Lecture: Science and Explanation
Lecture script · about 15–20 minutes spoken · Week 13
Opening
Picture three claims reaching you in one week. An ad for a cold supplement fills its page with 12 happy customers' stories. A seller says an energy bracelet works, and that no test can measure its effect. A friend says the old dorm must be haunted, since no one has ever proven it isn't.
A nursing student meets claims like these when asked whether a new remedy beats a placebo. A mechanic meets them when deciding which fault best explains a car's clues. A shopper meets them in every ad that promises more than it shows. In each case, one question cuts through: what would prove this wrong?
Here's where we are. Week 4 taught you modus tollens and affirming the consequent. Week 10 taught you what a controlled comparison adds. Week 12 taught you to check where a claim comes from. In this lecture you ask what would test a claim, and I'll walk you through eight steps, from a single hypothesis to one science check you can carry anywhere.
1. Ask what would prove it wrong
Your laptop fan starts roaring. You guess that a browser tab is running something heavy. Then you think: if a tab is the cause, closing the browser should quiet the fan. You close it and listen.
That guess is a hypothesis, a proposed explanation that is still being tested. What you expected to hear is a prediction, what you should observe if a hypothesis is true, stated before you look. And the whole thing was a real test because the fan might have kept roaring.
That's the heart of it. A testable claim is a claim that some possible observation could show to be wrong. A claim that fits every possible observation isn't tested by any of them.
Now the bracelet. The seller says it boosts your energy, and no test can measure its effect. Ask what observation could show the claim wrong. The seller has ruled out every one in advance.
The philosopher Popper built a famous proposal on this idea, called falsifiability. He set it out in a book first published in German in Vienna in the mid-1930s, and in his English edition, The Logic of Scientific Discovery, in 1959. Here's an honest caveat, though. It's one influential proposal, not a settled test. Critics note that many pseudosciences have been tested and shown false, rather than being untestable.
Pause and try this. The bracelet's claim can't be shown wrong by any test. Does that make it proven?
Think about what a test is for.
Here's the answer. No. The claim hasn't survived a test; it has avoided one. Unfalsifiable never means proven.
2. Read the logic of a test
Here's where Week 4 comes back. Every test rests on a conditional: if the hypothesis is true, then the prediction comes true. The hypothesis is the antecedent. The prediction is the consequent.
Take a kitchen smoke alarm that keeps chirping. Your hypothesis is that its battery is low. Your prediction is that a new battery will stop the chirping. You put one in, and the chirping goes on.
So you reason like this. If the battery is low, a new battery will stop the chirping. A new battery didn't stop it. So the battery isn't the cause. That's modus tollens, a valid form. If both premises are true, the conclusion has to be.
Now suppose the chirping stops instead. If you conclude "so the battery was low", you've affirmed the consequent, an invalid form. Here's a counterexample. Taking the alarm down to change the battery may have reset it. Both premises would be true, and the battery story could still be false.
So a passed test gives inductive support, never a guarantee. And a failed test, by a valid form, counts against the hypothesis, unless the test itself went wrong. Maybe the new battery was dead in the package.
Pause and try this. The new battery stops the chirping, and a friend says, "That proves it was the battery." Which form is your friend using?
Say "if P, then Q" to yourself first.
Here's the answer. Affirming the consequent. The result supports the battery hypothesis, but another cause, like a reset, could produce the same result.
A miss can refute; a hit can only support.
3. Compare with a control
Your roommate swears by a new pillow. "I slept great the first night!" Maybe the pillow helped. Maybe it was a quiet night, or a tired day, or the hope of a good night's sleep. One night with nothing to compare can't tell those apart.
That single story is an uncontrolled observation, a case with no comparison group, so it can't show what would have happened without the treatment. The cold-supplement ad prints 12 of them.
Three tools turn stories into a test. A control group is a comparison group that doesn't get the treatment. A placebo is a look-alike treatment with no active ingredient, so the effect of a treatment can be told apart from the effect of expecting one. And blinding keeps people from knowing who got the treatment, so expectations can't tilt the measurements.
Here's an illustrative trial. 120 adults with colds are randomly assigned, 60 to the supplement and 60 to a placebo pill that looks the same. Neither the adults nor the staff checking their symptoms know who got which. Within a week, 51 of the 60 on the supplement feel better. That's 85 percent. And 51 of the 60 on the placebo feel better too. That's also 85 percent.
Pause and try this. Both groups recovered at 85 percent. What do the ad's 12 testimonials show?
Ask what happened to the people on the placebo.
Here's the answer. They show that colds end, not that the supplement ends them. The 51 placebo takers who recovered could have written testimonials too.
No comparison, no conclusion.
4. Choose the best explanation
You come home and the kitchen floor is wet near the fridge. Did the dog knock over its bowl? Did the freezer defrost? Is a pipe leaking? The bowl is full, the freezer is cold, and the drip comes from under the sink. You settle on the pipe, because it explains what you see.
That move is called inference to the best explanation: reasoning from the evidence to the hypothesis that would best explain it. You judge rivals by four criteria: fit with the evidence, simplicity, scope, and consistency with what else is known. Peirce called a related form of reasoning abduction. The modern phrase comes from Harman, in 1965, in an article called "The Inference to the Best Explanation".
Now a car that won't start. There are 3 clues: the engine only clicks, the dashboard lights are dim, and the headlights were left on overnight. The fuel gauge reads half full. There are 3 rival hypotheses: a dead battery, an empty tank, and a failed starter motor.
