Is "theoretical" computer science too narrow?

Question

The CS Theory Stack Exchange is humming along nicely. It's a terrific, robust, and growing community that is clearly making an increasing contribution to the state of CS Theory research.

Something we keep thinking about, though, in the context of the entire Stack Exchange network, is how broad a Stack Exchange site should be. I blogged about this a little bit here:

So: the right size might be somewhere around the size of a university department. Somehow, the cultural anthropologists don’t mind sharing a building with the physical anthropologists, and when they both find themselves at the Yale-Harvard football game, you can bet that they’ll sit together and find something anthropological to talk about. Similarly, at Stack Overflow, the Java Entity Bean programmers at insurance companies don’t mind all the iPhone developers asking Objective C questions about the horrible, horrible game they’re working on. Heck, they might become iPhone developers one day. And they both share the humiliation of not being able to fix their uncle’s virus-infested Windows XP machine when they’re home for Thanksgiving.

So that leads to the obvious question... is this site too narrow? I don't know of many universities that have a Theoretical Computer Science department; they all have Computer Science departments and the algorithms purists don't seem to mind having Friday tea with the computer vision geeks. And the idea of meeting somebody who is working on AI is, it turns out, not entirely too horrible to contemplate.

I'm concerned that there are non-theoretical aspects of Computer Science research that don't have a home right now. If you look around Area51 you'll see proposals for crpytography, operating systems, compilers, cognitive science, numerical modeling, computer vision, bioinformatics, history of computers, and a lot of other fields that aren't quite reaching critical mass. We had an AI site that failed to ignite due to a lack of actual AI researchers.

I'm not saying all of those are on topic, but if a university had, say, two professors interested in cognitive science, I'm pretty sure they'd just stick them into the CS department and nobody would think that was strange or upsetting or crashing the party.

Right now, though, I see researchers all over the computer science department that don't really have a place to participate, and the idea of making dozens of thin little sites to cover every possible ACM SIG strikes me as ignoring one of the great reasons that Stack Overflow succeeded in the first place: because we brought together all programmers to talk about code, and we created a reliable and robust tagging system so that you could quickly find the topics and people that you cared about, and because nobody minded (and many people enjoyed) the occasional brushing-of-shoulders with people who shared the same values but worked in different technologies.

Can we do something here? I think that sticking to research-level conversation is a great common denominator, but there are an awful lot of interesting topics that are excluded right now which could really use a home. See, for example, arXiv's idea about what constitutes "computer science":

Artificial Intelligence; Computation and Language; Computational Complexity; Computational Engineering, Finance, and Science; Computational Geometry; Computer Science and Game Theory; Computer Vision and Pattern Recognition; Computers and Society; Cryptography and Security; Data Structures and Algorithms; Databases; Digital Libraries; Discrete Mathematics; Distributed, Parallel, and Cluster Computing; Emerging Technologies; Formal Languages and Automata Theory; General Literature; Graphics; Hardware Architecture; Human-Computer Interaction; Information Retrieval; Information Theory; Learning; Logic in Computer Science; Mathematical Software; Multiagent Systems; Multimedia; Networking and Internet Architecture; Neural and Evolutionary Computing; Numerical Analysis; Operating Systems; Other Computer Science; Performance; Programming Languages; Robotics; Social and Information Networks; Software Engineering; Sound; Symbolic Computation; Systems and Control

Some of that stuff is already on topic here, but a lot of it is sadly homeless in the Stack Exchange network.

My belief is that broadening this site to "computer science" and allowing discussion on any research-level topic that fits in one of the arXiv categories of Computer Science (and possibly even using those categories as tags) would make this site grow quickly into a much larger and stronger resource that attracts people in computer science departments around the world, rather than a thinner site that attempts to stitch together just the purely theoretical disciplines. Of course through the use of tags and our tag following feature, every participant could define their own filtered view of what kinds of topics they are interested in following, but if you want to reach outside of your narrow research interest once in a while there will be lots of other interesting stuff going on here too. And the opportunity to connect, even accidentally, with researchers who are ostensibly working outside of purely theoretical computer science but might, nonetheless, face the same types of problems seems like too good an opportunity to pass up.

What do you think?

