It always fascinates me that often these "heads in the clouds" type stuff has an enormous impact on history, politics, and other disciplines while what is supposed to be more practical often is ignored or belittled. The most plausible reason is that we have firmer convictions in the little things rather than the large things. This seems to reveal the contemporary, and speaking in philosophical lingo, "modern" tendency to place epistemology as prior to metaphysics. The exception to this thought is the field of ethics. Logic too has been explored in depths through computer science, but this is still -- historically speaking -- in its infancy. The difference between formal logic and computer science is also a gap that cannot be underestimated.
Keep your audience in mind. This is Hacker News. If this were a website for lawyers, political scientists, and historians I would have given a different answer to the question of "What, exactly, do philosophers do?" For them, the language of thought and issues at the foundations of mathematics qualify as "heads in the clouds" type stuff.
I would have also given a different answer if the primary group here were, say, physicists, astronomers, and chemists. For them, the answers I would have given to the lawyers, political scientists, and historians would have likewise been "heads in the clouds" type stuff.
Philosophy has unseen tendrils in almost everything. You can only get someone to see it (or, really, care about it) when it's something they're already interested in.
HN isn't particularly philosophy-friendly. I've talked about how philosophy was crucial in laying down the philosophy of science and working out the scientific method, only to be drowned in a sea of "wasn't necessary, it's self-evident" sorts of comments.
It's funny because it took philosophy centuries to see that—and even now, the scientific method is showing limitations; its too difficult to perform in the social sciences and the less rigorous method of extracting trends off big data also often has great value.
"Trends off big data" can be very useful in a machine-learning sense, if you want to predict very accurately, but you're ok with being far off when you do make a mistake. They're not very useful at all for designing interventions, which in the end is more what we care about.
