What this covers

Sean Carroll and Ned Hall begin with a deceptively simple question: what grounds our confidence that tomorrow's laws of nature will resemble today's? Their conversation unfolds the philosophical architecture beneath that question, revealing how disagreement over laws of nature splits into two camps. Humeans, following David Hume, treat laws as convenient summaries of what actually happens—no more. Anti-Humeans insist laws have independent existence that constrains which futures are possible. This fundamental divide, Hall argues, cannot be resolved in isolation; it entangles with how we understand possible worlds, counterfactuals, and what causation really is. The discussion begins with causation but quickly branches into the conceptual groundwork those disputes require.

Hall moves through several distinct territories. He reconstructs Hume's problem of induction—the puzzle that justifying extrapolation from past experience seems to require assuming nature's uniformity, which is itself an empirical claim—and explores how Hume's psychology of projection offers an alternative to grounding necessity as real. The conversation examines the interventionist theory championed by Judea Pearl, which analyzes causation through minimal-change comparisons between worlds, and discusses cases where such comparison becomes incoherent or useless. Hall also unpacks how possible worlds clarify ambiguities ordinary language obscures, and how the vocabulary problem—illustrated through Nelson Goodman's grue paradox—affects which laws look simple and which gerrymandered. Throughout, the pair wrestle with whether modal powers exist in the world or function as a useful conceptual grid we impose, a tension that resurfaces wherever causation, explanation, or induction appear.

Sharpest takeaway

Hall argues that our talk of laws, causes, and counterfactuals splits philosophers into Humeans (laws are convenient summaries of what actually happens) and anti-Humeans (laws have independent existence that constrains what would happen), and that resolving causation requires first settling how we compare possible worlds.

  • The Humean/anti-Humean divide turns on whether modal 'powers' are real features of the world or a useful conceptual grid we impose
  • Scientific explanation relies on counterfactuals, which anti-Humeans say require robust laws to ground
  • Causation is analyzed via comparison of possible worlds, but which minimal-change world to compare against is unresolved

The argument · threads7 threads · 32 claims
0.72

Humean and anti-Humean accounts of laws diverge fundamentally on whether laws are mere summaries of actuality or possess objective necessity independent of what happens.

5 pointscentrality 5/5
  • There are competing views about the 'knitting together' of causes: some hold a two-factor view where reality contains both the stuff that happens and the necessities/powers, while others hold that causal/lawful talk is merely a conceptual grid we impose that need not latch onto any real structure out there.

    “there are some people, philosophers and non-philosophers alike, who think really you need a two-factor view of what reality is like. There's kind of the stuff that happens and then there are the necessities, the powers.”

  • One Humean approach treats laws of nature like axioms in mathematics: a small set of statements singled out not because they shine more brightly as fundamental truths, but because they effectively organize our practice—an anthropocentric choice serving our interests in prediction and explanation rather than tracking objective fundamentality.

    “it's a view that's anthropocentric in a certain way, 'cause it's we who find these things obvious or epistemically powerful... But it's not out there in the mathematical universe. If we imagine like we're Platonists and we think there's a universe of mathematical truths, it's not that some of them somehow shine more brightly as like being the fundamental truths.”

  • Imagine a unique experiment that, under theory A succeeds but under theory B triggers a new Big Bang; an anti-Humean treats it as a fact which laws are correct so we should cautiously avoid risk, while a strict Humean—who holds laws are summaries of what actually happens—would find that if the experiment is never run, the unique conditions never occur, leaving it indeterminate which laws are correct, which conflicts with how we actually reason.

    “If we don't run the experiment, then there will be no such conditions to, as it were, include in the summary... So it looks like you ought to say, 'If we don't do the experiment, then it's simply indeterminate whether the laws are the A laws or the B laws.' But that's not the way we would actually think about this.”

  • If fundamental dynamical laws are irreducibly probabilistic, an anti-Humean can hold that the probabilities are real facts over and above frequencies, allowing for the possibility that observed frequencies are 'unlucky' and don't match true probabilities, whereas a Humean who treats probability talk as part of the conceptual grid encapsulating frequencies has no clear sense in which we could be 'unlucky.'

    “I will think that while it's likely that those probabilities are reflected in observable frequencies, there's no guarantee of that... Whereas, if I think that, no, talk of probabilities is just another component of this conceptual grid, I have no particular reason to grant that possibility.”

  • Because the Humean treats laws as mere summaries of the actual world, comparing it to other possible worlds requires stipulating which statements count as laws; the anti-Humean worries this stipulation is governed by what is useful to us, whereas what is genuinely explanatory should not depend on our purposes and interests.

    “The worry that the anti-Humean has with this approach is that the stipulation seems like it's governed by questions about what's useful to us, whereas what's explanatory, what's genuinely explanatory shouldn't be governed by those questions.”

0.62

Induction faces an unsolvable circularity: justifying extrapolation from limited data requires assuming nature is uniform, but that assumption is itself empirical and cannot be empirically justified without presupposing uniformity.

