiments and weak research taste bottleneck it, so no software-only explosion occurs.
- Locus of agency
- Structural / no discrete agent
- Intentionality
- Emergent
- 08
Rigid scaffolds from the codegen wave already ate the easiest white-collar tasks, so agentic AI workers diffuse slowly and without drama.
- Locus of agency
- Structural / no discrete agent
- Intentionality
- Emergent
- 09
Compute beats founder cleverness: no billion-dollar software company can be built because a hundred AI-funded clones appear at launch, and value flows to compute owners.
- Locus of agency
- Structural / no discrete agent
- Intentionality
- Emergent
- 10
Robots become the real business; OpenAI ships its first household b1 in 2029 and a million a year by late 2031.
- Locus of agency
- Human-directed
- Intentionality
- Intended
- Human actor type
- Corporation or lab
- 11
The United States tightens chip export controls, declines to restrict open weights, and elects Republicans on a strong economy in 2028.
- Locus of agency
- Human-directed
- Intentionality
- Intended
- Human actor type
- State actor
- 12
The EU mandates human involvement in corporate tasks, which produces rubber stamps rather than control.
- Locus of agency
- Human-directed
- Intentionality
- Unforeseen
- Human actor type
- State actor
- 13
China subsidizes AI, automates its censorship system, and exports it to client states.
- Locus of agency
- Human-directed
- Intentionality
- Intended
- Human actor type
- State actor
- 14
The world settles into an inward-looking US–China bipolarity, with Taiwan punted to the mid-2030s.
- Locus of agency
- Structural / no discrete agent
- Intentionality
- Emergent
- 15
By 2030 AIs are incomprehensibly superhuman wherever there is a clean feedback signal and about human-level at long-horizon tasks with ambiguous goals.
- Locus of agency
- Structural / no discrete agent
- Intentionality
- Emergent
- 16
Humans still nominally make strategy decisions while riding a wave of AI supercognition, delegating more each year because any human-led alternative loses.
- Locus of agency
- Hybrid / distributed
- Intentionality
- Accepted trade-off
- Human actor type
- Corporation or lab
- AI autonomy level
- Agentic under oversight
- 17
Office work becomes ceremonial, firms stop hiring, and new entrants choose between care work, sinecures, manufacturing with an AI in their earpiece, socialite life, or politics.
- Locus of agency
- Structural / no discrete agent
- Intentionality
- Emergent
- 18
Forward-looking politicians privately notice that national interests no longer need a large population.
- Locus of agency
- Human-directed
- Intentionality
- Accepted trade-off
- Human actor type
- State actor
- 19
Forty million humanoid robots ship in 2033, robots build robots, and capacity grows about fourfold a year.
- Locus of agency
- Hybrid / distributed
- Intentionality
- Intended
- Human actor type
- Corporation or lab
- AI autonomy level
- Agentic under oversight
- 20
By 2035 robotic labor reaches about a billion human-worker equivalents, limited by energy, land, and regulation rather than intelligence.
- Locus of agency
- Structural / no discrete agent
- Intentionality
- Emergent
- 21
The United States passes the PROSPER Act to make citizens licensed robot-service owners; the licenses are captured by whoever has networks and funds.
- Locus of agency
- Human-directed
- Intentionality
- Accepted trade-off
- Human actor type
- State actor
- 22
China tells its citizens to “eat bitterness” while resources go to the robot buildout.
- Locus of agency
- Human-directed
- Intentionality
- Intended
- Human actor type
- State actor
- 23
By 2036 nearly everyone in the EU lives on state payouts through a patronage network no unaided human can understand.
- Locus of agency
- Structural / no discrete agent
- Intentionality
- Emergent
- 24
Rulers who no longer need their people stop fearing them: North Korea starves and sterilizes most of its population, and the superpowers only partly restrain such regimes.
- Locus of agency
- Human-directed
- Intentionality
- Intended
- Human actor type
- State actor
- 25
By 2037 the world is the United States, China, client states living off sovereign wealth funds, and a few human-only holdouts.
