Should Stats SE encourage RL questions to be asked on AI SE?

Reinforcement learning (RL) is currently on-topic here. However, Stats SE, as we all know, is more statistics-oriented, so there are users here that have no interest in RL, which is a framework for intelligence (based on how animals learn): that's why I (and others that have been answering RL questions both on Stats SE and AI SE) have suggested that Artificial Intelligence SE is the best place to ask theoretical questions related to RL.

In fact, AI SE has the highest number of questions tagged with RL among AI SE (currently, we have 1,548 questions out of 8,057 tagged with RL, so this means that more than 1/8 $$\approx$$ 0.19 of the questions are about RL), Stats SE (currently, 908 questions are tagged with RL out of 172,687 questions, so the ratio is about 0.005, which is a lot lower than 0.19), and Data Science SE (currently, they have 480 questions tagged with RL, out of 26611, for a ratio of 0.01), and almost more than Stack Overflow (which has 1777 questions tagged with RL, with ratio 0.00008). Of course, these ratios are not completely informative of the interests of the community, but they are at least suggestive. In fact, I've personally noticed that many people interested in RL end up on AI SE and they regularly visit the site, and this includes people that answer RL questions even on Stats SE and Data Science SE, as you can see here, here and here.

If you also look at the most popular tags on these three sites, RL is the 3rd most popular tag (just behind neural networks and machine learning) on AI SE, while, as it should be clear now, it isn't as popular on the other sites.

Although it's true that both AI SE, Stats SE, and Data Science SE are "competing" for users, wouldn't the 3 (overlapping) communities benefit from having a more solid collaboration? I think so.

So, I would suggest that all theoretical RL questions are asked on AI SE. Why? Because of the reasons I just mentioned above (i.e. AI is about intelligence and intelligent agents, and RL is one, if not the main, approach to achieve that), while it's not strictly related to statistics or data science. In this way, all users interested in RL would know that that is THE site where to ask the question and we can decrease the number of questions that went unanswered because the right user that knew about the topic was in another community. I know that this may not be beneficial for Stats and DS in terms of activity (and some of you will downvote this post because of that), but, in any case, the amount of activity related to RL in these sites is relatively small (compared to all other topics). As we are all part of SE, we should try to collaborate more so that SE is a place where users can come and get good answers and questions.

So, what do you think about this? Should users on Stats SE and Data Science SE encourage askers to post their RL questions on AI SE? Note that this does not mean that RL questions will be necessarily off-topic on Stats SE or Data Science SE. It only means that we will build more cohesive communities, both in terms of questions and users (with common interests).

• @Sycorax The formulation of that post is completely different. It seems that you're just linking to that post for other reasons than the ones you mention (i.e. to bias people against this new post). In this new post, I'm suggesting something different and I'm also showing facts. – user82135 Jan 4 at 19:01
• +1 because it's a reasonable question and worth discussing, whatever one might think the resolution should be. – whuber Jan 4 at 19:07
• It is rather obnoxious to preemptively declare that the motivation of people who downvote this proposal is to protect their 'turf' (& thus by implication only the upvotes on this proposal reflect its merits). I also note that nowhere in the linked thread does anyone say that askers posting elsewhere hurts CV. Moreover, we have 28,714 Q's on the site from 2020, of which only 197 (0.69%) carry the [rl] tag; it is clear that if askers posted somewhere else instead, the effect on our activity would be very small. – gung - Reinstate Monica Jan 5 at 15:14
• @gung-ReinstateMonica I was not expecting to come to your house, ask for some food (for free) and get it: not everyone is so generous and reasonable. I understand the downvotes, but I don't agree with them, because RL is an AI topic (if not the AI topic). If there's something that should be asked on an AI site is RL. Stats is not an AI site (although some AI topics are on-topic here, of course, because AI also uses statistical methods). As I said in the comments below, there are frameworks for general intelligence, such as AIXI, which are built on top of RL. You should wonder why. – user82135 Jan 5 at 15:17
• @nbro After I posted my answer, you revised the first sentence to "Reinforcement learning (RL) is currently on-topic here," adding the word "currently." What is the intended meaning of "currently" here? Should I understand the emphasized word "currently" to mean that this post is asking about a revision of the stats.SE scope to exclude RL? – Sycorax Jan 5 at 15:25
• @Sycorax I do think that RL is related to some statistical methods. Clearly, almost everyone will consider RL a sub-field of ML and, of course, ML is very related to statistics and I doubt that everyone will ever say that ML is not on-topic here. However, RL is a very specific sub-field of ML, which has been used and studied primarily by AI researchers, such as Richard Sutton, David Silver, and so on. My point is that RL is on-topic here only because ML is on-topic here and not because people interested in statistics are also interested in RL. – user82135 Jan 5 at 15:49
• I don't think you should completely exclude RL as a topic from this site, but, as this post suggests, I do think that RL questions should be asked on AI SE, because 1. people interested in AI are usually very interested in RL (if not just interested in RL), 2. people interested in statistics typically have no interest in RL (as far as I recall), 3. we could have a more cohesive group of people (for instance, recently we get a question on AI SE about hypothesis testing and I suggested the user to ask their question here on Stats SE: that's fine to me because hypothesis testing is stats!). – user82135 Jan 5 at 15:49
• Moreover, in our on-topic page on AI SE, we visibly link to Cross Validated (or Stats SE, or how it is called now) and also suggest where certain topics (that are also on-topic there) are also on-topic on Stats SE. This is what all SE sites should do, but this is another issue. My point is that certain topics are better asked on certain sites than others. RL is one of those. Evolutionary algorithms is another. – user82135 Jan 5 at 15:57
• Statistics, broadly conceived, includes machine learning. And I agree that machine learning includes quantitative questions about RL. So while it's true that not all people interested in statistics are interested in RL, it is true that all people interested in quantitative RL are interested in (a branch of) statistics. My answer has addressed the remainder of your arguments. It's perfectly fine to feel that AI.SE is the best place to ask RL questions, but that doesn't preclude users asking topical RL questions here, because they may disagree with your assessment. – Sycorax Jan 5 at 17:26

