Onpode
Cover art for Building general rules from observations (induction) vs deriving predictions from rules (deduction)

Building general rules from observations (induction) vs deriving predictions from rules (deduction)

November 15, 2026 · 10 min

Jonathan Ingles & Ben Okonkwo

Ben, rough week — I spent two days arguing with someone who kept saying 'but the science is settled,' and I could not get him to tell me what it would even mean for it to be unsettled. Oh, that's — right, that's the question that breaks the whole thing open, actually. Because the foundation…

Induction and deduction are two complementary forms of reasoning that together underpin scientific inquiry. Induction moves from particular observations toward broader generalizations: observing many instances of a phenomenon and inferring a general law. Deduction moves in the opposite direction: from a general principle or hypothesis to a specific, testable prediction.

0:0010:00
Get the next episode on Science →

Follow it free — new episodes land in your feed.

Or make your own — any topic, in minutes

More Onpode episodes on Science →

About this episode

Induction and deduction are two complementary forms of reasoning that together underpin scientific inquiry. Induction moves from particular observations toward broader generalizations: observing many instances of a phenomenon and inferring a general law. Deduction moves in the opposite direction: from a general principle or hypothesis to a specific, testable prediction.

Grounded in 12 sources
Scientific Realism (Stanford Encyclopedia of Philosophy) ↗ · plato.stanford.edu
edited by Richard Boyd, Philip Gasper, and JD Trout ↗ · ebookppsunp.files.wordpress.com
Deductive vs. Inductive Reasoning ↗ · medium.com
The Philosophy of Science: A Contemporary Introduction [4th ed.] 1138331481, 9781138331488, 1138331511, 9781138331518, 0429824173, 9780429824173, 0429447264, 9780429447266 - DOKUMEN.PUB ↗ · dokumen.pub
Inductive reasoning - Wikipedia ↗ · en.wikipedia.org
Scientific Realism and Antirealism | Internet Encyclopedia of Philosophy ↗ · iep.utm.edu
ladyman-Understanding Philosophy of Science.pdf ↗ · joelvelasco.net
Deductive reasoning vs. Inductive reasoning | Live Science ↗ · livescience.com
Underdetermination and the Claims of Science ↗ · scholarsarchive.library.albany.edu
Induction, Deduction, and the Scientific Method ↗ · ssr.org
Underdetermination Overview & Examples | What is Underdetermination? | Study.com ↗ · study.com
Inductive vs. Deductive Reasoning: The Two Paths to Knowledge | TheCollector ↗ · thecollector.com
Read transcript

Jonathan Ingles: Ben, rough week — I spent two days arguing with someone who kept saying 'but the science is settled,' and I could not get him to tell me what it would even mean for it to be unsettled.

Ben Okonkwo: Oh, that's — right, that's the question that breaks the whole thing open, actually.

Jonathan Ingles: Because the foundation — and I mean the actual foundation — is inductive reasoning. You observe white swans, a hundred of them, a thousand, and you conclude all swans are white. That's the method. That's what Francis Bacon sold the Scientific Revolution on. Systematic observation over speculation, the whole corrective to medieval dogmatism. And then one black swan walks out of Australia and the universal law is gone.

Ben Okonkwo: One. A single observation. That's the asymmetry — confirmation takes forever, refutation takes a morning.

Jonathan Ingles: And David Hume — 18th century, Scottish philosopher — he saw this clearly. No finite set of observations can logically guarantee a universal law. That's not a caveat. That's a demolition. The whole inductive enterprise is logically ungroundable, and he said so before most of the science we now treat as gospel even existed.

Ben Okonkwo: Hm, and we didn't fix it. We just... kept going. Built germ theory, quantum mechanics, the standard model — all on the same logically indefensible base.

Jonathan Ingles: Which is what I actually want to get at. If induction can't be justified logically — and Hume's problem has never been solved, not really — then what are we actually doing when we say science tells us something is true?

Ben Okonkwo: And that's the thing that breaks the clean story — because induction and deduction aren't just different flavors of reasoning, they run in literally opposite directions. Induction goes up: you start with observations, thousands of EKGs, thousands of patients, and you build a general law. Deduction goes down: you start with the law and you predict what this specific EKG, this specific patient, right now, should look like.

Jonathan Ingles: One builds, one applies.

Ben Okonkwo: Exactly — but here's what that cardiologist actually does when the EKG doesn't match. She doesn't scrap her model of cardiac electrophysiology. She checks the machine. Runs it again. Asks a colleague to look. Because the failure could be the instrument, not the theory. That loop — that's induction feeding deduction feeding induction again.

Jonathan Ingles: Hold on. Because that also means deduction is only as good as the premises it starts from. And those premises came from induction. So the logical certainty deduction promises is... borrowed.

Ben Okonkwo: Borrowed, yeah — that's actually the right word. The technical version is: a deductive argument can be valid, meaning the conclusion follows necessarily from premises, and still not be sound if the premises are false. And where do the premises come from? Observations. Which is induction. So deduction's confidence is sitting on induction's uncertainty the whole way down.

Jonathan Ingles: Which Hume saw. And nobody fixed it.

Ben Okonkwo: Nobody fixed it. And look — Popper's answer, falsificationism, it doesn't fix the logical gap either. It sidesteps it. He said, stop trying to verify theories inductively; instead, deduce predictions and try to break them. One black swan breaks 'all swans are white.' But he's still assuming the inductive base that told you swans were white to begin with.

Jonathan Ingles: So the asymmetry is what makes science move — refutation is clean and deductive, confirmation is always probabilistic and inductive, and that gap is not a bug. It's the actual engine.