2 pointscentrality 5/5
  • Hume's problem of induction holds that extrapolating from limited data requires a background assumption that nature is uniform, but that assumption is itself an empirical claim; if every empirical claim must be backed by empirical data while the very process of substantiation presupposes uniformity, the justification runs in a circle, so induction cannot be grounded as rational reasoning—though we must do it anyway.

    “if you have the background attitude that any empirical claim about the world has to be substantiated by empirical data, but you also think that the very process of substantiation, in order for that to get off the ground, we need to pre-suppose that nature operates according to simple principles or is uniform, then it looks like you're running in a circle here.”

  • On Hall's reading, Hume conceives of human reason as analogous to digestion—a faculty that serves us and can serve us better or worse but should not be over-inflated—so one can deny that any rational argument grounds extrapolation while accepting that we inevitably extrapolate and that some ways of doing so are better for us than others.

    “the way I read Hume, he thinks of human reason as sort of like digestion. It's something that serves us in a certain way. And it can serve us better or worse. But it's a mistake to over-inflate it.”

0.55

Determining which possible worlds to compare when assessing causation—not merely whether a counterfactual differs from the actual world—is the central difficulty of causal analysis.

4 pointscentrality 4/5
  • To say one event caused another (e.g., bad sushi caused getting sick) is to compare our world to another possible world where the cause is absent, but not just any such world—comparing to a world with a different but similarly harmful cause (bad chicken wings) won't do—so the central difficulty of causation lies in determining which possible worlds to compare and how.

    “Here's a possible world, in which you didn't have bad sushi, you had bad chicken wings instead... So that's where all the fun lies. It's like comparing, how do we compare the possible worlds?”

  • The interventionist account has severe limitations for variables that cannot be manipulated or where it is unclear what manipulation would even mean—such as social class or race—and for cases where a minimal localized intervention introduces confounding violations of physical law (e.g., God instantly changing the color of someone's shirt would itself become the salient cause of the conversation's course).

    “in the social sciences, we might want to attribute effects to features of the world where they're... It's not just that we can't manipulate them, it's not even clear what we would have in mind by manipulating them.”

  • Tracking the causal relations we care about often requires choosing the right temporal scale: we ask whether a decade-long history of heavy smoking caused cancer, not whether a single puff did, so intervening on such protracted events in a localized way without introducing confounding variables is genuinely problematic for interventionism.

    “We really feel like in order to track the causal relations we're interested in, we need to zoom out to a longer time scale. So what would it be to intervene on that history in a localized way that doesn't introduce potentially confounding variables.”

  • For some causal questions, the relevant interventionist counterfactual is incoherent or uninformative: asking how stressed Joey would be if he had two parents but were systematically teased for having one parent produces a Kafkaesque scenario that fails to illuminate the original causal question, showing minimal-change comparison can mislead.

    “How stressed out would he be if he had two parents and were systematically teased at school for having one? And as Alex points out, he would be in some Kafkaesque scenario... whatever our answer to that question is, it simply does not seem to be illuminating the causal question we were asking in the first place.”

0.51

Anti-Humeans argue that robust laws grounding genuine counterfactuals are essential to explaining why scientific explanation appeals to what would have happened under different conditions.

3 pointscentrality 4/5
  • Scientific explanation, unlike mathematical explanation, routinely cites not just what happens but what would have happened under different conditions; for example, Newton's inverse-square law explains orbital precession by entailing that orbits would not precess absent other planets, and making sense of such explanatory counterfactuals is where the anti-Humean argues we need robust laws.

    “Newton puts us in a position to evaluate a hypothetical; if there had been no other planets then this orbit would not have precessed. And that actually looks like it's an explanation of the precession of the orbits, and it looks like an explanation that cites, not just facts about what does happen, but facts about what would have happened”

  • Some philosophers like David Armstrong argue that unless laws possess genuine power or force, induction is irrational and we should expect nothing about the future beyond 'something will happen'; thus attributing real necessity to laws may provide a positive justification for extrapolation that mere regularity cannot.

    “David Armstrong, who's... He's deceased now, but he was a very influential Australian philosopher, who argued that unless we see laws as having some genuine power or force, induction is irrational.”

  • Consider a world of two Newtonian particles in circular orbit due to mutual gravity versus a world where each particle independently follows a simple law of perpetual circular motion; Humeans see these as identical descriptions of the same facts, while anti-Humeans insist they differ because they make different predictions about what would happen if a third particle were introduced.

    “here's something that's true of the Newtonian world. If there were a third particle somehow introduced, the motions going forward would no longer be circular... Whereas, in the other world, where there's just a simple law that applies to all particles separately, if you introduced a third particle, then, by stipulation, it too would move in a circular motion.”

0.44

Possible worlds are best understood as complete and consistent stories, clarifying modal ambiguities language ordinarily blurs and enabling analysis of modality itself.

9 pointscentrality 2/5
  • Philosophers conventionally introduce a contrasting hypothetical character by prefixing 'schm' to an existing name (e.g., Schmisaac Newton).

    “That's what philosophers do when they want to introduce a character, they just add 'schm' to the front.”