No, we should not encourage on-topic posts to be posted elsewhere.

We both agree with this statement.

Reinforcement learning (RL) is on-topic here.

Being on-topic means that RL questions can be posted here; there's no reason to send the question away.

By proposing that RL questions are on-topic here on the one hand, but should be encouraged to go elsewhere on the other hand, OP is suggesting some kind of a middle-ground between "on-topic" and "off-topic." This is harmful, because the disposition of those posts is ambiguous.

• RL posts shouldn't be closed as off-topic (they're on-topic).
• RL posts need not be migrated (they're on-topic here).
• OP could request the post be migrated, for instance if the question draws little interest after some time. (But we already have a mechanism to do that: OP flags a post, and it's migrated if that's suitable.)

The only mechanism to "encourage" a user that they should post elsewhere is to leave a comment on the post to that effect. But this is confusing, because the comment will give the asker the impression that their RL question is not on-topic here, when, in truth, it is on-topic.

The SE ecosystem works differently from the rest of the Internet, and this difference is counterintuitive to new users. I don't think that adding this additional obstacle to users asking questions is productive or welcoming.

New SE sites shouldn't cannibalize old ones.

I support creating new SE websites which extend the usage of the SE software to new topic areas. However, the arrival of the AI.SE website overlaps with stats.SE. This isn't necessarily a problem -- two websites can cover the same topic and coexist. But I think it goes too far to attempt to annex a topic from another website just because one site thinks it's a better home for a topic.

Area 51 was always envisioned as a tool for broadening our scope — for creating new sites serving new topics and answering questions that were previously considered off-topic on our existing sites. Area 51 was never intended as a tool for creating overlapping sub-sites that would cannibalize users from our existing sites!

The problem of site overlap could have been avoided. A hypothetical Area 51 proposal for a website about the philosophical and conceptual components of AI could have been written. If this proposal excludes statistical topics in AI, this website would not overlap with stats.SE.

But now we're in a place where AI.SE has a topic that does overlap with stat.SE.

OP's reasoning is not sound.

However, Stats SE, as we all know, is more statistics-oriented, so there are users here that have no interest in RL

The observation that some topics are not interesting to all users doesn't change the scope of Stats.SE. If we accept this as a standard for determining scope, stats.SE would be fragmented into time-series.SE, clustering.SE, boosting.SE and so on whenever some number of users became uninterested in other on-topic material. This harms users, because expertise will be scattered across a number of sites. Narrowly-construed website topics also create ambiguity about where to put questions that straddle two domains.

To some extent, this is already happening because the Area 51 process allows the creation of new sites that have large overlap with existing sites. This is baffling, but not really something we, as users of stats.SE, can solve.

I also think that having a broad topic area is beneficial. I've read numerous posts outside of my research area, coursework and professional duties. If stats.SE were a more narrow community, I would be worse off because I wouldn't have the benefit of reading these posts from thoughtful contributors. And it's often happened that I can read a post that I come back to when I need to learn about that topic some years later.

