The world needs computational social science | Hacker News Reader

The world needs computational social science

petterhol.me

105 pointsAnon8479 comments

To clarify a bit, as this used to be my field when I was an academic...

"Computational Social Science" is a slightly cumbersome term introduced in 2009 [1] that essentially corresponds to using Data Science/Machine Learning/Statistical Mechanics/etc techniques to study questions typically addressed by social scientists. Stuff like:

- Epidemic modeling using cell phone traces

- Social Media data to analyze linguistic trends

- Wikipedia talk pages to analyze consensus formation

- etc...

A somewhat less kind definition is "Physicist cum Data Scientists doing Social Science". You can checkout some of the leaders in the field on Google Scholar [2]

[1] https://www.science.org/doi/10.1126/science.1167742

[2] https://scholar.google.com/citations?hl=en&view_op=search_au...

Maybe it's just me, but when I hear "computational social science" I first think of agent-based modeling. But yes, when I look at recent work, it's all about statistical analysis of historical data. Ugh... So why can't we just call it that? Maybe it's justified to emphasize the computational aspect because of the scale of operation these days.

Ah, yes... that's the original version of Computational Social Science before the Physicists came into the science. You're probably familiar with R. Axelrod's work [1], and J. Epstein has an excellent book on it[1] C. Cioffi-Revilla has another excellent book that tries to straddle/review both sides of the fence: [2]

[1] The Complexity of Cooperation - https://amzn.to/3ZDJ5GW

[2] Generative Social Science - https://amzn.to/3PE6gw7

[3] Introduction to Computational Social Science - https://amzn.to/46cNMtW

Maybe I'm in the minority but as a physicist it makes perfect sense to model micro-interactions to estimate emergent macro-scale patterns. What is your experience with "physicists" in the derogatory?

Three issues to distinguish here. Parent is referring to a change in usage of "computational social science" from meaning agent-based modeling to meaning statistical data analysis. This change in usage may have been driven by physicists moving into social science.

You ask about modeling based on micro-interactions, which parent mentioned but did not comment on.

You seem to ask why social scientists might be skeptical of physicists contributing to social science. My observation is that physicists tend to be extremely naive about the high dimensionality and nonlinearity of social science phenomena.

Agent-based modeling is a form of modeling which explicates micro interactions to imply macro patterns. That is the connection here.

Your comment looks like it was written by ChatGPT. It misses basic implications about agent-based modeling and has a lot of emphasis on "you seem to mean..." phrasing. Am I close, and if not, was this intentional?

It's just my natural pedantry from years of helping students clarify their research statements. :-)

I'm a physics PhD... so it's a bit of self-deprecating humor. As a wise man once said, the 10 scariest words for a social scientist are, "I'm from the Physics department, and I'm here to help".

On a slightly more serious note, Physicists (and I've certainly been guilty of this as well) aren't always humble enough to learn why a specific field does things the way they do before immediately trying to impose their own approach (Ising models all the way down!)

I have absolutely zero sense of what this actually is.

All kinds of economists estimate models on very large data using computers. Does that count or not?

What good is it specifically to “know habermas as well as feature selection” ? What problems are you solving?

Maybe “the world needs an explanation of what you mean when you say computational social science” would make a nice follow up…

I don't think it's a matter of whether a certain project falls under the CSS nomer depending on the scale of the data and methods employed, but that there is as much 'sociality' to computation as there is 'science', and that studying both social and computational aspects of contemporary society and their interactions can broach insights into the 'digital stack' modern life is folded into.

I gathered the quote is not about Habermas and feature selection per se, but a platitude about bridging theory and practice--both social and computational theories and practices. Knowing when, for instance, not to divvy up social features that may constitute a larger (digital) public, or conversly combining/throwing away features that may in fact constitute multiple publics or subject groupings, paves the way for building more accurate or generalisable computational social science models.

Why would we want such models? Perhaps to map 'hunches' about social life that were previously only in the realm of rhetoric or simply too difficult to map before the advent of large scale data collection and computation.