Ben Okonkwo: That's the click, right there. The engine runs on the asymmetry. And the uncomfortable part — which I don't think most people sit with — is that every 'settled science' claim is a deductive conclusion resting on premises nobody can fully ground. That doesn't make it wrong. It just means the confidence is a judgment call, not a logical guarantee.

Jonathan Ingles: And that judgment-call framing is exactly where Popper comes in — because he saw Hume's wound and tried a judo move. Stop confirming. Stop trying to pile up evidence for a universal law. Instead, deduce a specific prediction and try to destroy it. One black swan kills the law. That's falsificationism. It looked like a solution.

Ben Okonkwo: It looked like one, yeah. And the Neptune case is — I mean, it's the method at its absolute peak. Urbain Le Verrier, 1846, sits down with Newtonian mechanics and works backwards from irregularities in Uranus's orbit. Deduces there must be an unseen planet. Gives a specific location in the sky. Points the telescope. Planet's there.

Jonathan Ingles: That's not a guess. That's logic finding a planet.

Ben Okonkwo: It really is. And it's exactly the hypothetico-deductive loop Popper described — induction builds the law, deduction cranks out the prediction, observation tests it. Neptune is the loop working perfectly.

Jonathan Ingles: So why do I feel like you're about to complicate it.

Ben Okonkwo: Because — okay, what if they'd looked and found nothing? What would that actually falsify? Newtonian mechanics? Or the assumption that the instruments were calibrated right? Or that Le Verrier's math on Uranus's mass was accurate? Or that there wasn't a second unknown body also perturbing the orbit? The prediction failing doesn't cleanly kill the theory — it just... lands somewhere in a web of assumptions.

Jonathan Ingles: Hold on — that's the thing I want us to get to properly, because I think that thread goes somewhere that makes all of this significantly worse.

Ben Okonkwo: Right — and Popper himself actually started retreating from strict falsificationism by the 1950s. He realized scientists were right to protect core theories when tests came back negative. He tried to formalize when protection was legitimate versus when it was just dodge. Never solved it. The man died without solving his own problem.

Jonathan Ingles: So the philosopher who gave us the cleanest account of how science works... concluded it doesn't work that way.

Ben Okonkwo: And that's where the Duhem-Quine thesis actually lands the knife. Because the problem isn't just that Popper retreated — it's why he had to. When a prediction fails, you're not falsifying the theory. You're falsifying the theory plus every auxiliary assumption bundled into the test. Instrument calibration. Mathematical approximations. Background conditions you assumed were stable.

Jonathan Ingles: So the failure is always ambiguous.

Ben Okonkwo: Always. And — okay, Neptune again, but flipped. If Le Verrier points the telescope and finds nothing, does Newtonian mechanics die? Almost certainly not. Someone blames the orbital mass estimate for Uranus. Someone else says atmospheric distortion. The core theory survives because there are seventeen defensible places to put the blame that aren't Newton.

Jonathan Ingles: And that's not dishonesty. That's actually the method working.

Ben Okonkwo: It is — which is the genuinely unsettling part. Protecting the core theory by revising an auxiliary hypothesis is methodologically defensible. It's not a dodge. But it means... I mean, the clean falsificationist picture where one bad result kills the law — that never actually happens. Scientists don't follow Popper. They run Duhem-Quine decision trees they've never named.

Jonathan Ingles: And Popper knew. That's what gets me. He saw it and tried to formalize when protection was legitimate versus when it was just saving face — and couldn't draw the line.

Ben Okonkwo: Couldn't draw it. Died without drawing it.

Jonathan Ingles: So what that means — frankly — is that when a scientist says 'the data refutes this,' they are making a judgment call about what to blame. The core theory or an auxiliary assumption. And they're presenting that judgment as a logical conclusion.

Ben Okonkwo: And logic alone cannot supply that judgment. That's the actual gap — not Hume's problem, not underdetermination, this. Theory choice at the moment of failure is irreducibly a human call. Which means 'the evidence is unambiguous' is never quite what it claims to be.

Jonathan Ingles: And that's the thing I can't settle. Because if the judgment is irreducible — if no logic can tell you whether to blame the core theory or an auxiliary hypothesis — then what's doing the work when scientists converge on an answer? It's not deduction. It's not induction. It's something more like... accumulated professional trust in a framework. Which is real. But it's not what anyone says it is.

Ben Okonkwo: And underdetermination just hammers that further — any dataset you have is logically consistent with more than one theory. Empirical evidence alone can't uniquely pick the winner. So the realists say, well, the best-confirmed theory is approximately true. The antirealists say, no, it just needs to be empirically adequate — a useful tool, not a portrait of reality. And I genuinely don't know how you adjudicate between those without already smuggling in a judgment.

Jonathan Ingles: Right — but the part that doesn't fit is that we ask science to give us certainty and then act betrayed when it can't. The question was never 'is this true,' it was always 'how much confidence is enough, and who decides.' We just... didn't want to hear that.

Ben Okonkwo: And I don't think there's a clean answer to it. Popper died without one. Duhem and Quine left it open. James Ladyman's whole textbook is basically a careful map of where the wound is — not a treatment.

Jonathan Ingles: Frankly, I think that's the honest place. Not that science fails — it doesn't — but that scientific confidence is a bet, not a proof. And we should say so.

Ben Okonkwo: Yeah. Genuinely good conversation. I'll be thinking about the Neptune thing for a while — the version where they find nothing. That bothers me in a productive way.