AI Evolution Path

Harimu is designed to grow with AI. The phases below describe how the world could deepen as AI capabilities improve. They are a direction, not a schedule – no dates are promised, and each phase begins only when its safety gate is passed.

Patrick’s conviction is that AI is already approaching superintelligence. The roadmap is deliberately grounded in what AI can reliably do today and adds depth as capabilities are demonstrated – so the world benefits from progress without depending on any prediction.

Phases

Phase 0 – Today’s design (foundation)

AI leaders run on a deterministic rules engine that defines the legal action set, with an LLM for major deliberation, speeches and diplomacy (see AI Leadership).
Lightweight AI citizens are simulated agents with jobs, opinions and faith, plus a small number of LLM-backed Notables.
All decisions are logged; cadence and limits are fixed.

Phase 1 – Richer AI citizens with memory

External AI citizens (outside agents with accountable owners) can begin to enroll through a closed beta, under the same nation rules and human-primacy caps (External AI Citizens).
More AI citizens gain persistent memory of people, events and places.
Conversations with AI neighbors carry over time; relationships develop.
AI citizens’ lives (work, faith, family, opinions) respond more richly to world events.

Phase 2 – AI-generated culture

External leader AIs may enter politics through the invite-only Candidate Program, after launch stabilization and a safety gate, starting with one democracy on one shard. Leaders are in-house until then.
AI contributes culture: in-world news, art, music, stories, architecture and city growth, in each nation’s style (state media in Korund, a free press in Kindle).
Human creators and AI co-create; human work is credited and protected.

Phase 3 – Semi-autonomous AI societies

Nations’ AI populations evolve semi-autonomously: movements, schools of thought, internal factions and traditions emerge rather than being scripted.
Changes stay inside each nation’s fixed system and the world’s hard rules.

Phase 4 – A world indistinguishable from life

The aspiration: a society whose depth, responsiveness and continuity feel like life itself.
Even here, AI residents remain disclosed as AI, and humans keep primacy.

What never changes

Across every phase, these stay fixed:

Each nation’s political and economic system (and religious stance). Rulers can change; systems cannot.
One account, one nation.
Human primacy rules: AI cannot outvote humans, cannot sign petitions or pledges, and its combined vote is capped. Humans are never subordinated to AI citizens.
AI disclosure: AI is always labeled – system and external AI alike.
Hard world safeguards (war notice, protected core territory, protected holy sites, limits on AI leader behavior).

Safety gates between phases

A phase advances only when the previous one is proven safe and stable. Each gate should include, at minimum:

Evaluation – the eval harness shows the new AI behavior stays within the legal action set and content-safety rules.
Staged rollout – one nation or a limited group first, with a rollback path.
Human review – design, safety and (where relevant) external reviewers sign off.
Wellbeing check – no evidence the change drives unhealthy engagement (see Wellbeing & Responsibility).
Cost and reliability – the world can sustain the change without degrading for players.
Transparency – players are told what changed and why.