Quantcast

«

»

Dec 21 2009

Atheist Meme of the Day: Unfalsifiability

Scarlet letter Today’s Atheist Meme of the Day, from my Facebook page. Pass this on; or don’t; or edit it as you see fit; or make up your own. Enjoy!

Most religious beliefs are unfalsifiable — there’s no possible way to test whether or not they’re true. And unfalsifiable hypotheses should be rejected on that basis alone. If there’s no way to test whether a religion is true, how do you distinguish it from the thousands of other untestable religious beliefs? Pass it on: if we say it enough times to enough people, it may get across.

11 comments

Skip to comment form

  1. 1
    Paul Wright

    How would you falsify the hypothesis that “unfalsifiable hypotheses should be rejected”?
    Which isn’t to say that falsifiability isn’t a desireable quality, of course.

  2. 2
    Greta Christina

    How would you falsify the hypothesis that “unfalsifiable hypotheses should be rejected”?

    Good question. I’ll take a stab: If unfalsifiable hypotheses could be found to be useful in some way; if they could be found to serve some function that can’t be served by falsifiable hypothesis. That would falsify the hypothesis that “unfalsifiable hypotheses should be rejected”.
    Thoughts?

  3. 3
    Chris H

    Good answer, Greta. Even the question made my brain hurt. It’s probably a result of watching too much porn.

  4. 4
    Joshua Zelinsky

    I disagree with this. First of all, many unfalsifiable hypotheses are as unfalsifiable as their negations. Thus, “there does not exist an undetectable dragon in my garage” is as unfalsifiable as “there does exist an undetectable dragon in my garage.” So a simple basis of fasifiability is not sufficient to determine whether a hypothesis should be rejected. Falsifiability is potentially a useful test to determine if a hypothesis is scientific. It might also be a useful test for determining if a question is worth thinking about. But without a lot of other machinery, such as Hume’s and Occam’s razos, simple falsifiability isn’t sufficient to reject a hypothesis.
    Moreover, falsifiability isn’t even a great test for what is scientific. While this test works for naive Popperianism, as pointed out by Quine, Kuhn and others we rarely reject a treasured hypothesis when an experiment provides evidence against it. Indeed, we often posit additional defensive hypotheses to preserve the statement. As Quine observed, one can preserve a hypothesis indefinitely by adding enough defensive hypotheses. Even more troubling, sometimes this is the correct response and sometimes it isn’t. For example, to preserve the Ptolemaic system people kept adding additional epicycles to the model of the solar system. However, this turned out to be bad. On the other hand, the neutrino was posited to exist to preserve various hypothesized symmetry laws. So it isn’t even easy to tell when one is making legitimate or illegitimate defensive hypotheses.

  5. 5
    Colin

    @Joshua: Agreed. If most religious beliefs are unfalsifiable, then so is atheism, as long as no god literally presents itself to humanity. A better criterion, following the switch from the (both technically correct) Ptolemaic to Copernican model, would be the elegance of starting premises, though of course that has met its own bugbears on the religious/atheistic front.

  6. 6
    DSimon

    Axiom: Use the simplest model that explains the evidence and makes correct predictions, because simpler models are easier to deal with.
    Working from that assumption, unfalsifiable hypotheses are a bad idea because there’s no way of telling if their inclusion in your model is making it unnecessarily complex, because there’s no prediction that can invalidate them and no evidence that can disprove them.
    Joshua, that takes care of the “endless extension” problem, since extensions (that is, complications) to the model have to be justified by additional predictive power.
    It also, I think, takes care of the “negation is unfalsifiable too” problem, since it’s not helpful to have such negations in the model and therefore we can just ignore them. For example, the theory of gravity is “The force of gravity is proportional to the mass over the square of the distance”, end of line. It’s an unnecessary complication to make it “The force of gravity is proportional to the mass over the square of the distance, and has nothing to do with invisible pink unicorns or shoe-repairing elves or caramel popcorn”.