Answer 1

sheesh. I go offline for 6 hours and possibly the most interesting question on meta shows up.

I think firstly that it's a tribute to the community here that Joel has even raised this question: it makes me feel that we're doing something right enough that he'd like to expand the scope.

I'm skeptical about the likelihood of success for an expanded CS-level version of this site. Part of my skepticism is because of things I don't know and could be wrong about, and part of it is things I have experience with.

The major part of Joel's argument is that IRL, there aren't theoryCS departments, and so it's not unnatural to have generic CS departments here. I think that's exactly the wrong way of looking at the value of such sites. University departments are market-driven: they want to provide a generic-enough service to attract a large number of "customers", while not watering down their mandate so much that they lose definition. They are also slow-moving lumbering beasts, so it takes a long time for them to evolve, and entrenched interests often prevent true change from coming.

But research communities form transversally: representatives of the community within departments meet up with each other, and as a lone theoretician, I can tell you that I find MUCH more community at conferences, and here, than in my home department. I think theoreticians would LOVE to have a theoryCS-only department (otherwise known as Princeton :)), but since that will never happen in the university environment, we are ecstatic to have it here. I would suspect this to be true for other subdisciplines in CS as well.

Why isn't this the case in math ? it's because all mathematicians speak the same language and have the same overarching goals: to understand mathematical structures. CS subdisciplines don't have this common language: there are the electrical engineers, the software hackers, the AI researchers, the graphics/imaging folks, and then the mathematically minded theoryCS people. In fact, CS departments are making moves towards splits: witness Georgia Tech and CMU, for example.

I think we could be a lot broader, and we need to work on that. But we have to be broader within theory. As Sylvain, Dave and Neel have argued, we need more theory B (logic, formal methods, programming languages). We also need more of the active researchers to show up here. I know many colleagues IRL who aren't here, and they should be. That's where I'd want to spend my efforts to make a community that I'd enjoy more.

Finally, an argument that Joel makes regarding SO is that even if the communities are largely disjoint, the occasional serendipitous connection achieved from a merge might be worth it at little cost. I see how this happens on SO, and it might work here. However, there are many features of the site (most importantly reputation and the associated power that comes with it) that are global, and can easily get swamped by a few high popularity groups. A personal example: while I'm fairly active on MO, I prefer to spend my time here, because only a tiny fraction of the questions on MO are within my areas of expertise, and even though I've set up elaborate tagging filters, most of the first few pages of MO are not relevant to me. I fear that happening here too.

Answer 2

Clearly there is potential benefit and potential cost to Joel's proposal. The benefit would be that we're exposed to more questions and have larger participation. The cost is that the site might fill with all sorts of questions that don't interest us and we would get lost in the flood.

A good way to judge these factors is to imagine how often the typical user of this site would read / answer questions on memory management, context switching, or image acquisition. We could also try to imagine whether a systems researcher would be interested in strange complexity classes.

My feeling is that the cost would outweigh the benefit. However, it wouldn't have to be that way if the stackexchange software were more flexible!

If you think about a CS department, while all areas of CS are covered, there are separate research groups, labs, etc. We don't all go to each others' talks because, frankly, not all talks are interesting to everyone. So if we structured this more like the CS department in the analogy, I would be all for it.

Here is an idea:

I think it would be great to have a CS stackexchange site that had multiple "forums" like cstheory. The membership could be shared. The reputation points could transfer across the forums. But each site should have its own homepage which featured only its own questions - tags and filters are not nearly enough. Perhaps the featured bounty questions could be shared across forums, as could questions with very high scores. That way, we could see what exciting things are happening in the other fields and would be tempted to occasionally browse the other forums without their questions clogging the theory ones. This would increase the user base as users from all the forums would occasionally do some browsing and answering each others' questions.

However, I see no way of implementing this idea in the current software. Stackexchange clearly sees the value of combining the various fields because their membership and viewership would grow. So they would have to make it happen by making their software be flexible enough.

Also, of course, the community would have to think it's a good idea :)

Answer 3

Moving in this direction would be contrary to early Meta discussions, and discussions that preceded commitment, for that matter. It is also a genie that could never be put back in the bottle. Nothing is preventing anyone, CS professor or otherwise, from asking a research-level question about theoretical computer science, as things currently stand.

