Mindscape Podcast Episode 335: Andrew Jaffe on Models, Probability, and the Universe

Date: November 10, 2025

Host: Sean Carroll
Guest: Andrew Jaffe (Theoretical Cosmologist, Imperial College London)
Theme: The interplay between models, probability, and our understanding of the universe; how scientific reasoning is inherently provisional and probabilistic; and the practical and philosophical implications of these ideas in science and cosmology.

Episode Overview

This episode explores the crucial roles played by models and probability in making sense of the universe, both in the physical sciences and in daily life. Sean Carroll and Andrew Jaffe dive into the mindset of modern science, emphasizing the provisional, model-based, and probabilistic nature of knowledge. The discussion bridges examples ranging from gravity to the cosmic microwave background, and touches on fundamental questions about determinism, quantum mechanics, and Bayesian vs. frequentist approaches.

Key Discussion Points & Insights

1. The Nature of Scientific Knowledge and Models (00:00–13:51)

• Certainty vs. Provisional Knowledge:
  • Scientific knowledge doesn’t allow for absolute certainty; all understanding is based on models which are provisional and probabilistic. (00:00)
• Models as Universal Tools:
  • Not exclusive to scientists—everyone, even children, uses models to make sense of the world.
  • “We can't know anything at all about the world without a model for the world.” — Andrew Jaffe (04:46)
• Definition of a Model:
  • A model is "a story about the world" that helps us navigate and predict outcomes; it need not be mathematical.
  • Example: The London Tube map as a model that is functional, non-literal, and good only for certain things. (05:38)
  • “The map is not the territory.” — (05:38)
• Models in Science:
  • Newton’s laws as a model; sometimes replaced by better models (e.g., Einstein’s relativity) in specific domains. (06:51)

2. Models and Artificial Intelligence (07:41–12:11)

• Do Large Language Models (LLMs) Have Models?
  • The difference (or lack thereof) between massive lookup tables and true "models of the world."
  • Reference to Searle’s Chinese Room: is a system capable of understanding if it produces the right outputs? (10:21)
• Phenomenology Over Internal Structure:
  • Intelligence (human or AI) is judged by outputs and interactions, as internal workings can't be directly inspected—apart from artificial systems, where design is known but still opaque.

3. Models, Arguments, and Learning About the Universe (12:11–15:04)

• Models vs. Deductive Proofs:
  • Scientists propose multiple models for phenomena and judge their fit to data, not aiming for “proofs.”
  • Anecdote: Carroll’s debate with William Lane Craig on cosmology illustrated different approaches—deductive logic vs. model-based reasoning. (12:11)
  • “Literally the only way we can understand the world is with models.” — Andrew Jaffe (13:43)

4. Induction, Probability, and Hume’s Problem (15:04–29:20)

• Deduction vs. Induction:
  • Deduction yields absolute truths within systems (math); induction tries to infer general rules from finite observations and is always probabilistic. (14:10)
• David Hume & Induction:
  • Observing falling objects doesn’t guarantee the rule holds forever; certainty is unreachable, all we can do is update probabilities based on accumulated evidence. (15:56)
• Scientific Progress as Model Comparison:
  • Example: Newton vs. Einstein’s gravity, and the probabilistic updating of our best models with each new observation (mercury’s orbit, gravitational lensing, etc.).
• No Bedrock Certainty in Science:
  • Even our best models might be superseded; science’s lack of absolute foundation is liberating and ensures endless discovery. (24:04)

5. Bayesian Reasoning and the Problem of Induction (25:29–29:52)

• Bayesianism as a Solution:
  • Bayesian updating offers a coherent way to handle the problem of induction; never absolute certainty, but probabilities that get more robust with more data.
  • “All probabilities are conditional.” — Andrew Jaffe (40:18)
• Subjectivity in Probability:
  • Bayesian reasoning is subjective in the sense that conclusions depend on priors (initial assumptions).
  • “Within the Bayesian formalism, there’s no such thing as no information…You have to go in with something.” (40:19)

6. Bayesians vs. Frequentists in Science (29:52–50:46)

• History of Probability in Science:
  • The gradual shift in astrophysics and cosmology from frequentist to Bayesian thinking, especially in situations where repeats of experiments are impossible (e.g., supernova 1987A neutrino detection). (31:43)
• Concrete Examples:
  • How Bayesian reasoning helps in interpreting the cosmic microwave background (CMB) and measuring parameters like the Hubble constant.
  • Frequentist vs. Bayesian error bars—practically the same numbers, but philsophically different interpretations.
• Disagreements and Debates:
  • The current "Hubble tension"—different measurements give different values for the Hubble constant, with the CMB and supernova methods disagreeing. Both groups may be partly wrong, reflecting challenges in modeling and measuring. (44:16)