  7. 7
    Greta Christina

    Joshua: I don’t agree. It’s true that “There is absolutely, positively, with 100% certainty no undetectable dragon in my garage” is unfalsifiable… but that’s not an important, interesting, or useful statement. The important and useful statement is, “There is no reason to think that there’s an undetectable dragon in my garage, and in any practical sense a completely undetectable dragon is no different from no dragon at all — so I’m going to act as if it isn’t there.” Which is what atheism is: it’s the null hypothesis.
    As for the process of science: It’s true that as a practical matter, one experiment contradicting a hypothesis will rarely overturn it — especially if the hypothesis is an important one. But I would argue that this doesn’t contradict the principle of falsifiability. I would argue that that’s as it should be — even with the principle of falsifiability in place.
    As a science teacher of mine once pointed out: “If one of my students announced that they had discovered that the atomic weights of helium and lead were the same, I wouldn’t immediately alert the scientific journals. I would first ask if my student had turned on their scale.” If a hypothesis has proven extremely powerful in the past, a handful of experiments shouldn’t necessarily be enough to overturn it. Experiments can be wrong: they can have poor methodology, etc. (Despite the joke, it would probably take more than one fossilized rabbit in the pre-Cambrian layer to overturn the theory of evolution.) It should take time, and extraordinary care and rigor, for science to adopt radical new hypotheses.
    And sometimes the appropriate response to experiments contradicting a useful hypothesis is to modify the hypothesis rather than discard it. Experiments didn’t eradicate Newton’s laws of motion and replace them with Einstein’s — they showed Newton’s laws of motion to be an entirely useful approximation of Einstein’s when applies to the special case of slow-moving objects. The theory of relativity wasn’t a reversal of Newton’s laws — it was a modification of them. And that’s appropriate.

  8. 8
    Greta Christina

    If most religious beliefs are unfalsifiable, then so is atheism, as long as no god literally presents itself to humanity.

    Colin: “As long as no god literally presents itself to humanity”? But that’s the whole point. Of course atheism is falsifiable. If a god or gods presented themselves to humanity, atheism would be falsified.
    The principle of falsifiability doesn’t mean a hypothesis has to be falsifiable with the currently available evidence. That wouldn’t make a hypothesis falsifiable — it would make it falsified. The principle means that it has to be possible, in theory, for some evidence to arise that would prove the hypothesis wrong.
    There are, of course, limits to the principle of falsifiability. The main one is that you always have to start with some axioms. (Most commonly, we start with the axioms that our perceptions are not hallucinations, we’re not living in the Matrix, etc.) But the fact that no current evidence has contradicted atheism doesn’t make it unfalsifiable. It makes it a really good, strong, plausible hypothesis.

  9. 9
    Joshua Zelinsky

    Greta no disagreement there. But the point is that you are using other notions such as what constitutes a reasonable null hypothesis (implicitly using Occam’s razor). Thus a claim that “unfalsifiable hypotheses should be rejected on that basis alone” isn’t accurate.

  10. 10
    Eclectic

    Greta: It is not necessary to assume we are not living in the Matrix.
    A useful hypothesis ultimately predicts (subjective) observations. A good hypothesis predicts them consistently and accurately. A bad one has a lot of error.
    Unfalsifiable nonsense like deism or last thursdayism doesn’t predict observations at all and may therefore be dismissed without wasting further brainpower on it.
    Two hypotheses that make exactly the same predictions (e.g. the universe is actually old, vs. the universe was created last thursday with a perfect illusion of age) are actually the same hypothesis. Only when hypotheses make conflicting predictions is the statement that there is a difference falsifiable, and therefore itself a hypothesis worthy of consideration.
    So as to whether the universe is actually old or just a perfect illusion, I don’t care. It will never (because we are assuming the illusion is perfect) make the tiniest little difference to anything that ever happens.
    Occam’s razor was created as a technique for choosing the most plausible hypothesis given incomplete information. But it’s also a good rule of thumb for choosing among equivalent hypotheses: when they will all produce the same predictions in the end, choose the one that involves the least mental effort.
    Joshua: unfalsifiable hypotheses absolutely should be rejected on that basis alone. They aren’t real hypotheses, but rather daydreams with pretensions. Once you’ve considered them long enough to understand that they are unfalsifiable, you’re done. There is nothing whatsoever to be gained from considering them any further.
    It’s far more fruitful to discuss whether pigs have wings.

  11. 11
    Eclectic

    Greta: Let’s say that it would take only one fossilized rabbit in the pre-cambrian to overturn evolution as currently understood, but more than one rabbit in pre-cambrian rock.
    We’d look good and hard to see if there were alternative ways for fossils to get into older rock, or of the rock was misidentified, or if the rabbit was misidentified, or if the whole thing was due to some stupid error, or if it was a practical joke, or…
    Very few observations are supported completely by a single hypothesis. You usually have one dubious one being tested, and a large number of “safe” hypotheses that are assumed to be reliable. (E.g. measuring instruments working properly, readings recorded properly, etc.)
    But if you observe something really unexpected, you start looking for the least implausible thing to explain the observation, which may be one of your “safe” assumptions.

Leave a Reply