My gist was that the author wants the data vacuumed up by big tech to be used to continue analyzing the questions posed by social science, why do we have inequalities, why do we have class division, that sort of thing.

I already felt the name of the field was terrifying, conjuring up thoughts of famines and quotas for workers with some arbitrary trait the network hallucinated. If we bring big corp into it, it gets so much more scary.

No, that's not it. It's about using computational techniques to address questions of social science. Using big tech data is not a goal, but it will sometimes be a source of data.

I worked with a sociologist on a variety of projects which used sensors and cellphones to do computational social science. In the past, sociology was based on observing people, writing those observations down, and then thinking about them. Some statistics were gathered, but it was a very fuzzy science. Now, electronic devices and networks have made detailed big data about people available and sociologists are using it. Electronic surveys have replaced paper surveys and greatly increased scale. This happens mostly behind the scenes, such as research on the use of electronic medical records, marketing, employee motivation, etc. Take a look at her research projects here (and look up the people she worked with to see what else they are working on):

https://sph.umich.edu/faculty-profiles/anthony-denise.html

One of the things we did was build survey apps for phones which would sense activity and only pop up a question when a person wasn't doing much. The sensor data would also serve as a ground truth, telling us for example how much activity a person really did, versus how much they think they did. One early result from such research was that people think they are always doing things, rushing around, when the truth is they are sitting doing little for 95% to 99% of the day. We used the survey app to build a system to train a gait recognition machine learning algorithm for user authentication, where the survey app collected both sensor data and quick answers about what a person thought they were doing, where they were, were they walking, where they were... A lot of todays AI projects are essentially computational sociology, algorithms trained on peoples behavior that is used for predictions.

Business has bought up a lot of the sociology talent, to work with marketing, motivating employees, and changing peoples minds (somebody had to design those motivational posters in the hallways and some of them were based on sociology, lobbying campaigns to change public opinion are based on sociology, political campaigns use sociologists work).

I have a strong suspicion that all the Big Tech companies that collect lots of intimate data about people are employing sociologists to analyze that data and use it for many purposes, mostly related to advertising and motivating people to buy stuff, but also swaying opinion, making market decisions, investing, hiring and firing, and more. A common refrain in economics is that markets are unpredictable because you can't know what all the individual players are thinking, well now you can measure them and build models/AI. Big computational social science already exists at scale. I wonder what is being done with the health data from fitness trackers.

Similar things have been happening in economics. Unfortunately in both cases the data is being used mainly to gain special private knowledge about the social behavior of people, rather than informing the public or public policy. Because money. There are a few researchers who are more interested in the public good though.

> I have a strong suspicion that all the Big Tech companies that collect lots of > intimate data about people are employing sociologists to analyze that data

No need for speculation Facebook got caught experimenting on their users to see the impact of the news feed on their emotional state:

https://www.theguardian.com/technology/2014/jun/30/facebook-...

Glad I’m not the only one who felt this way, given all the vagueness.

At the moment, social science is an oxymoron.

I think where some contention originates from, compared to for instance the German and Dutch terms 'Wissenschaft' and 'Wetenschap', is how the Angelo-Saxon 'science' nowadays emphasises universality and systemisation, whereas the former encompass more broadly all 'practices of knowing' [0].

From this point of view, Sozialwissenschaften is less the act of pinpointing universals, but rather engaging with the plurality of social life and knowledge making.

[0]: https://en.m.wikipedia.org/wiki/Wissenschaft

There's a perfectly good English term for that, Natural Philosophy. What these people do could perhaps be called "Social Natural Philosophy". But if you don't have the ability to do a controlled experiment, you can't do science (or else the term loses all meaning, e.g. is history a science? There are plenty of historians using rigorous methods (often more rigorous than social science, frankly), and it's certainly a "practice of knowing"; nevertheless we would generally say no).

No. There are many methods beyond the controlled experiment. The natural experiment is one, which is extremely important in epidemiology for example. You don't show smoking is harmful by setting up a control group to smoke incessantly for 30 years. Your view of science is far too narrow.