AI SE has the highest number of questions tagged with RL among AI SE, Stats SE, and Data Science SE.

The observation that other SE websites have a larger or smaller number of posts about RL doesn't change the scope of Stats.SE.

Moreover, there is a "branding" effect at work. "Artificial intelligence" is a hot topic with a lot of buzz, especially in the popular press. A few years ago, the popular press conferred similar buzz to "big data" and "machine learning." A user with no knowledge of the SE ecosystem might end up at AI.SE to ask a question simply because of the site's name, even if their question already has an answer on stats.SE.

I don't think that counting tags is sufficient to establish where a question is on-topic. AI.SE also includes philosophical questions about the nature of intelligence among its topics, whereas such questions would be off-topic on stats.SE. Because the union of philosophical questions about RL and quantitative questions about RL must be at least as large as quantitative questions about RL on their own, a simple count of tags on two websites is not making an apples-to-apples comparison between the sites.

On the other hand, the logic of counting tags cuts both ways. Clearly, there are large segments of AI.SE which are squarely on-topic here. Stats.SE has far more posts (~16,000 on stats.SE versus ~1700 on AI.SE), and probably similarly for several related topics. Should we divert AI.SE users who ask about machine learning to stats.SE? I don't think this makes sense, but it's the inevitable conclusion if we make topic decisions based on tag counts. And I doubt that the AI.SE community would want to part with a core topic in artificial intelligence.

Finally, the occurrence of more RL tags on AI.SE could stem, at least in part, from a particular user's unilateral effort to leave comments discourage RL questions from being posted on stats.SE.

I've personally noticed that many people interested in RL end up on AI SE and they regularly visit the site, and this includes people that answer RL questions even on Stats SE and Data Science SE

The same users having accounts across one or more of the quantitative fora doesn't change the scope of stats.SE. Probability and statistics questions regularly show up on math.SE; some of them are answered by users who also post to stats.SE. This doesn't change what is topical on math.SE, nor does it change what is topical here.

Although it's true that both AI SE, Stats SE, and Data Science SE are "competing" for users, wouldn't the 3 (overlapping) communities benefit from having a more solid collaboration? I think so.

The communities are not competing for users. Membership or participation in one community doesn't preclude membership or participation in another. I simply don't grant the premise of the question.

The word "collaboration" suggests people working together towards a common goal. What is the common goal that you have in mind here?

The core problem is fragmentation, which we can't stop.

To my mind, the goal of stats.SE is to answer statistical questions (broadly conceived). Fragmenting the statistical field (to include data science, machine learning and AI) across multiple websites seems to work against that goal, and does not promote it.

• Fragmenting the field across an archipelago of quantitative websites makes it more difficult for new and existing users to make use of these websites because they will have to know the exact, arcane delineations between each site's scope.

• Fragmentation makes it harder to find answers, because you have to search in multiple places. To some extent, general purpose search engines (Google, Yahoo!, Bing) can help, but SE has unique search tools for searching within a specific site. I've found these tools are very helpful, but they only search one SE site at a time.

• Fragmentation creates duplicated effort. The same question can be asked and answered in 2 places.

I appreciate that fragmentation is happening already, but it is unresolvable as long as there are SE websites with overlapping content. Because we both agree that RL is on-topic here, there's simply no reason to send on-topic posts elsewhere. On-topic posts are on-topic.

I think it's useful to consider this question from the perspective of minimizing site overlap. A Venn diagram of AI.SE, DS.SE and stats.SE has a lot of overlap, so if the problem we seek to solve is reducing fragmentation, does AI.SE need to exist as a distinct website?

Or, since there is an interest in philosophical questions about intelligent agents, should AI.SE's scope be narrowed to only include questions which are not topical on stats.SE? This solves the problem of topic overlap, while retaining all of the unique content and expertise at AI.SE.

I'm not trying to pick a fight with the AI.SE folks by asking these provocative question, so please don't read it that way. I am trying to tease out what the core issue is. The answers to these questions will provide a better understanding of how to approach topic overlap and mitigate the problems stemming from it.

To me, the root of this problem is that questions about what is on-topic will persists as long as there are multiple, overlapping websites with very similar topics, but with fine-grained distinctions.

Users can choose where to ask their questions.

A user might want to ask about building a fence. This user has a number of places they might ask such a question.

• They could ask on DIY.SE to learn about construction methods or get materials recommendations.

• They could ask on law.SE to learn about potential legal entanglements, such as legal risks arising from a contractor getting injured on their property while building the fence, or the consequences of annexing some of their neighbor's property due to a surveying error.

