[gnulinux]

I strongly disagree with this view. I sympathize with you that it's probably true that scientists are inclined to cargo-cult their science after physics. But the endeavor here is not wrong, so even if it's cargo-cult, it's good.

Science's primary objective is to find models. A model should do predictions, and then scientist should collect data of interest, and make sure the data isn't falsified by the model's predictions. Any scientist who cannot do predictions, and verify that data doesn't contradict predictions, is no scientist at all.

Once you have models, it's simply too tempting to formalize them and build mathematical theories for them. If nothing, for computational advantage, so you can make computers make predictions, this way you can eliminate human errors. So, it seems like any science will eventually build models that can generate predictions from first principles.

It is one thing to claim the entire human history can be predicted from first principles of a theory X. Clearly, we have no such X. Maybe we never will. It's another, and totally reasonable thing, to build a theory X from first principles that correctly predicts some data.

[jdm2212]

No one has ever built a historical model that predicts the future with any accuracy. There are plenty of overfit models that "predict" the past, but none hold up on new events.

Interesting book on the topic from the world's leading researcher, if you're curious: https://www.amazon.com/Superforecasting-Art-Science-Predicti...

[drdeca]

Does Moore's law count as a historical model?

Perhaps it is too recent and limited in scope,

but it is about, like, some measurable things about society, which is based on past observations, and it has been somewhat predictive? (though, perhaps it being somewhat predictive has been in part due to it becoming somewhat prescriptive?)

[madhadron]

Moore's law is more of an economic necessity. Circuits get cheaper to run, cheaper to make, and faster the more you shrink them, so you just keep shrinking them. This was observed by Carver Mead way back in the middle of the 20th century.

Moore's law requires an average of 3% increase a month in the number of transistors packed on a chip. With a few hundred thousand people working on the problem, that's not crazy. Of course it's going to be spikey, but combined over the lifetime of a chip project...

[jdm2212]

I don't know that Moore's Law is what most people would mean by a historical law, but it has held up surprisingly well even for a non-historical law :)

[gnulinux]

Just because it has not been done before, doesn't mean trying to do so (i.e. make a science of history) is invalid or just cargo culting. Especially since GP referred to other social sciences, and named neuroscience. It has not been done for history, maybe it never will be. I don't think generalizing this to all social sciences holds up. Surely there are some falsifiable predictions of psychology, sociology etc that has been tested.

[gmtx725]

The social sciences don't work like the hard sciences though. And the primary reason for this is because humans, unlike say, atoms or molecules, are not independent objects. They are subjects with agency and can react and change their behaviour in response to what your models predict about them. Essentially predictions can become self-fulfilling prophecies.

[madhadron]

> I sympathize with you that it's probably true that scientists are inclined to cargo-cult their science after physics. But the endeavor here is not wrong, so even if it's cargo-cult, it's good.

Having been trained as a physicist and then worked as a biologist and dabbled in history, it isn't good. Physics cannot be used as the ideal of sciences because it is an extremely strange science. Consider: physics deal with simple systems with only a few essential observables that can be repeatably measured. History deals with very complicated systems with an inordinate number of possible observables none of which can be repeatably measured. Why would you expect them to resemble each other? I actually wrote a book about this...

> Science's primary objective is to find models. A model should do predictions, and then scientist should collect data of interest, and make sure the data isn't falsified by the model's predictions. Any scientist who cannot do predictions, and verify that data doesn't contradict predictions, is no scientist at all.

A model is supposed to recapitulate observations. Prediction is only relevant in the case of repeatable observations or the discovery of new observables. Astrophysics and history don't get the former, but you will find prediction in both in the latter case...but once you've measured it, it's not prediction anymore.

> Once you have models, it's simply too tempting to formalize them and build mathematical theories for them.

People do this to a limited extent with toy models, but the relationships among observables in history that you can get from the historical record tend to be fairly simple and rough, while the number of observables is enormous, so formal modelling doesn't buy you much. When people have done this on a large scale, you get the Club of Rome...trying to peer a few decades into the future with an enormously complicated model.

> It's another, and totally reasonable thing, to build a theory X from first principles that correctly predicts some data.

First principles are overrated, and I think the training of physicists overemphasizes them. In condensed matter there's a phrase, "more is different." We can go from atomic descriptions to macroscopic ones...in very, very simple cases. History has no such simple cases. You always have a huge number of parameters, which leads you back to the old adage that with three parameters you can draw an elephant, and with four you can make him wag his tail. There's too much slop for a model from first principles predicting things in history to be interesting.