  • Possible-world talk clarifies ambiguities ordinary language blurs—for example 'I could have been taller' is false on a reading requiring a world where Ned is taller than himself but true on a reading where Ned in another world has a greater height than Ned in the actual world.

    “we see this just in microcosm here how talk of possible worlds can help clarify distinctions that ordinary language can blur.”

  • David Lewis argued for modal realism: that possible worlds are not mere useful devices or stories but literally concrete chunks of reality, each a separate region bearing no spatiotemporal relation to ours, motivated by the demand to analyze possibility and necessity into non-modal terms rather than leaving an unanalyzed notion of consistency at rock bottom.

    “a possible world is literally a concrete chunk of reality as a whole. How is it related to the chunk that we're in? It bears no space-time relation to our chunk at all.”

  • A possible world is intuitively a way reality (or our corner of it) could have been—best understood as a complete and consistent story about the world, where completeness means the story answers any question we might pose and is so full that any addition would render it inconsistent.

    “a possible world is something like a complete consistent story about the world.”

  • Historically the ultimate anti-Humean view was theistic—Newton took writing laws to be reading God's commandments—and a benevolent intelligible deity gives a positive reason to expect nature to be uniform enough to discover, whereas a deity with unknown intentions leaves Hume's problem of why we should expect simple laws to re-arise.

    “someone like Newton would have happily said, 'In writing down the laws, I'm literally seeing into the mind of God, like laws are God's commandments.' And if you had that view, that's a very anti-Humean view.”

  • On one view (which Hall presents without endorsing), color is an illusion projected by our visual system—useful for survival but not really in the world—though a richer modern physics of charged matter and photons reopens space for treating colors as real elements.

    “it is an illusion, although unlike other illusions, like seeing a stick in water that looks bent or something like that, a highly serviceable and useful illusion.”

  • If our cognitive equipment was shaped by evolution to navigate the world at our scale, there is reason for unease about whether it can grasp phenomena at extreme scales such as quantum gravity, since the tools that catch everyday causal regularities may not transfer to fundamentally different regimes.

    “you might start to get uneasy, like, great, why should we think that the cognitive equipment that evolution's bequeathed us with, is going to let us do quantum gravity?”

  • Statistics as a field began in the early 20th century fairly hostile to the notion of causation—an early book-length treatment (attributed to Pearson) included a chapter arguing causality should be banished from good science—because the precisely definable notion of correlation seemed far more scientifically rigorous than ordinary causal talk.

    “statistics as a field started out in the early 20th century fairly hostile to the notion of causation... a book-length treatment of statistics early in the 20th century, where there's a whole chapter on why the notion of causality should be banished from good science.”

  • Hume added that humans naturally project features of their own psychology onto the world—just as color may be a serviceable illusion projected by our visual system rather than a real feature—so our sense of necessary connection in nature is the projection of our psychological habit of expectation rather than something we passively detect.

    “a kind of view that we humans naturally tend to project kind of features of our psychology out into the world.”

0.34

Causal and lawful talk introduces powers and capacities as a dimension distinct from spatiotemporal regularities, capturing how we ordinarily understand the world's structure.

2 pointscentrality 3/5
  • Talk of cause and effect goes hand in hand with talk of the powers or capacities of things (e.g., ibuprofen has the power to cure headaches), introducing a dimension to our picture of the world distinct from the blow-by-blow account of what happens where and when.

    “ibuprofen has the power or capacity to cure headaches. And that talk of powers and capacities seems to introduce a dimension to our sort of picture of the world that's different from the dimension we're focusing on when we simply give a kind of blow-by-blow account of what happened, where and when.”

  • On the interventionist approach (championed by figures like Judea Pearl), causal relations are relations exploitable in principle for manipulation: our interest in causation is an interest in facts we can use to settle what would happen if we did one thing or another, with the cleanest test being a god-like localized intervention that tweaks one variable while holding everything else fixed, modeling an ideal controlled experiment.

    “the core thought that is that causal relations are relations that, at least in principle, are exploitable for purposes of manipulation.”

0.34

The gerrymandering problem and vocabulary-dependence of simplicity show that empirical adequacy alone underdetermines which descriptions and extrapolations we should adopt.

2 pointscentrality 3/5
  • Nelson Goodman's grue/bleen problem shows that whether a pattern looks simple depends on the descriptive vocabulary: green and red are inter-definable with bent predicates like 'reen' and 'gred' that flip at a date, so a world looks simple in one vocabulary and bent in another, though grounding descriptions in fundamental physical details may reduce (but not eliminate) this flexibility.

    “It's the world in which you know very young tomatoes are reen, and ripe tomatoes are gred, always.”

  • A 'gerrymandered' law—like Schmisaac Newton's gravity that works normally until 2020 then drops to zero—is designed to make identical predictions to the standard law within all currently testable ranges while differing outside them, illustrating that empirical adequacy alone cannot select between rival extrapolations.

    “This is a way of fashioning a force law that by design gives you the same predictions for any phenomena that we're presently in a position to look at or test, but differs outside the range that we have direct access.”