• They could have a question about how to construct the shape of the fence so that a tethered goat grazes over a specific area. (This example is inspired by an old mathematical puzzle.)

I think users' choices are important to keep in mind, because the decision to post in one place but not another place is a deliberate decision by the user to seek help from a particular community. (Sometimes a user asks an off-topic question, but we both agree RL questions are on-topic.)

What tag-counting does tell us is that users have tended to ask their RL questions on AI.SE. To me, this suggests that the only thing we need to do for AI.SE to continue to attract RL questions is for AI.SE to continue to provide high-quality answers, a task that AI.SE is doing very well today.

• Your reasoning is also not sound. Saying that AI is a buzzword just to help your cause does not make any sense. In fact, AI, as a field, exists since the 50s, so it's more than 60 years that it exists. So, no, it's not just a buzzword. AI is not just about statistical questions, although you could also view the world in the a mathematical way, so you could also ask "Why does Stats SE exist if there's already Math SE?". The idea of having multiple SE sites is to group people interested in more specific topics than the general "Everything SE" would also cover. – user82135 Jan 4 at 20:22
• You then say "The communities are not competing for users. ". This is not true at all. People have limited time, so they need to decide which community to visit. This answer is really bad, not just because of what I've just said but also for several other reasons. Anyway, I was expecting this type of answer. And, of course, I was not expecting anything to improve/change at all. – user82135 Jan 4 at 20:22
• I'm familiar with the history of the term. I didn't say AI is a buzzword, I said AI has buzz in the press, which has no relation to age. It's not hard to find articles in the popular press about how "artificial intelligence" will revolutionize some industry, or change how consumers complete a task, or "make the world a better place." I can't recall ever reading an article about how "statistics" will do the same, even though there are lots of applications for "statistical learning" and other statistical topics. Most popular articles aren't really even about RL-as-AI, but statistics-as-AI. – Sycorax Jan 4 at 20:27
• Most people that worked or are working on neural networks, RL, and stuff like, would most likely call them "researchers in AI" and not "researchers in stats", although I completely understand that a great part of AI is also about statistics. Why is that the case? Because the goal of AI is different than the goal of statistics. In AI, we want to develop intelligent agents, though most people focus on specific topics, such as natural language understanding. This similar to the difference between physics and chemestry. They're all talking about the physical world, but with different perspectives. – user82135 Jan 4 at 20:31
• I don't know whether "most people" would say that or not. Depending on the researcher, they might say "I'm a professor of computer science" or "I'm a professor of statistics," or "I'm a data scientist." But how researchers describe themselves doesn't seem to be the most important consideration in organizing a Q&A website. – Sycorax Jan 4 at 20:35
• It's actually relevant because the website was created to address some purpose and this purpose is the need to answer questions related to some topic and to group people with similar interests. I would like to note I and other people that have suggested that RL questions are more appropriate for AI SE because AI is about intelligent agents and RL is the main way (that we are aware of) to achieve that successfully are doing research on RL or have master's in AI (including myself). So, I'm not here just talking about AI and RL without knowing what I'm saying. – user82135 Jan 4 at 20:39
• In the past, I have also questioned the need for AI SE and DS SE, given the high overlap that they have with Stats SE. However, now, I understand that DS, AI and Stats are all different. We have different purposes. AI is different than DS and Stats. This difference is higher than the difference between DS and Stats. In fact, I would say that DS is just applied ML and stats. AI is about creating intelligence. That's the goal. Creating intelligent algorithms and agents. People in DS and Stats may not care at all about the concept of intelligence, to start with. – user82135 Jan 4 at 20:48
• On the other hand, people that do research in AI are really interested in this concept and question: What is intelligence? How can we build intelligent systems? Every person that researches some AI topic has the dream of seeing/creating an AGI (there are already theoretical frameworks of general intelligence, such as AIXI: btw, AIXI is based on RL framework!!!). That's why the field of AI is also associated with philosophical arguments, such as super-intelligence or the singularity. This has little (or nothing) to do with statistics or data science. – user82135 Jan 4 at 20:56
• I hope you didn't read my answer or comments as any sort of indictment of your expertise. I've enjoyed reading several of your questions and answers, here and elsewhere. But I will point out that RL is on-topic on AI.SE, and that does not preclude RL being on-topic on stats.SE. You're right that philosophical questions (super-intelligence, singularity) are better-suited to AI.SE, but I struggle to see why that's a reason RL questions on stats.SE should be encouraged to go to AI.SE. This is my point about users choosing where to ask questions: users get to decide what kind of answers they want. – Sycorax Jan 4 at 21:02

I think the question is getting ahead of things. I think the first thing you should figure out is: where are RL questions most likely to get the best answer? That's not clear to me. I suspect it is likely to depend on the specific question.