A dead battery explains all 3 clues and fits the gauge. A failed starter explains 1 of the 3, the clicking. An empty tank explains none of them, and the half-full gauge counts against it. So the dead battery wins on all 4 criteria.
Pause and try this. Does the dead battery's win mean the case is closed?
Think about what kind of reasoning this is.
Here's the answer. No. Inference to the best explanation is inductive. The battery is the hypothesis to test next, with a jump-start.
List the rivals, weigh the clues, then test the winner.
5. Tell a theory from a guess
A friend shrugs off a well-supported idea: "It's only a theory." In everyday talk, a theory sounds like a hunch. In science it means nearly the opposite. A scientific theory is a well-tested, broad explanation of many observations, not a guess. The ladder runs from guess, to hypothesis, to theory, and each rung is earned by a test.
Some claims borrow the look of science without the climb. That's pseudoscience: a claim presented with the look of science that lacks the testing science requires. Five warning signs help you spot it. No possible result could count against the claim. Testimonials stand in for controlled tests. Methods and data are kept secret. Persecution is claimed in place of evidence. And the claim never changes despite contrary evidence.
Back to the bracelet's ad. It says no test can measure the effect. It fills its page with customer stories. And it claims big labs want to hide the truth. That's 3 of the 5 warning signs.
Here's the careful part. The signs are a classroom tool for deciding where to look harder. They're signs to investigate, not proofs.
Pause and try this. The bracelet's ad shows 3 of the 5 signs. Does that prove the bracelet fails?
Hold on to the word "signs".
Here's the answer. No. Three signs are a strong reason to investigate, not a proof that the bracelet fails.
Guess, hypothesis, theory: each rung is earned by a test.
6. Ask who carries the burden
Back to the haunted dorm. You ask your friend how they know. "Prove it isn't." Notice the flip. They made the claim, and now you're asked to disprove it.
Here's the passage. "No one has ever proven that the old dorm is not haunted, so it must be haunted." The only support offered is a lack of disproof. That's the appeal to ignorance, or ad ignorantiam: concluding that a claim is true because no one has shown it false. Locke named that argument in 1690.
The cure is the burden of proof, the duty of the person who makes a claim to offer support for it. A lack of disproof doesn't shift that duty to the listener. With this passage, all 20 passages in the course's fallacy set have been keyed.
Every fallacy here has a reasonable neighbor. An inspector checks every beam in a building for termite damage and finds none. The search would likely have found damage if it were there, so the conclusion is reasonable, though not certain.
Pause and try this. What does the inspector have that the haunted-dorm passage lacks?
Ask where each one looked.
Here's the answer. A careful search. The inspector looked where the evidence would be; the dorm passage names no search at all.
Whoever claims it carries it.
7. Face the deep question
Every morning you expect the floor to hold and your coffee to taste like coffee. You've never checked tomorrow. You expect it to resemble the past because the past has been reliable. But why should the past be a guide to the future?
That's the problem of induction, Hume's question of what justifies expecting the future to resemble the past. Hume raised it in A Treatise of Human Nature, in 1739 to 1740, and again, more briefly, in An Enquiry Concerning Human Understanding, in 1748.
Look back at this lecture. The passed test with the smoke alarm leans on that expectation. So does expecting other adults to respond as the trial's adults did. So does trusting the dead battery, because dead batteries have caused such clues before. Every inductive step leans on it.
Hume's question doesn't tell you to stop. Philosophers still disagree about the best reply, and this course takes no side. What it does is explain why inductive conclusions are never guaranteed.
Pause and try this. If induction can't be proven, is one anecdote as good as a controlled trial?
Think about what the question asks.
Here's the answer. No. Hume asks why induction works at all, not whether some inductive arguments beat others. A controlled trial still beats one story.
Strong arguments still beat weak ones.
8. One science check
You now have four questions, and they fit on an index card. What does the claim predict, and could any observation show it wrong? Compared with what: was there a control group, a placebo, and blinding? What are the rival explanations, and which fits best? And who carries the burden of proof?
Run them on this week's claims. The bracelet's claim is untestable as stated, and its ad shows 3 of the 5 warning signs. The trial's 85 percent in both groups shows the supplement did no better than a placebo. The dead battery is the best of 3 rival explanations of the car's 3 clues. And the haunted-dorm passage commits the appeal to ignorance.
Notice what the check doesn't do. It doesn't show that no cold remedy helps or that the dorm isn't haunted. It shows how much support each claim has so far.
Pause and try this. The bracelet was never tested. Is the honest verdict "false" or "unsupported"?
Ask what the evidence actually shows.
Here's the answer. Unsupported. No test was run, so there's no evidence either way, and the seller blocks the test.
Predict, compare, rival, burden: four questions, one card.
Closing
Let's pull it together. A testable claim could be shown wrong by some possible observation. A failed prediction refutes by modus tollens, while a passed one only supports. A control group, a placebo, and blinding turn stories into a test, and the trial's 85 percent in both groups showed the supplement did no better than a placebo. The best explanation wins on fit, simplicity, scope, and consistency, and then gets tested. A theory is well tested, and the warning signs of pseudoscience call for a closer look. A lack of disproof isn't proof. And Hume's question sits under every inductive step.
Next in this module come the slide deck, the Readings and Resources page, the lecture tutorial, the practice set, and the quiz. Next week, in Bias and Decision-Making, you turn these tests on your own reasoning.
Built with The Course Maker · samples.thecoursemaker.com