Your view is far too broad. There are many rigorous (and less rigorous) fields of study that are not sciences; if everything is a science then the term becomes useless. (I'm quite content with saying epidemiology is not a science; it draws from scientific knowledge but involves non-scientific study as well, much like history).

I agree about the controlled experiment aspect, but before we arrive at that stage social description allows us to explore a range of possible theories.

In any case, the language example used here was more to demonstrate the historic and perceptual differences as to what amounts to science, and that for some social science is not necessarily an oxymoron.

Edit: wissenschaft and wetenschap are furthermore direct translations, while natural philopsophy moreso indicates a field or discipline (akin to 'Naturphilosophie' or 'wijsbegeerte'). While this can be regarded as 'just semantics', the categories are subtly different enough, at least to a second-language speaker.

It’s clear cut and dry. None of the social studies can replicate their theoretical models. That’s why they’re not considered a science. It has nothing to do with naming in other languages.

Are they still useful? Yes, they still give us a needed picture. However, until we can build a near perfect simulation of the world at the atomic level, social studies will be likely remain as an alchemy or astrology instead of a chemistry or astronomy. There are just too many variables not being accounted for and they depend on too many other disciplines that are also still relatively young like computer science and chaos theory.

From this view, science (singular) is only achieved or perfected once reality can be (nearly) mapped 1:1, which, as we know, is not a realistic goal ("all models are wrong, some are useful"). Beyond replication, wissenschaften (plural) additionally encompasses descriptive methods, which is less about predicting social dynamics and more about summarising sociality to a certain level of achievable granularity.

Hence why I brought up the difference and why I think language differences are important--the ways in which scientists, academics and researchers understand practices of knowing and knowledge building informs their scientific process and the models of reality they hold as true. Besides, models frequently 'drift' and are rarely cut and dry[0], just as blanketing all social science models as irreplicable is simply not true (see Lotka's Law or the Barabási–Albert model; statisticians contributing greatly to social science) [1][2].

More problematically, and not necessarily directed to your views alone, is the continued 'dunking' on 'non-scientific' disciplines. Why are humanist methods held to indefinite scrutiny, while similar criticism could as easily be directed elsewhere? For instance, dismissing physics as 'not a science' because models fail to predict quantum phenomena, mathematics as 'not a science' because there are limits to provability, machine learning as 'not a science' because source corpora are subjectively chosen, and so on (just examples, not my critiques of them).

It is precisely why I regard computational social science as important, seeking to bridge the seeming incompatibilities between exact and inexact phenomena. Even after more than a century of social science, we are only at the cusp of understanding the foundational patterns that make up social life, with contemporary computational methods and data collection capabilities offering new levels of detail into phenomena that were previously difficult to investigate.

[0]: https://en.wikipedia.org/wiki/Concept_drift

[1]: https://en.wikipedia.org/wiki/Lotka%27s_law

[2]: https://en.wikipedia.org/wiki/Barabási–Albert_model

And some how people seem to have a strong urge to dunk on social sciences but not economics, even if the fields are rather similar. And economic models often are even harder to fit into real world data.

Economics is a social study despite all the math

Science isn’t perfect, but they have repeatable theoretical models. That’s literally what separates them from the social studies.

> It is precisely why I regard computational social science as important, seeking to bridge the seeming incompatibilities between exact and inexact phenomena.

Yes, this is what I meant with a near perfect simulation. We agree here

> Even after more than a century of social science, we are only at the cusp of understanding the foundational patterns that make up social life

This is what’s meant for social studies lacking repeatable theoretical models

I got my degree in Computer Science and Anthropology which in my opinion quite fits in line with "Computational Social Science". That being said, it feels that the more applied aspect of computer social science often ends up being human computer interaction or UI/UX design

I don't know you, but I have no idea what this person is talking about.

If I had to guess, I'd say it's an academic talking to other academics about some internal fight about whether or not using big data crap to study social behaviors.

> Maybe the biggest challenge is that we need to embrace our yin and yang—being iconoclasts and friendly neighborhood soccer coaches at once

This makes no sense whatsoever to me.