If this change is implemented, I will log in significantly less, because the discussions will not be at the current level that interests and challenges me.

Why remove the theoretical focus? Theory is marginalized enough in CS as it is.

I support the objective of constructing a site for computer science theorists, broadly defined (and, as a practical matter, defined by moderator action and Meta discussion), to share research knowledge with themselves, and with anyone else who wishes to participate.

Answer 4

If you judge by conferences (and not university departments), then Theoretical Computer Science is the natural community. We have the theoretical computer science conferences FOCS, STOC, ICALP, but there's no all-computer-science conference. There are big math conferences (internationally, every four years; in the U.S., every January), and big physics conferences (in the U.S., in March and April), but not CS.

Answer 5

I agree with many answers here. In brief:

• It works. Let's not try to fix it.

Some opinions:

• Not everyone agrees with the premise that StackOverflow is exemplary and that all other sides should follow its model. I find its huge volume of traffic exhausting and daunting. It is too fast-paced for my taste. Questions are asked, very quickly answered, and then forgotten. It might work fairly well for technical programming-related questions, but it is not necessarily a model that I would like to try with theoretical research problems. I think that reasonably low volume is an important feature of this site.

• I think that this site is too narrow, but it should expand the scope within TCS, not outside TCS. There are subareas of TCS that are virtually nonexistent here.

• Universities should have TCS departments (and some universities do have TCS departments). As Suresh pointed out, CS departments mix people who don't share a common language, and in practice, it does not make that much sense.

Answer 6

Thanks for raising the issue. I don't have a clear opinion on the issue and have to think more. But note that we started this separating from MathOverflow. A general research level computer science might be too big, and the cultures seem quite different. We might end up facing the same problems that lead us to separate from MathOverflow. There are reasons why departments have different groups, ACM has different SIGs, and researchers in Systems, AI, and Theory have different conferences.

I don't know if this is correct in general but in our department the amount of interaction between different groups is not very big, AI people work mostly together, similarly theory, systems, and other smaller groups.

---

I think AI researchers have already a site (metaoptimize) so it is not strange that the AI.SE did not get enough users to get off.

The research-level Crypto questions are definitely on topic here, but based on the questions posted on the proposal it is not clear to me that is what the Crypto proposal is aiming at. Similarly the practical algorithms proposal.

---

Lets go over the list you had there:

1. Artificial Intelligence; Computer Vision and Pattern Recognition; Computation and Language; (and Vision)

2. Computational Engineering, Hardware Architecture; Hardware Architecture; Networking and Internet Architecture; Operating Systems; Systems and Control; Robotics; Multiagent Systems; Software Engineering; Performance;

3. Graphics; Human-Computer Interaction; Multimedia; Sound;

4. Databases; Information Retrieval; Digital Libraries; Social and Information Networks; Computers and Society; Finance, and Science;

5. Mathematical Software; Numerical Analysis; Symbolic Computation;

6. Computational Complexity; Computational Geometry; Computer Science and Game Theory; Cryptography and Security; Data Structures and Algorithms; Discrete Mathematics; Distributed, Parallel, and Cluster Computing; Formal Languages and Automata Theory; Information Theory; Learning; Logic in Computer Science; Neural and Evolutionary Computing; Programming Languages;

I have dived the areas in the list to make the discussion more detailed and constructive (and I know that others would probably have different divisions). I think those in group 1 already have a site and are happy with it, it is unlikely that they will leave MetaOptimize. It seems to me that group 2 and 3 will probably want to have their own sites, and I have doubts about how interesting their questions would be for theory researchers. Group 4 and 5 are closer to theory. And most of group 6 areas are already on-topic on cstheory AFAIK.

---

Answer 7

In my opinion there are a few, more-or-less independent, issues discussed here. I will try to list them:

1. do we want to broaden the scope of the audience of TCS.SE?
2. do we want to broaden the topics discussed on TCS.SE?
3. do we want to broaden the level of the discussions in TCS.SE?
4. what is the real-world kind of community that we want TCS.SE to resemble most?

In my opinion, almost everybody agrees that it would be beneficial to enlarge the audience of the site, but the directions of such expansion is not unanymously agreed upon.

