Why can social sciences apply hard science concepts to other things than what they apply to?
E.g. suggesting that social groups can be modeled using thermodynamics:
http://www.eoht.info/page/Sociological+thermodynamics
To me this makes very little sense and I don't understand what could be the usefulness of such.
What's the usefulness of such theories and why are soc. scientists allowed to make such? Can they prove that the outcomes of the theories?
First, and I know this is somewhat subjective, I dislike the distinction between "soft" and "hard" science. When it boils down to it, all science is "soft" because at best we can falsify our theories. The issue with many social sciences is that it is hard to construct formal experiments or repeat an experiment, but we can still collect data to build more robust theories and to knock down old ones.
Now, as for application of hard sciences like physics and chemistry to "soft sciences" like psychology and sociology, the answer is fairly simple. It is because there is a sort of hierarchy from more foundational sciences to more derived sciences.
Physics drives chemical reactions. The reasons why certain molecules form, energy is released or absorbed during reactions, etc is all a matter of the physics of atoms. Biology is driven by chemical reactions. Thought is a manifestation, as best as we can tell, of the brain, a biological component of the body. With humans, behavior of individuals, couples together in social interactions in a meaningful way, and so our psychology and biology drive culture, group behavior, and so on.
Basically, while we might be able to understand the "what," such as "what do humans do when they interact" without these more foundational sciences, we cannot really understand the how or the why unless we look at the mechanisms which underlie them. So rather than being silly that "soft science" uses "hard science," it is necessary.
Example
This connection does not mean that one can easily throw any theory from physics at more derived fields of study. However, there are cases where there is a direct enough connection. For instance, evolution is essentially a stochastic process, modulated by environment and energy dynamics. There are fairly direct theories connecting thermodynamics to biology. Specifically, Jeremy England has proposed that looking at entropy within an open system (usually entropy is thought of in terms of closed systems) within an energy bath, it seems that "life" is really natural consequence of thermodynamics (Quantum Magazine).
So my take on this is that it's not about science (soft, hard, or otherwise) and more about mathematics.
I study physics myself as a hobbyist, and I can remember sitting in a concert a couple of years ago where the mosh pit was deep and narrow. Looking at the concentration of people in it, and running some rough calculations in my head, it became obvious that the distribution of people in the mosh pit conformed to Boyle's Law.
Does that mean concert mosh pits share the same scientific principles as the pressure of liquids at depth? Of course not. What it does mean is that there's a correlation in the math used to describe both scenarios.
It's important to note here that correlation is not equal to causation. Just because two concepts can be described using the same math, it doesn't make them the same process, or even related.
This happens in statistical and data analysis all the time. When you get right down to it, the math we use to detect insurance fraud is the same math we use to highlight people at high risk in hospitals, which emergency workers are more at risk of suicide, and which offer you should be made by a phone company when you call to renew your account.
There's a very simple reason for this; the math doesn't take into account meaning. That is something only we can supply, which in turn makes the same math against different data (or even applied against different fields in the same data) carry a VERY different meaning or solution.
To that end; AFAIK there are no social scientists out there actually saying that social behaviour and thermodynamics have the same root drivers; what they're saying is that the two concepts can be described using the same math and that as you apply more of the math used in thermodynamics, it continues to describe social behaviour. Part of this is by definition interpretation; the rest is a classic example of the diversity of mathematics as a field of study in itself.
Thermodynamics is an unusually direct application of mathematics to physics. It is entirely a structure that satisfies a given set of axioms. Any other field where the constituents meet the same axioms is an application of this same theory.
To the extent that clusters of humans share a microstate, and we can combine those statistically, we have made a sociological model that satisfies the axioms that allow thermodynamic theory to apply.
The question then is not whether the model works, but how well these axioms actually describe the situation at hand. People's states are not independent the way molecular states are: we can notice patterns as they arise and actively contribute to them or avoid them. So to the degree the statistics requires independence of interactions, the model no longer applies and will not be helpful. But in situations that are brief or confusing enough to prevent the use of intelligence like this, we have a model.
Since all models are approximate anyway, there is some range of situations where this model works.
I think the first thing to realize is that all sciences do this. Consider the ideal gas law, PV=nRT. When was the last time you saw an ideal gas? Can you name an ideal gas? The answer, of course, is no you can't. Ideal gasses don't exist. They're just an approximation of reality which creates a simpler model of how gas molecules interact (in particular, no intermolecular forces, and negligible atomic volumes). Every science uses models to model things which aren't really perfect matches for the model. In the case of the ideal gas law, it tends to behave quite well until you go to extremes (such as supersonic flow).
So likewise, the social sciences get to use simplified models. They even get to use models from other branches of science, such as thermodynamics. Why? Well the major models (like those for thermodynamics) tend to have a very small set of axioms describing the assumptions we make about the system. Thermodynamics makes very few assumptions, other than that entropy is always increasing. If you have a concept of energy and a system which exhibits entropy, you can typically apply thermodynamic models and get reasonable results.
Now soft sciences tend to be less rigorous in their falsification of hypotheses. As you suggest in your question, it's harder to prove that thermodynamics doesn't actually model a social network. But that's a general state of being for soft sciences. It's not a special feature that comes into play when they borrow models from the hard sciences, it's the way they always operate.
Hard sciences tend to be more rigorous. The most powerful model which comes to mind is modeling heat transfer as electric circuits. It seems preposterous that one could model the heat transfer from the cold outside into your house as a network of resistors, but that is indeed what they teach in school. If you look at the equations for the heat transfer, which are rigorously tested, you see that they are identical to the equations for current flow through a resistor network. So why not just use resistors? As it turns out, the way electrical engineers think about resistor networks is pretty much the simplest way of thinking about these sorts of problems, so we all pretend these thermal problems are actually electrical engineering problems!