I have often banged on about "MOOP" - massive open online psychology. The idea is a form of behavioural epidemiology - we can watch through smartphones and other devices people at enormous scale - and answer questions like "do people who save 10% of their income in index linked isas have less stress / better life outcomes" or "do people who meet a mate for a drink each week live longer" ...

The answers may well surprise us, but at some point the answers will also benefit us.

Don't we already have many answers? -- We just do nothing to address the problems.

I had ideas for exactly such an app.

Very interested in the unexpected correlations it can find, especially with respect to health.

The main trouble is how to deal with keeping it secure/private/anonymized, while also making the data open for others to run queries on.

I don't see it as one "app". It is everything. Record my conversations, who was in the room with me (bluetooth, facial recognition), which chat room was I in, was my tone of voice loud or angry when I was talking to my daughter about homework, do I do that often, how does she react. Out of 1 million parents who raise their voices discussing homework, what is the stress level of the children, the amount of times they spend doing their homework after the dinner conversation. Of 1 million children doing homework which app is most effective e for their academic success - and is their stress negative correlated with academic achievement or not?

Then take all of this and dump it toNHS / NICE researchers whom I am happy to let have this because the legal regulations are in place on supra-national levels such I know that if a data-brocker gets it they shit their pants and send it back to the government with an apology.

It's a long way off, but one day your iphone will trill in your ear and say "hi paul, I know you are angry about the tone of voice your wife used just know but she is worried about her unfinished report, and out of five million men who react the way you are about to 5 million did not get positive responses."

It’s been done. The secret is to reward participants and make it long term. Check out the General Social Survey conducted by NORC. It’s one of the longest running longitudinal surveys with multiple cohorts tracking the public’s changing views on a variety of issues combined with detailed demographic data.

I've thought about this as well. I wanted to create anonymity pools and then use zero knowledge proofs to justify predicates about the entire pool.

So it would be like:

1000 participants formed 10 pools, assigned by random numbers generator. Preficatefoo was true of 19% of pool foo, 24% of pool bar...

Participants could then attest: yes, I was in pool foo. But there wouldn't be enough data to reidentify them individually.

> The reasons follow below and I also cover what a computational social scientist should know [and] do

It doesn't seem to do any of that. There's nothing actionable here AFAICS.

I wouldn't be too happy about such a divide into high-level analysis and interpretation on one hand, computational social science, on the other hand. A certain degree of intimacy with your data is essential. There are even cases, where high-level descriptors would suggest one thing, while in the actually data, there emerges an entirely different picture.

As long as political correctness is programmed into these systems, we’re just lying to ourselves. It doesn’t matter if it’s a human, a programmer behind the scenes, or biased data sourced to train a model.

But we do have computational social science. Simulations, interaction networks, opinion spreading models, voter models etc. Quite an interesting area of research (and practically useful).

There aren't a lot of people being paid to do this type of work though and social science tends to lag behind on the technology front because people with said skills tend to find more lucrative venues.

I've worked with a few computational social scientists. Personally I've always found the area a bit fascinating. There's lots of agent based models, tie ins with economics, social models, etc. trying to determine how groups of people will behave and interact. Lots of understanding behavior and exploring implications policies may play on behavior from a governing standpoint or how things naturally occur.

Well, as a rule, you don't come to the academia for money. I am a software engineer who now works as a researcher in the university for 3x less salary, because I just can't stand bigtech anymore (and probably unfit for corporate life in general).

Another thing is that software engineers are extremely highly paid only in United States, and even there perhaps only in a few highly concentrated places. Everywhere else the difference is not that dramatic (and not everyone in the world with the skills can move to Bay Area).

Good idea. What happens when math uncovers something in the social sciences that you don't like or people will find offensive?

Such as what? Sounds like you already have a conclusion in mind.

There are plenty of examples where research or researchers have been cancelled for factual objective analysis.

Your comment seems disingenuous.

Can you share some then? Asking honestly.