I share the belief that CS is not a field sharing a common language, therefore I would not like TCS.SE to accept all possible CS topics. Also, I agree that we should welcome people coming from Theory B (and I would add OR people to that).

A more controversial opinion of mine is that we should also welcome questions that are almost research-level, provided that they are clearly tagged as such (I am thinking about something similar to the soft-question tag). The reason for that comes from my opinion on point 4.

The question originates from the idea that the TCS.SE community shoud actually model the one of a CS department. My opinion is that it should instead model a conference/workshop. A bolder statement is that TCS.SE should be an ever-running whole-TCS conference with over a thousands partecipants that the real-world TCS community sometimes talks about but has never materialized.

A second motivation for such a model is that it should be a vehicle for promoting TCS to potentially interested students, by showing what academic research in a decent, friendly community (surely TCS.SE qualifies) is like. The point of allowing almost research-level question is to provide a gentle introduction to advanced TCS topics suitable to undergraduate interested students.

Answer 8

Disclaimer: I started to participate this website only during the public beta. This means that I do not know how its scope was determined to be “theoretical computer science” instead of something broader like “computer science” or something narrower like “algorithms and data structure.” And I do not value much the opinion of a person who does not know the history. You should consider whether my opinion should count at all.

I like the idea of broadening the scope. Theoretical computer science has much to offer to and hear from applied fields. However, broadening the scope is dangerous to the quality of answers, especially to topics with fewer users. Therefore, I support broadening the topic only if we have a plan to keep the quality of answers in minor topics high.

Stack Overflow fits the model with a broad scope with many subtopics, and therefore I think that it is useful to see what happens there:

1. I guess that Stack Overflow has many C# users who are not usually interested in PHP and many PHP users who are not usually interested in C#, and I guess that they coexist just by using different tags. If this is true, we may use the same method. Occasionally we can see questions in the other tags to see if anything interesting is going on in the next room.
2. However, less popular topics are miserable on Stack Overflow. I have seen several high-score answers to questions about Haskell which are incorrect apparently because the answerer lacks real-world experience with the language. I imagine that not many Haskell experts are interested in Stack Overflow because the top page almost certainly contains no questions about functional programming. If there were a Stack Exchange website dedicated to programming in functional programming languages, the situation might have been different. I do not want to see the same thing happen on the proposed SE site with the broader topic.

Answer 9

The question of broadening the scope is indeed a good one. But recall that before thinking for enlarge the scope to CS outside TCS, we should first think about really broadening the scope to TCS theory B...

Answer 10

I wanted to comment on Robert's point, but the comment box is too short for my answer.

I think Robert's rephrasing is close to the spirit of what Joel intended, and unfortunately I think that this whole discussion has triggered some old fears. Without attempting to channel too much what the community thinks, let me make some points in response to Robert:

• Would we like to be part of a larger, harmonious CS gathering ? Of course ! I don't like being marginalized as part of a 'useless, mathematical' CS ghetto. The responses though come from fear of having to constantly defend ourselves and protect ourselves from the reams of 'but theory isn't useful' and 'who cares about theory' comments we're used to hearing anyway as part of a larger CS discussion. Call it battle fatigue if you will
• But do we think it has value ? that's more complicated. Your entire argument from value is based on the assertion that a large group has network effects that outweigh the niche community benefits we perceive here. The question is, for whom ? There's value for students who want to be exposed to the virtual university, but there isn't value for us, who want a community of our own: we get enough exposure to general CS in our departments, and many of us work a lot with outside groups (I have papers with DB, ML, vision, graphics and parallel folks, for example)
• We've worked very hard to create what we have so far. And by no means are we done growing our community. It's a pity that the other area51 proposals haven't taken off: I don't know how well they've tried to proselytize, but I know that many of us worked very hard to get us over the definition and commit phases with relentless promotion. I think therefore we're a little defensive about any attempts to change it. But I can see a future in which we do become part of a larger community - just not at this instant.

Answer 11

If expansion is considered, why is it considered under orthodox university department lines? For instance, I am in a department called Combinatorics and Optimization under the faculty of Math and I enjoy the site greatly. I would argue that $||\text{C&O} - \text{TCS}|| < ||\text{Software Engineering} - \text{TCS}||$ in any natural norm on common interests, tools, and ideas.

