> it creates false confidence in results it produces in domains it doesn't fit

I don't think that's correct, people simply love to extrapolate beyond the evidence. You do a study where you take 6 grad student and tell them to push a button that sends electricity (supposedly) to a subject. You measure their behavior in that context. Okay, you've got a pretty small sample of something.

Now here comes the non-scientific part, a professor looking for fame and glory, a bunch of news site looking for clicks and views, and you get an article: "Science proves that people are willing to hurt others on command if told it is for an experiment."

See, the thing is that science is about slowly acquiring more knowledge and understanding and updating our mental models and understanding of things as more information shows up.

Here the experiment wasn't useless, it tells us one small extra bit of info we didn't have before. Doing experiments, testing hypothesis, those are still valuable. Extrapolating the result of those to mean more than they demonstrate is just a follow up hypothesis that then itself needs an experiment, etc.

In which domains do you believe it's ineffective?

What alternative methods do you suggest for the "domains it doesn't fit"?

The obvious answer would a philosophical approach. And an example of a domain it doesn't fit would be ethics.

I think it can very well fit ethics.

A lot of ethics argues that acting in such and such "ethical" way will result is some outcome, like more happiness, less harm to the environment, less harm to workers, less injustice, etc.

It would be great to test those assertions empirically as well. If it is true, devise an experiment, implement your ethical framework, measure outcomes, validate it, prove that it does improve on your claimed metrics.

Just to be clear, there is a place for thinking through first principles in science, that's part of the hypothesis phase. In that phase you can use inductive and abductive reasoning to guide yourself to a likely hypothesis. But science will want you to go further and test the hypothesis, to ascertain more strongly that it is a likely hypothesis.

The only part which isn't really applicable to science is deductive reasoning, because it relies on assuming truths and perfect formal logical rules of derivation. These can hold within a total imaginary setting, mathematics does this all the time for example, and so does a lot of philosophy, but when applied practically (which is where science comes in), you need to show that those assumptions hold in the real world and its current setting, and that the deduction rules do as well, and that brings you back to inductive and abductive reasoning, because you no longer control the range of the domain, it is now dictated by reality or our perception of it.

When I studied philosophy, we applied scientific methodology to understand complex topics, categorising them, and evaluating them against each other. The theories of moral philosophy are under scrutiny in a way very similar to the (other) natural sciences.

That works to inform discussion. Having access to the modern knowledge of how the nervous system powers muscle movements helps elevate discussions around the intent to act and the deterministic natures of that intent well beyond what folks digesting the topic in the 1500s could comprehend. But there's a really hard limit with how far it can get you.

Provable things often point us toward certain answers - but provable science has historically shown us that it isn't afraid to pull the rug out from under us and throw a curveball. Physics is highly deterministic and ordered - extremely so - under you get subatomic - then everything goes out the window.

Philosophy obviously needs to not go directly against things we have a high confidence in - but there's still a lot of grey area in there especially where the philosophy of motion and ethics are involved.

One helpful thing would be to separate the hard sciences from the soft sciences. Calculating the mass of a hydrogen atom is fundamentally different than determining which factors and how much of each are required to diagnose ADHD. The latter is very important and it deserves an important category that clearly distinguishes it from the former in the common language.

I'm not suggesting any, I'm saying it's better to make decisions knowing that you don't know than acting like you can model the system.

I don't really have a problem with people doing this kind of research either, I'm bothered when they try to use these models to prescribe behavior.