IME, "factual objective analysis" relies on the underlying data being accurate and clean. That's never, ever the case, but people see "data" and think of it as a baseline truth. So I'm curious if you can share some that don't have that fault.

What courses of study and applied methods could further computational social science? What are some good resources for foundations in the requisite methods and their impacts?

"Applied causal inference with observational data (for benevolent humans) and classical counterfactuals"

And then maybe something like,

"Quantum statistical foundations for causal analysis of emergent patterns in adaptive, nonlinear, possibly complex systems"

Though already suggested is:

"Validated Evidence Based Computational Thinking and Cybernetics" https://twitter.com/westurner/status/1118822798217101313

Causal inference > Approaches in social sciences https://en.wikipedia.org/wiki/Causal_inference#Approaches_in...

"Answering causal questions using observational data" (2021) [PDF] https://www.nobelprize.org/uploads/2021/10/advanced-economic... https://www.nobelprize.org/prizes/economic-sciences/2021/pop...

/? causal inference site:github.com awesome https://www.google.com/search?q=awesome%2Bcausal%2Binference...

/? causal inference from: https://westurner.github.io/hnlog/ :

https://news.ycombinator.com/item?id=20178068 ; Python packages for causal inference

"The limits of graphical causal discovery" (2021) https://towardsdatascience.com/the-limits-of-graphical-causa... :

> Structural causal models support three key operations: observations, interventions and counterfactuals.

What is the difference between counterfactuals in structural casual models and counterfactuals in (quantum) Constructor Theory?

How can causal inference identify causal linkages between nonlinear quantum complex adaptive systems (that are affected by other [super-]fluidic fields), in computational social science and data science?

Do we need even social science? Does social SCIENCE even exists? Isn't behavioural biology 'nuff to describe the pathetic monkey horde?

Psychohistory ...

If the world needs it, it’ll be there. Subsidizing it does not suddenly make the world need it.

I believe shopping this assertion around a social science department (or economics, if we’re not calling that social science) would yield a rather more nuanced view.

The internet was subsidized....

no, it was invented at UCLA and Stanford and financed by what is now modern DARPA. It wasn’t a matter of financing, it was a matter of national security. we are subsidizing plenty of advanced technology that puts us on the bleeding edge of military, but that doesn’t mean they wouldnt find private funding

The concept reminds me of 'psychohistory' from Asimov's Foundation novels.

The social "sciences" are a political tool, whose usefulness depends on their ability to reach any desired conclusion.

The assumption that better (that is, more accurate) social science is what people want shows a hilarious lack of awareness of what is actually going on.

Did you read the article?

"In summary, CSS is, first and foremost, a division of social science whose goal is to generate knowledge about society and the social human and make it accessible to everyone."

Huh? That's a field?

Computational social science should be devoted to developing testable theories with predictive power. "Making it accessible" is a job for popular science writers, not researchers. Of course, if you can create theories with proven predictive power, you're probably going to do better at McKinsey or Black Rock than in academia.

The beginning of psychohistory?

No. It is just too powerful. It will be used in panopticon-like setting.

lots of ignorant takes in this thread about sociology...not surprised tho

> It should do that in total epistemological freedom, being informed by everything from the latest in ML to the oldest in sociology and not bound by conventions of what constitutes a scientific explanation, etc.

So the author wants all the shine that being labeled a science brings without any of the heavy-lifting that science requires (i.e., following the “conventions of what constitutes a scientific explanation”).

I can play this game too: my life is evidence-based and data-driven, by which I do not mean that I gather and use data to help decision-making like crusty old “data scientists” learn to do nor do I follow any accepted standard of evidence, because I am an epistemological anarchist! My data is my experience and my evidence is how I feel in any given moment. I should get grants from grant-making agencies just for existing!

If I’ve read the article correctly, all that stuff seems to have mainly to do with avoiding becoming tied to the particular, distinct norms, methods, and standards of the various disciplines CSS might touch, interact with, or be applied to.

