Is there any particular evidence or even just an example that demonstrates the blogposts thesis? It’s seems to take as given that you agree with the premise, but I’m not sure that’s something you can take as given.

I’ll say that ML/“”””AI”””” has the opposite problem - no one wants to do theory, because it’s hard. For empirical studies, you can brute force them with slave labor (aka undergrads and phd students).

I am not sure why the author thinks science is stagnating — it is hard to measure progress, and even if we were stagnating, it is possible that many fields have matured and now have to tackle extremely difficult problems. (I guess in complexity theory, we would want to show that P is not NP, although I think very few people actually work directly on this problem.) There are many people working on more empirical questions in all of the natural sciences, for example. There are fewer theorists in these departments than experimentalists, and I have the feeling that the proportion has not favored theorists more in recent years.

It is true that academia tends to work on more "fundamental" or "theoretical" questions, but this also seems acceptable — industry has the financial power and incentives to work on problems that are closer to having impact, like "cheaper rockets, cures for cancer, [and] software that is efficient." And in the modern day, we do see many companies working on these problems. There is not enough money or people in academia to successfully compete with industry in these areas, so they work on problems that have longer timescales. When they do see translational application, those who work in academia often partner with their industry counterparts to make it happen (this is common in software and in pharmaceutical applications).

I do agree that empirical work is often undervalued by people who work on more theoretical topics, and they do not necessarily have the same respect for their creativity and intellectual capability as they would for other theorists. However, as a person who works on computer science theory, most of my colleagues recognize that working on theory, versus, say, compiler design or machine learning, just require different skillsets. I think the point about college courses not suiting their audiences needs is an important point, but is mostly a separate discussion.

Huh. I think the other commenters have provided sufficient challenge to the substance of the article. I would agree this is untrue about science more generally; empiricism is alive and well in physics, biology, chemistry, etc.

The author seems to care about fast software, which is an interesting field for this discussion. I only briefly studied speed of software in undergrad, but from what I remember there was a lot of emphasis on asymptotics, theoretical optimization of problems, etc. It’s likely true that theoretical computer science is drifting farther and farther from empiricism.

I don’t have any opinions on if this is good or bad. @localizer correctly mentions that fast software is a priority in industry, which can allocate resources to the problem more efficiently by virtue of financial incentive. Seeing that allocation not prioritized in academia disappoints the author. I wonder if focusing on the direction of the community as a whole is misguided. As a respected professor, the author has enough career freedom to do what he wants to do; it’s a valuable position to have.

I'm sorry, but this is not a serious dichotomy, let alone article. It stinks of "philosophy is just a bunch of eggheads building a peerless stack of meaningless papers to look down on everyone else from, who needs it?"

Theory is a strictly necessary component of finding empirical evidence. Without the right questions, one can construe almost anything from the sheer amount of information that exists in the world. There are so many people on the planet that medical researchers are publishing meaningful descriptions of genetic disorders with only dozens of cases. That's an incidence rate on the order of impacting 0.00000001% of the world. In IT, datacenters broadly refuse to use RAM that doesn't have algorithms for error-correction, because of the inevitability of those submillimeter 1s and 0s being struck by cosmic rays.

The p-hacking crisis in psychology (really, all fields to some extent)? That data is showing something, those researchers are almost certainly wholehearted believers in their fraudulently-supported conclusions. Why? The universe, wait 'til you hear this: is BIG. Humanity has been collaboratively conducting rigorous investigation of it for a tiny fraction of our existence, which has itself only spanned the temporal equivalent of a mote of dust that landed on a parking garage mural, and we're still screwing up the fundamentals of science daily, trying our best.

Ask the Big Data firms how accurately they can measure anything that wasn't designed to go on a spreadsheet. Ask the CIA, or the FSB or any given intelligence agency, how confident they are in predicting what will happen in the tiny, entirely comprehensible and human-driven, extremely constrained domain of geopolitics. Geopolitics, the things done by collective action of billions of more-or-less rational humans, not the shapes of particles too small to have shapes, not the currents that lead to the waves cresting on the ocean, not the impact of wind turbines on the migration patterns of birds, not the intelligence of fishes, not the questions of supersymmetry or the implications of black holes or how to handle the fact that we still don't know how those problems can coexist coherently.

Actually, don't ask the intelligence agencies these things, they're stuck in the same paradigm that this article is: that to understand is simply a matter of scrying the infinite tea leaves of Information. That all which stands in the way of escaping war, or death, or bad sales next quarter, is that ours is not the eye of God. For this strain of thought, all that defines deity is to know all. But we were not there when He struck down Behemoth, when He wrestled with Ishtar, when the first Cosmic Turtle wondered "Upon whose back do I stand?" We are monkeys who still don't know how to share, who don't say thank you, who don't accept our own damn place. Hell, we barely know our own address.

That doesn't mean this piece is fueled purely by hubris. Of course, we are the monkeys who ventured out of the forested valleys of what we call Tanzania, who crossed the savannahs and deserts and seas, who braved lions and tigers and tamed wolves and aurochs and drove countless living tanks the size of our houses to extinction, who watched the stars and learned to call them Grandma and Friend-of-my-Heart, who shaped those into figures and those constellations into the maps by which to navigate these treacherous plains which we today take for granted to be our domain. If we have a place, to explore and understand is it.

The frustration of seeing countless hypotheses and paradigms accumulate without testing, of watching a loved one pass only for a cure to be found in what amount to the days following, of seeing Robin Redbreast leave your tree for good when his song had always arrived in your dreams before the cresting of dawn ushered in the morning when, if you just knew what was insufficient now that had not been, you would uproot every ounce of order in your life to bring him back, just to hear that tune again? That pain is unceasing, and the very river by which the drive to discover is turned. Another spin will not ease it, this is our burden.

So what is to be done here, what should Daniel Lemire bemoan instead of the instinct to stew and digest that which they wish we would preoccupy our attention with gathering? The scarcity of resources, funding. Theory is cheap, people do it constantly for free, and plenty will spend 20 years preparing and live out the remainder of their lives all-but-drowning in debt just to be offered a subsistence wage and told their conclusions are worth considering by others. Laboratories and materials don't provide themselves. Someone has to chop down the forests, and unearth the mummified corpses of algal colonies, and distill and part and reconfigure those into forms sterile enough to "wipe clean" and set the stage to play-act the questions those thinkers are refining and compiling into ideas. Those people play-acting the scenarios need to fund their lives so they can watch the tar pitch drop, or invent whole new fields of science and build industries just to have a webcam film it.

Only for the arrogance of the observer to make that collected information worthless by changing it to fit publishable standards. Anon, anon, unto oblivion.