There are people in linguistics departments that ask questions about the learn-ability/structure of languages in clear TCS-friendly terms (as a dated example take the Chomsky hierarchy). People is biology departments that ask questions about the complexity, evolvability, and robustness of systems in clear mathematical ways. On the flip side, there are researchers in, say, software engineering or HCI that ask questions about the usability of a code base or interface that seem completely foreign to someone in TCS and much more friendly to a psychologist.

Wouldn't it be more exciting to see how similar ideas are used by a linguist or biologist that you seldom get to meet at your university, instead of reading about completely foreign ideas that a distant part of CS uses that come up at every other colloquium in your department?

Why expand towards unrelated parts of computer science before expanding to other parts of the natural (and even social) sciences and math that adopt the algorithmic lens and use the same language? Why not provide a service that you don't have at your home institution instead of modeling ourselves on some orthodox division of study?

Answer 12

For what it's worth, my belief is that a major contributor to the success of this site was the previous success of MathOverflow, and massively multi-collaborative efforts of Gowers, Tao, Kalai, polymath. These phenomena were very influential upon certain components of the theoretical computer science community (myself included) and sparked the urge (subsequently satisfied) for a site of our own. My contention is that in as little as a few months, members of other computer science research fields will in turn be influenced and encouraged by the success of this site (as well as perhaps the massively multi-collaborative efforts last year to pick apart Deolalikar's manuscript) to start their own site.

To be clear, I should also say that I do not favour an expansion of the scope of this website.

Answer 13

There are certainly risks in expanding the charter for the site, but there may be opportunities.

Theory B is a messier part of TCS than theory A, with more connections to the rest of CS. In particular, looking at the proceedings of such programming languages conferences as POPL or PLDI, you'll see a mix of theory B and non-theory contributions.

Some number of theoreticians will find a broader site attractive. The question is, how many? But if we did have the broadr scope, I would expect the first non-theory users to be from those who have some interest in theory, and that might well establish the tone for the site.

Lev's idea of having a series of fora would reduce risks. That we already make a lot of use of Arxiv-derived tags would make this easy to work with, at least from the user's point of view.

Answer 14

I think broadening scope now could achieve next to nothing, even assuming that people from other areas moved in here.

In my experience, theorists just think differently than software engineers, to create the hardest contrast. There would be next to no overlap in questions and too much divergence in mentality for one to provide helpful answers for the other.

Sure, we could peacefully coexist, no doubt. But I believe that not much mingling would take place.

Answer 15

Excellent discussion. I am intrigued by the passion poured into it.

I would like to get your thoughts from a different angle — Rather than contemplating whether you would allow this expansion of topics, I would like to see what you think about being included in an inevitable site covering the larger field of "Computer Science." There's a big demand for these topics and they inevitably have to go… somewhere.

What we are really proposing here is the creation of something like a virtual Grand University of Computer Science. I've been looking at the ACM SIGs while Joel cites the Cornell arXiv: Computer Science library. Either way, I very much envision this hypothetical site being a higher-level academic setting where you have that sense of place of belonging to a top-branded university; a place where a community of students, professionals, and scholars collaborate about higher education topics and research.

The simple question we are coming to you with… Can we break ground here and expand your college into a top-tier, well-equipped, full-fledged University? Or, do we have to go a few miles down the road and find another piece of property and build our campus there?

You've already created your own specialized, great virtual college campus here. You've certainly earned your respected place to keep as you see fit, no doubt. But the (currently) top-voted answer laments that "Theory is marginalized enough in CS as it is." In that vein, I am having trouble envisioning a site purported to be a virtual CS University, but always having that ^little asterisk that says: _{^{but for Theoretical questions, go here}} →

My question may be a bit loaded, but I would like to point out that there is a lot of value in having a strong site that includes your topics of interest. Topics don't simply disappear because people are also talking about related topics. The community doesn't let it. If the tag-set is strong, the subject stays strong. These sites scale wonderfully. The traditional religious and platform wars simply do not happen; all are welcomed and we all get along… somehow.

We've already proven that the network effects of scale far, far outweigh the niche closeness you might feel here.