It explicitly opposes having no norms or standards. It appears to be intended to stake out CSS as an independent discipline, rather than a study of pure methods that can end up “living under” any of several other academic areas, with mutually incompatible norms and cultures. (I’ve no idea whether that’s, in fact, a good idea)

Yeah, I think that’s fair, as I just said in response to a different comment. [0]

Of course, the institutional and political factors of universities and funding will make establishing an entirely separate discipline an extreme uphill climb. I suspect the author realizes that when they emphasize the importance of having some traditional disciplinary knowledge.

[0] https://news.ycombinator.com/item?id=37746815]

A more charitable interpretation would be that the author is being tactical in addressing a discipline where the dominant fashion has been to critique or dismiss scientific methods. By packaging “let’s be more scientific” in an argument that sounds like “let’s not be narrowly prescriptive about epistemologies”, they may have a better chance at getting through to the most relevant target audience.

Yes, and now that I reread it I also see there’s a bit more emphasis than I perceived on first reading on the author speaking to people who do computational work to convince them that social science disciplinary knowledge is important to have. In other words, the author is saying to computational workers, “Don't pigeonhole us as totally outside academics” and to academics, “Don’t pigeonhole us as merely methods people.”

Of course, a manifesto proclaiming the independence and usefulness of computational social science will never get you as far as actually demonstrating its usefulness.

Things like total epistemological freedom always end up being a euphemism for chasing fashions.

To be fair, it’s not like the traditional academic social sciences need much excuse to chase fashions as it is. If anything, that very tendency might be what gets computational social science’s foot in the door. “Hey, we’re the hot new thing!” (That they can leverage the ML/AI hype to generate grants will be the main thing, of course, that secures their place.)

I always took it as a warning that you're about to be asked to make a gigantic leap of faith.

Is playing make believe (like mind reading the goals of the author) just for fun, or is that more of that scientific rationalism that everyone claims is so powerful, yet is powerless to turn around the warming of the planet that it kicked into gear?

Was that written by AI? Because I have no idea what the author was trying to say, and I'm not sure they were saying anything at all. The confused haircut metaphor did not help.

Quantitative social science exists. What is the big revelation here? What makes it computational? I don't understand. Can somebody enlighten me?

If you want to hybridize computer science and social science, I've got some good news for you. The ranks of data scientists across the corporate world are filled with ABD grad students from a range of quantitative fields, who applied their own problem solving skills to the question of "how do I turn my skillset into money." And in that regard, maybe computational social science in a broader sense is the art of converting quantitative social science into profits for major corporations. The utopia that CSS aficionados longed for is already here.

Well for one thing, the need for better consensus, conflict resolution, and fact (or dictionary) propagation in society is obvious and likely never-ending, considering how many corners of life it's applicable. I've been considering trying to find a way to study sociology/computer science fusion in grad school for a moment now myself.

Going through Petter's archive it doesn't feel like it, writing about the topic for some time.

I gathered the grooming metaphor referenced how it's time for the CSS practitioner to don the complete methodological anarchy that once dominated the field, while remaining open to disruptive changes coming out of CS or elsewhere.

> Because I have no idea what the author was trying to say,

That sounds like a 'you' problem.

"The world needs another field where nothing replicates and those main function is to produce fake results for airport books aimed at management consultants"

more data to brainwash and control society by dark money yay!

People don't want truly good social science. The observations won't be politically correct and the use cases are controversial. Otherwise Cambridge Analytica would be a unicorn.

I'd argue it's the opposite. Everyone wants truly good social science, it's why when it matters they go to data scientists and not social scientists.

Election tampering is truly good?

Election tampering was an application, not the science. You cannot tamper successfully in an election without having a good understanding the electorate. And if you have that good an understanding of the electorate, you could put it to all kinds of other uses. The 'burning' of Cambridge Analytica could definitely bee seen as a baby and bathwater situation, or an admission that the what they had created was too powerful a tool to be allowed to exists.

Why would anyone would want to corrupt anything with the anti-science that Social Science is today unfathomable, unless corrupting another field is the goal.

Reply on news.ycombinator.com