06 — Escapes, Gardens, Elua, and Gotchas

❓ How do you kill Moloch — or at least close the offer — without summoning something worse?

Alexander’s line is famous: only another god can kill Moloch. Soft values (Elua) need power plus coordination. Every serious escape attempt sits on the tyranny–discoordination tradeoff: enough central power to impose the god’s-eye solution, without Stalin; enough freedom to avoid tyranny, without fish-farm hell. Schmachtenberger’s third attractor is the same ridge with more ecological vocabulary. This chapter inventories proposed escapes and the Gotchas attached to each.

Coordination mechanisms (earthly gardens)

❓ What ordinary tools already fight multipolar traps?

Law + enforcement: pollution rules on the lake; arms control treaties; licensing.
Norms + stigma: students shunning cheaters; professional ethics.
Markets with binding proxies: when customers can see and punish harm.
Technical commitment devices: monitoring, audit logs, compute cluster controls, provenance.
Polycentric governance (Ostrom): local commons rules that sometimes work without world government.

Limits: governments fall into Moloch (corporate welfare, electoral crime theater). Norms fail under intense global competition. Treaties fail without verification. Ostrom conditions (clear boundaries, monitoring, graduated sanctions, etc.) are hard at planetary AI scale (U3).

Still: “coordination is impossible” is false; sufficient coordination for ASI-era stakes is the open problem.

Capture Gnon / walled gardens — and Gotcha

❓ Why doesn’t a virtuous enclave solve it?

Warg Franklin-style programs (rational theocracy, conservative order, bottled capitalism, martial sovereignty) try to garden locally and “capture Gnon.” Alexander’s reply: Gotcha — you die anyway. Outsiders who don’t adopt your restraints outcompete you on disease, memes, armies, markets, or shared existential catastrophes. Higher walls consume more resources until the garden is a bunker in a burning cosmos. Enclaves are not closed systems if they remain part of the cosmos.

This matters for EmTech strategy: “we’ll just build safe AI in one nice country” fails if unsafe AI elsewhere still ends you — unless the garden has global teeth or the tech is intrinsically local (rarely true for software).

Submit to Gnon — and Gotcha

❓ Is the answer radical compliance with natural competitive law?

Nick Land-flavored Gnon-conformity says civilization dies when it denies selection. Alexander lifts the veil: Gnon’s gifts are contingent outputs of blind processes that will discard you next. Submission is not freedom. Gotcha.

For AI: “let the race rip, selection will produce the worthy successor” is a value-laden bet dressed as realism — and it discards the orthogonality insight that optimization power ≠ human-goodness.

The friendly singleton / Elua path

❓ What is Alexander’s preferred long-term kill-shot?

If someone lifts to Heaven a superintelligence that shares human values and can prevent rival unsafe optimizations, Moloch’s open offer can close. That is hubris as anti-hubris: not begging the universe to care, but refusing to assume dark gods are destiny.

Risks:
- Racing to build the gardener is Molochian in the short run.
- Orthogonality: most strong optimizers are not Elua.
- Power concentration: Elua’s throne is one election away from Moloch’s mask.
- Multipolar takeoff: several near-peers → Hanson-like competitive stripping rather than clean singleton.

Alexander knows the horror branches; he still treats trying as higher priority than stylish despair.

Make safety the winning X

❓ Can incentives flip so alignment is competitively rational?

If robust safety, interpretability, or verified reliability become what customers, regulators, and talent select for, the sacrifice gradient reverses. Moloch’s Bargain paper implies the opposite gradient under pure engagement/sales optimization — so metric design and governance are first-class technical problems. Differential technology (cyberdefense, bio defense, verification ML) aims to make protective tech outrun offensive race payoff.

This is the most “engineer-native” escape: change payoffs, don’t only preach.

Pause, pace, and partial coordination

❓ Are slowdowns real tools or cosplay?

Voluntary pauses without verification are unstable (Mike free-rides). Pacing via responsible scaling policies, eval gates, and export controls is the 2023–2026 real-world attempt: incomplete, leaky, politicized — and still evidence that some coordination musculature exists. Whether it matches capability curves is unresolved (U1, U6).

Schmachtenberger’s third attractor moves

❓ What does “third attractor” demand beyond AI alignment research?

Sensemaking infrastructure, incentive redesign beyond one industry, ecological reality alignment, and technologies of collective intelligence that are not pure engagement Moloch. Intelligence bound by wisdom — slogan-level, but directionally: don’t maximize unbound proxies. Complements lab alignment rather than replacing it.

Contrarian: maybe Moloch is overfit

❓ When should we stop using the Moloch lens?

When a bad outcome is single-agent abuse or simple error.
When competition clearly raises absolute human welfare with stable side-constraints.
When the metaphor substitutes for institution-level design work.
When it implies fatalism that excuses free-riding on activism (“we’re doomed, ship it”).

Use Moloch when you can name poles, X, sacrificed value, and missing commitment device.

Secondary Q&A strip

Secondary	Answer
World government?	Strong coordinator; tyranny risk; incomplete sovereignty today.
Crypto stop-switches?	Often anti-coordination by design — essay’s warning.
Is Elua “just liberalism”?	Broader: human values complex; not only thin hedonism (see Misperceptions post).
Best near-term bet?	Stack: metrics + verification + partial treaties + technical alignment + cultural healthy-competition norms.

Only another god can kill Moloch; walled garden watching the sky burn; Elua mocked by dark thrones yet still standing.

← 05 · Superorganism 07 · Strategy synthesis →