The author makes some good points, but does not propose his own criteria for what distinguishes a good theory from a bad theory, which seems like a cop-out to be frank.<p>I imagine a big appeal of falsifiability in its early days, stemmed from its opposition to religious dogma. Imagine dealing with someone whose answer to every question comes back to God. <i>"Why did the patient die?"</i> <i>"Because God decreed it."</i> Talking about germ-theory must have been a massive improvement over theology, hence why ideas like falsifiability must have arisen to provoke people to think in more critical ways.<p>My own standard for evaluating theories comes down to the following: The <i>simplest</i> explanation that best <i>fits existing data</i>, and makes <i>useful predictions</i> for the future.<p><i>"Simplest"</i> because we need to prevent overfitting.<p><i>"Fits existing data"</i> because the whole point of a theory is to explain what we see around us.<p>And "<i>useful predictions</i>" because it will actually provide practical benefit to us. As opposed to a theory like <i>"God decreed it"</i>, which may be extremely simple and error-free, but provides us with no practical benefit at all.<p>As I see it, falsifiability is a side-effect of the second and third requirements. If a theory is making useful predictions, we might as well check to see if it's accurate. And if it's not, there's no point in holding on to a theory that's making faulty predictions. Falsifiability, in this perspective, isn't a requirement in and of itself - it's a natural consequence of the requirement that theories be useful and fit any data at hand.
This article attacks falfsificationism in part on the basis that scientists don’t seem to actually behave according to “naive falsficationism” in which you supposedly should immediately throw out a theory as soon as you get some conflicting empirical data.<p>The article notes that collecting empirical data to falsify a theory depends on auxiliary theories (eg testing the idea that smoke causes cancer in rats depends on our ability to precisely define and identify when cancer is present) and so most scientists will question these auxiliaries rather than the main theory if conflicting data comes out. This is presented as evidence that scientists don’t actually practice falsficationism.<p>This whole discussion completely ignores uncertainty and the fact that theories describe not only relationships between variables but also involve parameters that need to be estimated.<p>If I fill a rat cage with smoke and the rats don’t get cancer, I don’t immediately assume all previous studies were wrong. This isn’t ignoring my duty to falsify, it’s realizing that the theory that smoking causes cancer describes a causal relationship between variables (smoking -> cancer) it also implicitly or explicitly depends on parameters (in this case, how much and how long a rat must be exposed to smoke to cause cancer at some incidence rate). Given that many other studies found the causal relationship, it is more likely that my experiment messed up the parameters of the causal model (eg I didn’t administer smoke long enough) rather than messed up the causal relationship. And parameter estimation is subject to uncertainty.<p>Now if I’m absolutely sure I’m getting all my model parameters correct, and I can only be confident if other people get the same results. Then I’m moving toward falsifying the theory.<p>The point is that science involves positing theories, and theories are descriptions of causal relationships and generally involve parameters. Much of science is just estimating model parameters given that we assume some model to be tentatively true. You assume your model to be tentatively true, eg smoking causes cancer, now you need to estimate the effect size which is a parameter problem.
This article attacks a straw man. Being falsifiable is necessary but it is <i>not</i> sufficient to be considered a scientific hypothesis. Such a hypothesis also, and more importantly, has to provide a better <i>explanation</i> of some phenomenon than the current best theory. Experimental evidence is only brought to bear to decide among plausible alternative theories after the vast majority of candidate theories have been eliminated for not providing good explanations.<p>It's easy to see that this must be true because we can only ever have a finite amount of data, and that will always be consistent with an infinite number of falsifiable theories (c.f. Russell's teapot). So data cannot possibly help us choose from among those.
If there's no conceivable experiment that could contradict your claim then it's meaningless. Just because there might be other ways for a claim to be vacuous doesn't mean the necessary condition above doesn't hold.
> This means that a theory is never falsifiable simpliciter, but only relative to a set of background assumptions. Therefore, if we say that a theory is only scientific if it’s falsifiable, then it follows that no theory, not even a theory as successful as Newton’s law of universal gravitation, is scientific. Of course, this is absurd, so falsificationism is false.<p>You can consider the union of the theory and background assumptions to be a single theory. The premises do not imply that no theory is scientific.<p>An analogy to mathematics is pretty apt. The Riemann hypothesis (RH) is falsifiable in the crude sense described, but cannot be proven or disproven without an axiomatic basis. If you were able to derive a contradiction from RH and some other accepted theorems, you would think that falsifies RH. However, if you wanted a more concrete picture of what was proved, you might reduce the other theorems to Zermelo–Fraenkel set theory (ZF) so you have the result that RH + ZF is inconsistent/false. Maybe in some other axiomatization RH is true, but the general community accepts ZF so that means RH + ZF is false effectively means RH is false.<p>I do agree with some of the sentiment of the article. There should be thought about how to actually test what we say we are testing (not having a reliance on background assumptions) and to accept negative results rather than just find something to blame.
What is the point of this article? The content is interesting: falsification is more complicated than it seems at first, and scientists don't practice philosophically-pure falsificationism. But it's wrapped up in some kind of finger-pointing "the emperor has no clothes!" type of exposé format that doesn't make sense and doesn't lead to a satisfying conclusion.