345 | Adam Elga on Being Rational in a Very Large Universe

Episode Summary

Executive Summary: Sean Carroll and philosopher Adam Elga explore rational belief under uncertainty, focusing on self-locating uncertainty, the Sleeping Beauty problem, anthropic reasoning, Boltzmann brains, and cosmological inference. They debate whether observers should count equally, whether theories with more copies of oneself get Bayesian “boosts,” and how to avoid self-undermining skepticism in physics and cosmology.

Main Topics: Bayesian reasoning and belief updating (Priority: 5/5): Carroll frames science as Bayesian updating: priors plus evidence yield revised credences. The discussion uses dark matter candidates and cosmological models to show how rational agents compare competing theories without certainty. Disagreement among peers and deference to prior self (Priority: 4/5): Elga explains how to respond when an equally informed peer disagrees, favoring deference to what one’s prior self would have judged under the same anticipated disagreement, rather than simple averaging or stubbornness. Self-locating uncertainty and teletransportation (Priority: 5/5): The conversation uses transporter-style duplication cases to ask what probability should be assigned to being one copy versus another, highlighting uncertainty about one’s location or instantiation within a world. Sleeping Beauty and thirding vs halfing (Priority: 5/5): Elga summarizes the Sleeping Beauty problem and the debate between halfers and thirders. The issue turns on whether learning it is Monday preserves probability ratios or boosts the world with more awakenings. Anthropic reasoning and cosmological theory choice (Priority: 5/5): They connect self-locating beliefs to cosmology: whether universes with more observers, more copies, or more frequent observer-moments should receive higher credence, and whether this is ‘presumptuous.’ Boltzmann brains and self-undermining evidence (Priority: 5/5): A major section examines Boltzmann brains, the idea that random fluctuations could produce observer-moments like ours. The paradox is that if we are Boltzmann brains, our reasoning may be unreliable and undermine itself. Externalism, skepticism, and rational escape routes (Priority: 4/5): The pair discuss whether externalist epistemology, anti-skeptical priors, or special self-locating commitments can dissolve skepticism about brains in vats, simulations, or Boltzmann cosmologies.

Key Arguments: Scientific reasoning often looks Bayesian, but cosmology and self-location create cases where standard Bayesian updating does not straightforwardly tell us how to assign credences. In peer disagreement, a rational strategy may be to defer to what your earlier self would have thought conditional on the disagreement, rather than automatically privileging your present belief. The Sleeping Beauty problem turns on whether one should update by preserving probability ratios after learning it is Monday; thirders say the two-waking world gets a boost because it has more instances of the relevant experience. In anthropic settings, theories with more copies of an observer-like experience seem to gain confirmation, but this can appear ‘presumptuous’ because it lets sheer multiplicity drive theory choice. Boltzmann brain scenarios are troubling because if most observer-moments are random fluctuations, then the evidence supporting belief in the cosmology may itself be unreliable. A self-undermining argument can pressure us to reject Boltzmann-brain-dominated cosmologies, but Elga argues the issue is more subtle: some skeptical stances are stable and cannot be dismissed so easily. One possible escape is to assign intrinsically lower prior plausibility to radically deluded or random-observer scenarios, analogous to how one assigns low priors to skeptical hypotheses like brain-in-a-vat cases. Carroll’s alternative intuition is that a theory should be judged by the probability that at least one observer exactly like us exists in it, rather than by counting all observers equally, which may avoid runaway boosts from duplication.

Data Points: Sleeping Beauty awakenings: 3 possibilities - Heads-Monday, Tails-Monday, Tails-Tuesday are the relevant centered possibilities in the problem as discussed. Sleeping Beauty update under thirder view: 2/3 tails, 1/3 heads - Derived when preserving probability ratios after learning it is Monday. Transporter copies example: 100 Potemkins vs 1 Enterprise - Used to test whether one should assign more credence to being in the single favorable copy or among many unfavorable copies. Split-the-check intuition: 50-50 or adjusted by confidence - Used to motivate deference to prior self in disagreement cases and to show not all disagreements warrant equal weight. Cosmological constant / fine-tuning: small but not zero - Discussed as an example where scientists treat some numerical values as especially surprising or suggestive. Parameter count in theories: 10 extra parameters - Carroll notes that naive Bayesian reasoning may over-penalize theories with many parameters more than scientists actually do. Anthropic red/green example: 100% vs 50% existence of a matching observer - The second theory has a 100% chance of at least one observer like the subject, versus 50% in the first theory. Example of surprise number: π times the mass of the electron - Used to illustrate how specific measured values can suggest undiscovered explanatory structure.

Pivotal Quotes: "I characterize myself as addicted to rationality." — Adam Elga: Elga describes the organizing theme of his philosophical work. "You should think, well, I'm still right. So stay confident that I'm right. But I also think that's irrational." — Adam Elga: Explanation of the level-splitting view in peer disagreement. "The world that involves more awakenings gets a kind of boost associated with the fact that there are more instances of that state of mind being instantiated." — Adam Elga: Summary of the thirder/SIA-style intuition in Sleeping Beauty and anthropic reasoning.

Implications: The episode shows that cosmology, quantum mechanics, and epistemology run into the same deep problem: how to reason about observer-moments without circularity. For scientists and philosophers, it highlights unresolved choices about priors, reference classes, and when skepticism becomes self-defeating.