The fundamental problem with social sciences is it is far more difficult or impossible to isolate factors in an experiment as you would in a laboratory setting. Therefore the level of certitude that you can attain is far lower than that of an experiment in the physical sciences.

Take the development of a democratic society in various countries. You have an incredibly low number of samples or cases from history versus an incredibly large number of factors - or nuisance variables, if you're looking to investigate certain factors in particular. This makes it impossible to look beyond simply values such as correlations to more complex analysis because you will be getting negative degrees of freedom for the error term in your experiment. Yes if you drastically simplify the experiment by removing factors it will become possible, but you would be assuming away factors. In many cases such as the one mentioned here, there is simply no way to replicate it as in a laboratory experiment either.

That doesn't make it a useless science, although really Feynman never implied this, he simply suggested it was not as rigorous as he would have liked it to be.

Yes, taking economics as an example, it is founded on simplifying assumptions such as rational selfishness and yes, exceptions are constantly being found. When prices go up demand goes down unless it's a Veblen or Giffen good. Yes, markets will generally adjust to supply and demand automatically and price accordingly but wages and prices are generally sticky when moving donwards as workers tend to be unwilling to accept pay drops, and price adjustments are generally staggered and made relative to competitors rather than on the spot.

There's certainly a great deal of uncertainty, but surely this is preferably to being in ignorance over how to best stimulate an economy in recession or what factors contribute to a democratic society. The key point is to take this uncertainty on board when you apply it.