335 | Andrew Jaffe on Models, Probability, and the Universe
What this covers
Andrew Jaffe, a cosmologist at Imperial College London, sits with Sean Carroll to argue that science does not rest on certainty but on models and probabilities. The conversation moves through how Bayesian Inference works as a framework for choosing among competing models—from quantum mechanics to cosmology—and why induction, which can never reach absolute proof, finds its most honest expression in Bayesian updating. Jaffe defends the view that all knowledge of the world is necessarily model-based and probabilistic: we cannot step outside our frameworks to compare them to reality itself, only test whether they make accurate predictions. The core tension runs throughout: if probabilities are always conditional on a model, and all models are human constructs, how can we speak of objectivity in science at all?
The discussion ranges across quantum mechanics, where QBism reinterprets the wave function as knowledge rather than physical fact; statistical mechanics and entropy, where even the second law turns out to depend partly on what an observer knows; and cosmology, including measurement of the cosmic microwave background and the Hubble constant. A practical point emerges: Frequentist Statistics and Bayesian methods can yield identical error bars while answering fundamentally different questions about what those numbers mean. Jaffe also addresses the Problem of Induction and the Anthropic Principle, showing how Bayesian Inference handles both. Where the two thinkers diverge—on whether probabilistic reasoning truly explains thermodynamics, on the status of Many-Worlds Interpretation, on whether even large language models qualify as models of the world—the disagreement is precise and earns its place in the conversation.
Jaffe argues that all knowledge of the world is necessarily model-based and probabilistic rather than certain or foundational, and that Bayesian probability is the right framework for understanding how science chooses among models from quantum mechanics to cosmology.
- We cannot know anything about the world without a model, and all probabilities are conditional on a model.
- Induction can never reach certainty, but Bayesian updating lets us become progressively more confident as data accumulate.
- Even quantum mechanics and statistical mechanics are best read as probabilistic statements about what observers know, not absolute laws.
What counts as a model depends on compression degree, not on categorical differences in kind.
- Even a large language model conceived as a gigantic lookup table that produces sensible answers counts as a model of the world; physicists merely prefer vastly compressed, simple models, so the question of whether an LLM 'has' a model may collapse into a question of compression rather than kind.
“a gigantic lookup table that has all the right answers, is a model of the world. It's just that maybe as physicists, we're used to having these vastly compressed models”
Bayesian statistical reasoning explains cooling via overwhelming improbability, answering frequentist objections about thermodynamic irreversibility.
- Against David Albert's frequentist objection, Carroll maintains that knowledge-based statistical reasoning does explain why coffee cools to room temperature—at least it makes the usual cooling unsurprising—because with 10^23 particles departures from the expected behavior are overwhelmingly improbable.
“you can't tell me that the reason why coffee cools to room temperature in a room is because of our knowledge of what's going on in it. And I want to say, yes, I can”
QBism is a pragmatic coping strategy that physicists will eventually embrace because subjective Bayesian commitment is already common ground.
- Carroll conjectures that QBism is an effective coping strategy for physicists who do not want to accept many-worlds and that they will eventually arrive there anyway, since the Bayesian commitment is common ground; he supports this with the De Finetti slogan that probabilities are subjective rather than nonexistent.
“my current conjecture... is that QBism is an effective coping strategy for people who don't want to accept many worlds, and they'll eventually get there”