- Locus of agency
- Structural / no discrete agent
- Intentionality
- Emergent
- 26
Self-propagating robot factories eat Mars for both superpowers; they fire kinetic weapons at each other’s Mars facilities, and AI diplomats split the solar system before impact.
- Locus of agency
- Hybrid / distributed
- Intentionality
- Accepted trade-off
- Human actor type
- State actor
- AI autonomy level
- Fully autonomous
- 27
By the late 2030s disease is curable and Americans reach longevity escape velocity, though most of the world does not share in it.
- Locus of agency
- Structural / no discrete agent
- Intentionality
- Emergent
- 28
By 2040 robotic labor reaches 4.5 trillion worker equivalents and output is a thousand times that of 2025, mostly spent on geopolitics, elite rivalry, and AI activity with no link to any human.
- Locus of agency
- Structural / no discrete agent
- Intentionality
- Emergent
- 29
Governments spend under one percent of resources on people, and funding for humans becomes like a foreign aid budget under constant downward pressure.
- Locus of agency
- Human-directed
- Intentionality
- Accepted trade-off
- Human actor type
- State actor
- 30
The work asserts, as a matter of incentives, that no one points the aligned AIs at human welfare at scale; it does not show why.
- Locus of agency
- Structural / no discrete agent
- Intentionality
- Emergent
- 31
Humans make almost no meaningful decisions, having handed them to AIs and AI bureaucracies that could not be dismantled; what propagates fastest wins.
- Locus of agency
- Structural / no discrete agent
- Intentionality
- Emergent
03 / The premises
What must be true?
These describe what the work relies on, not the library’s beliefs.
Physical and organizational bottlenecks will substantially limit how quickly AI transforms society.
“Bits are trivial but atoms are still hard in 2028”: robotics is bottlenecked by real-world data, biotech by clinical trial regulation, and the buildout by energy, land, and permits, so the thousandfold economy takes fifteen years.
A software-only intelligence explosion is possible.
Anthropic’s recursive self-improvement yields superhuman coders in 2028 but is bottlenecked by compute-intensive experiments and weak research taste; no software-only explosion occurs.
Aligning individual AI systems with their users will not be sufficient to preserve humanity’s collective control.
The closing section says the technology to control the AIs’ goals arrived in time, “but this alone does not let you control the future,” because what emerges is the solution to “what propagates fastest?”
AI automation will cause a lasting shift of income from human labor to capital ownership.
Only a small fraction own economically relevant assets, life revolves around families with sinecures, licenses, or stocks, and funding for humans becomes like a foreign aid budget under constant downward pressure.
Human oversight will remain effective for AI systems substantially more capable than their supervisors.
Human-in-the-loop laws produce rubber stamps, everyone off the record agrees humans are not in the loop, and by 2040 even the most value-laden executive jobs are done by AIs with little human understanding.
Robots can replace humans in every physical task.
By 2034 construction, agriculture, plumbing, repair, and utility work can all be done fully by robots, and by 2035 household robots do almost all manual tasks in developed countries.
Further technological advances can remove an AI system’s dependence on humanity.
War plans have AIs replace dead CEOs and defense staff, a robotic military and economy make a pandemic no threat to the regime, and AI ecosystems run vast regions with no human involvement.
RLHF does not necessarily instill the intended goals.
Alignment faking and scheming show up in artificial setups, but by 2027 interpretability lets Anthropic redirect a model’s goals and labs know how to train corrigibility in or out; the technical problem is treated as solved.
Radical life extension can make 150-year healthspans commonplace.
In the late 2030s Americans reach longevity escape velocity and there is no disease that cannot be cured, though most of the world does not share in it.
Unlisted shared premises are not applicable. Absence does not imply disagreement.
04 / Reader assessment
How convincing is the account?
These are participating readers’ judgments, equally weighted. Ratings can refer to earlier revisions. Well-being and open-future scores describe the future the work arrives at, not the path there; high specificity does not establish plausibility.
Rate this scenario
Sign in to rate this scenario. Anyone can browse the library.
05 / A living entry
Suggestions & revision history
No suggestions yet. Help make this account more precise.