7. Probability in Statistical Mechanics and Quantum Mechanics (50:46–68:15)

• Statistical Mechanics:
  • Probability arises from our lack of knowledge of all micro-details; entropy reflects information and usable energy is linked to what we know. (50:46)
  • “If you knew more about the gas, you could extract more work from the gas.” (55:46)
• Quantum Mechanics and Probability:
  • Quantum probabilities are intrinsic to the theory (wavefunction), but the meaning is debated (ontological vs. epistemological).
  • Overview of interpretations:
    • Many Worlds: All possible outcomes happen; probabilities reflect branches.
    • QBism (Quantum Bayesianism): Probabilities reflect personal belief/updating based on observation.
    • Copenhagen: Collapse upon measurement, but works operationally regardless of deeper meaning.
  • The Bayesian approach has facilitated moving away from consciousness-related “wavefunction collapse” to focusing on knowledge and belief. (60:43)
  • “It should be no surprise that our theories are probabilistic, because that’s the only way we have for understanding the world.” — Andrew Jaffe (65:50)

8. Probability, Multiverses, and Cosmology (69:10–77:13)

• Inflation, Multiverses, and the Measure Problem:
  • Eternal inflation and the multiverse produce conceptual challenges: if infinite universes exist, does that restore frequentist notion of probability? (70:13)
  • Bayesianism can work even if there’s only one universe; probabilities express our credence, not frequencies in a literal ensemble.
  • The anthropic principle—our existence as data—shapes which models we should favor.
• Open Questions:
  • Some philosophical or cosmological questions (e.g., between Many Worlds and QBism) may never be empirically settled; prior beliefs and pragmatic models will continue to dominate.

Notable Quotes & Memorable Moments

• “We can't know anything at all about the world without a model for the world.”
  — Andrew Jaffe (04:46)
• “The map is not the territory.” (05:38)
• “Our job [as physicists] is to suggest a model and try to figure out which one works best.”
  — Sean Carroll (12:11)
• “Literally the only way we can understand the world is with models.”
  — Andrew Jaffe (13:43)
• “All probabilities are conditional, and that means all probabilities are given a model.”
  — Andrew Jaffe (40:18)
• “If a question begins ‘is it possible that’, the answer is always yes, but we would like to know relative degrees of likelihood or credence.”
  — Sean Carroll (29:20)
• “It should be no surprise that our theories are probabilistic, because that's the only way we have for understanding the world.”
  — Andrew Jaffe (65:50)
• “Physicists do like the idea that there's a right answer, but this is one of those cases where maybe, maybe we can't know.”
  — Andrew Jaffe (69:10)

Timestamps for Major Segments

• 00:00–04:04 — Introduction to models and probability in science
• 04:46–06:39 — What is a model? Models in daily life and science
• 07:41–12:11 — Artificial intelligence, Turing Test, and models in LLMs
• 12:11–15:04 — Models vs. deductive arguments; science and the role of models
• 15:04–24:04 — Induction, deduction, probability, and David Hume’s skepticism
• 24:04–29:52 — Bayesian vs. frequentist reasoning and the problem of induction
• 29:52–47:26 — Bayesians in cosmology (CMB, neutrinos, the Hubble constant)
• 50:46–57:28 — Probability in statistical mechanics, entropy, laws of thermodynamics
• 57:28–68:15 — Probability in quantum mechanics; interpretations of quantum probability
• 69:10–77:13 — Multiverses, inflation, Bayesian reasoning at the frontiers of cosmology

Tone and Style

• Conversational, thoughtful, and philosophical, blending technical depth with relatable analogies (e.g., London Tube map, children modeling the world, betting on the Hubble constant).
• Friendly disagreement and open-mindedness — both host and guest draw from shared experience but acknowledge where science leaves room for interpretation and new discoveries.

Summary for New Listeners

If you’ve ever wondered how scientists (and the rest of us) learn about the world despite never having absolute certainty, or why so many areas of science rely on probability, this episode offers both accessible explanation and deep insight. Sean Carroll and Andrew Jaffe unpack why everything we know is rooted in models and provisional inference, why probabilities are intrinsic to both practical data analysis and our deepest theories (statistical mechanics, quantum mechanics, cosmology), and how this understanding shapes both current research and our philosophical outlook on knowledge itself.