Answer 16

It seems to me that the broken thing that Joel is hoping to fix is that many computer science researchers do not have a StackExchange home, yet it seems difficult to create compelling enough Area51 sites to take off. In a first-pass analysis, there are probably just as many people in AI as in TCS so why did the AI site not take off?

However, many of the non-theory computing researchers and students I know are already on StackOverflow. Maybe these folks do not have the incentives to move away from SO to form new communities? If I had a 10K+ reputation on SO I would certainly find it jarring to have to build a new site from scratch, scrabbling initially for enough votes to even close a question as off-topic. Few of the people on cstheory came from SO, and MathOverflow had different software, so downgrade of functionality wasn't an issue.

CSTheory certainly became significantly less compelling for me post-beta, as high reputation users lost access to important site tools such as the list of all recent edits.

Answer 17

+1 I think you might as well do this. On the "Official ACM" Linked-In group, there was a question about whether or not Computation Theory should be required for Computer Science undergraduates.

What that revealed to me is that there is still vague understanding of the relationship of theoretical computer science and the broader computer science discipline, since some responders thought theoretical computer science included software engineering topics, analysis of algorithms, etc.

It would appear that making this a general Computer Science forum would assist in bringing together an understanding of the sub-disciplines, as it were. I trust that Computer Science would also allow for applied research, not exclusively theoretical research (however that strikes folks as different).

We still have the confusion of computer science and programming to deal with, but I think that is not something we can control and it is no reason not to broaden this StackExchange theme to Computer Science researchiness (however one draws the line about that).

Answer 18

Seems that not a single answer supported Joel's point of view, so here is one.

Point 1: Unclarity for a user

From a user prospective:

if my question is too theoretical for 'stackoverflow', then I post it to 'cstheory'.

This is a wrong behavior, but you can see this trend by looking at many-many "easy" questions on cstheory --- many questions should be moved/posted to cs. Hence, the original assumption separating cstheory/cs results in separation of question difficulty was not confirmed in real life. (I might be wrong here about the original intention of creating two communities; see below. I might also be wrong about "many-many".)

Point 2: Bad naming (point 1 cont.)

If cstheory is intended for research-level questions, then call it csresearch. If it is intended to split TCS/CS, then call it tcs. Why not cstheory? to avoid the wrong behavior mentioned in point #1: "lame" users are more familiar with the abbreviation cs than tcs, so it is more likely that they publish to cs. But now they see cstheory and the brain fires: "hey, it is about cs" and "hey, yes it is too theoretical for stackoverflow" so "hey, i will post it to cstheory", unaware of existence of cs. Alternatively, stackexchange could promote cs more aggressively than cstheory, to a "lame" user.

Answer 19

Maybe some voice from outside... IMO there exist a not so well visible but still solid border between Computer Science and something that I call Computational Science.

The first is mostly a part of mathematics dealing with algorithms, data structures, complexities, languages and stuff. This is what mostly happens here.

The second is an interdisciplinary movement of people using computer computations to make science; this group is rather united by common practical problems and methods than a common interest. Those are people that make all kinds of simulations, like computational physicists, computational chemists, computational biologists, computational sociologists, ..., meteorologists, economists, visual/physical effect makers, various kinds of engineers; people processing data, like various data scientists, bioinformaticians, chemoinformaticians, again economists, machine teachers, text miners, image and sound wizards and the rest of data-AI; finally all soft computing, robotics and steering theory debris.

There is some overlap between the two, mostly in areas of numerics, BLAS and some "golden" algorithms like Smith-Waterman (C-nal Scientists spent thousands of CPU hours in such so they would use any fast working black-box C-ter Scientists would gave them). Outside this C-nal Scientist don't care much about theoretical aspects of their algorithms. Also most C-ter Scientist don't care much about practical aspects of their work -- and that's why I think those two topics won't blend well.

Summing up, I second with integrating C-nal Science on Area, but rather in a new proposal than this site; see my Area51 discuss question about it.

Answer 20

I would like to see this site be broad rather than narrow, and I hope it will be friendly to non-experts trying to learn more. As far as the crossover between CS and Math, I looked at the names {CS,Math} Overflow, and thought, maybe "Overflow" could actually allow Stack Exchanges to talk to each other.