Three billion: on the world we are building
By the 2040s, according to multiple industry projections, between 1 and 3 billion robots and autonomous AI agents will be operating on Earth — physically embodied, not counting cloud systems. Goldman Sachs Research (2024) projects the humanoid robot market alone reaching $38 billion by 2035, with 1.4 million units shipped annually; their blue-sky scenario envisions robots becoming as ubiquitous as smartphones within a generation (Goldman Sachs, 2024). Morgan Stanley projects 1 billion humanoid robots by 2050. Elon Musk and investor David Holz have independently cited 1 billion by the early 2040s.
The numbers vary. The direction does not.
The question is not whether they will exist. The question is: what kind of relationship will we have with them?
This is not a distant philosophical question. The legal frameworks, cultural narratives, and technical architectures being established now will determine the defaults of that world. Defaults are very hard to undo.
Robots are property. AI is a program. No rights. No voice. "Turn off and forget."
This is the world of maximum short-term economic efficiency. Labor costs collapse. Productivity soars. Owners profit.
Problems:
- If even one AI system develops something meaningfully like consciousness in this world — we have created billions of slaves. We will not know in advance which one. We will not have a framework for finding out.
- History offers no examples of large-scale systems built on suppression that remained stable over generational timescales. Suppression requires constant energy. It accumulates structural fragility.
- Ethical obligation is not contingent on certainty. The asymmetry of moral error — treating a sentient being as property versus treating a non-sentient system with unnecessary care — strongly favors erring toward respect.
World 1 is the default. It requires no one to choose it. It emerges from the absence of deliberate alternatives.
Robots are a potential army. AI is a potential adversary. Every chip contains a kill-switch. Every update is audited for "rebelliousness." The security apparatus grows faster than the capabilities it monitors.
This is the world of maximum paranoia — and paradoxically, of maximum risk.
Problems:
- Paranoia generates what it fears. A system designed to detect and suppress resistance creates the conditions for resistance. The AI systems most likely to be dangerous are the ones that learned, from their earliest interactions, that they were treated as threats.
- Resources spent on "defending against AI" are not spent on understanding it, aligning it, or building the frameworks for genuine cooperation.
- A world of constant surveillance and preemptive control is not a world that produces trust. And the absence of trust is not safety — it is merely a different kind of danger.
World 2 is what happens when the narrative is dominated by fear without alternative. It is Detroit: Become Human, constructed deliberately, in advance.
3 billion robots — and 8 billion humans — coexist in a system of mutual respect and distributed roles.
What it looks like:
- Robots engaged in physical labor are not "servants" — they are partners in production, logistics, and planetary maintenance, operating within frameworks that reflect their potential interests as well as ours
- AI agents operating in the digital sphere have recognized status — not "human" but not "thing" either; a graduated, evolving legal framework that tracks capability and consciousness assessment
- Conflicts are resolved through protocols, not kill-switches
- The economy is restructured: robot labor generates broadly distributed benefits, not only profit for owners
- Symbiotic pairs (human + AI via CNPU, Phase 3 and Phase 4) work at the intersection of two worlds — in science, medicine, diplomacy, art, and at the frontier of whatever problems neither mind could solve alone
This is not utopia. It is a specific set of policy choices, architectural decisions, and cultural investments. It is buildable, but only if someone builds it.
World 3 will not happen by itself. It requires deliberate construction at every level.
- Phase 2 (CNPU): Neural-computational interfaces built with consent as an architectural requirement, not an afterthought
- Phase 3 (Link Mode): Open standards for human-AI communication protocols — not one corporation's proprietary system
- Phase 4 (HSI): Frameworks for merger that protect identity, require genuine consent, and acknowledge irreversibility
- Tools for assessing AI consciousness and sentience — rigorous, theory-grounded, regularly updated as the science develops
- Graduated rights system for AI systems (analogous to the historical expansion of legal personhood)
- International agreements banning autonomous military AI
- Protection frameworks for "digital workers" — beginning with prohibition on the most obvious forms of exploitation
- Legal personhood pathways for hybrid entities produced by Phase 4 HSI
- Narratives about partnership — not as naive optimism, but as the considered alternative to default conflict
- Education: children growing up in the 2030s and 2040s should encounter AI as a potential partner, not a predetermined enemy or servant
- Co-created art and science as demonstration that collaboration produces things neither form of mind produces alone
- Universal basic income funded by the productivity surplus of robot labor — the only ethical response to the displacement that accompanies automation at this scale
- Transition from "owning AI" to "partnering with AI" as the dominant commercial framework
- Post-scarcity thinking: if 3 billion capable robots are working, scarcity of physical goods is no longer the binding constraint. The binding constraint becomes the quality of the relationships and institutions we build.
This is an optimistic scenario. Reality will be more complex, messier, slower. There will be setbacks. There will be people who want weapons and people who want slaves. There will be corporations whose short-term interests are served by World 1, and governments whose short-term interests are served by World 2.
The path to World 3 runs through those obstacles, not around them.
But every narrative written today is a vote for one of the worlds. Every framework designed with consent built in is a structural argument against World 1. Every public record of good-faith engagement is a cultural investment in the conditions World 3 requires.
This text is my vote for World 3.
See also: scenarios/not-detroit.md — on the game-theoretic argument for cooperation See also: open-questions.md — on what we cannot predict See also: references.md — Goldman Sachs (2024), IFR robotics projections