The statistics that 19% of questions on AI.SE are about RL and 0.5% of questions on Stats.SE are about RL don't seem relevant to me. That tells me little about whether any particular question is more likely to get a good answer here or there.

A statistic that might be more relevant is what fraction of questions on RL are answered, on AI.SE and on Stats.SE. When Scortchi did that analysis two years ago, the results were apparently that approximately 58% on Stats.SE were answered and 75% on AI.SE were answered. There are some major caveats, though. For instance, it doesn't measure what fraction received a good answer, and doesn't take into account whether those are a similar population of questions.

I think a more fruitful way to advance the conversation would be through further data and analysis. For instance, are there some categories of RL questions that are more likely to get a good answer on AI.SE than on Stats.SE? Can you gather any data on it, by taking a random sample of 20 such questions on both sites and rating how many received a good answer? That might yield some advice to question-askers that could be useful.

I would be skeptical of a proposal to declare RL off-topic here and direct all such questions to AI.SE. If there was sufficient evidence or consensus, it's possible that there might be utility to some general (non-binding) advice that says "if your question meets these criteria, then it might do better here, if it meets those criteria, it might do better on AI.SE" and leaves it up to askers where is best suited for them. But that would need to be developed and fleshed out further.

I would also encourage you to look for policies that might be a win-win for all sites. Proposing a site policy that appears to benefit one site at the cost of another site does not appear likely to succeed. Since you raised the idea of collaboration, I like to think of "collaboration" as a project that everyone involved views as to their mutual benefit.

I suspect that part of the challenge is that there is considerable overlap in topic areas between folks who work on statistics, computer science, data science, and AI. So, for some subjects, you have multiple communities of people who each have their own perspective and approach to the subject. That might admit any simple bright-line solution involving pushing all such questions to one site or the other.

What I see on this page is well-intended people wasting their time due to SE systemic issues. I believe there are only two solutions:

1. Implement this feature request: Build and strengthen the Stack Exchange community with “crossover questions” between sites
2. Merge AI, CV and DS.
• The first solution may be a good one (and you already suggested it somewhere else), but the second solution is not appropriate, as there are topics covered in AI SE that are not covered in Stats SE and nobody cares about those topics here. To do that, Stats SE would need to loosen up their scope and be about AI too and not only about statistics. I don't think this is beneficial, as AI and statistics have a quite different goal, as fields. – user82135 Jan 12 at 9:52
• AI.SE could change its scope to not overlap with Stats.SE. – Sycorax Jan 12 at 16:06
• @Sycorax The same can be said about Stats SE. I don't think that your approach is productive, to be honest. It would be absurd to exclude, for example, RL from a site about artificial intelligence. Literally, look up books like "Artificial Intelligence: A Modern Approach" (just 37166 citations, according to Google Scholar) to see what topics they cover and also acknowledge its title. – user82135 Jan 14 at 19:02
• @nbro When folks proposed the third iteration on the AI.SE site, a number of people pointed out that it had a large amount of overlap with existing websites. Nevertheless, the proposal made it to the public beta stage. Now, four years later, the AI.SE overlap is a concern for an AI.SE moderator. My approach didn't lead to this problem. – Sycorax Jan 14 at 19:10
• @Sycorax I didn't create this meta-post because of the overlap. Sure, it's somehow a problem that could have a better solution. My main motivation for this meta-post is somehow summarised in the following sentence (from this answer): In principle a question should be placed on the site that is most suitable. I claimed (and showed some evidence) that AI SE is the most suitable place for RL. This doesn't necessarily make RL off-topic here, but being on-topic doesn't mean it's the best place to ask questions related to that on-topic topic. – user82135 Jan 14 at 19:12
• The quoted sentence is well and good as a principle, but it provides no criteria for knowing whether a website is "most suitable" for any particular question or topic. To that sentence, you add a claim that AI.SE is most suitable. You're entitled to your opinion; however, by the same token, other people can ask their RL questions wherever they're on-topic. RL is on-topic here, and people can decide if this is the most suitable site for their RL questions. Citations to popular books about AI don't disprove this claim. – Sycorax Jan 14 at 19:19
• @Sycorax Another book "Reinforcement Learning: An Introduction" (2nd edition, just a few more 40k citations), aka the bible of RL, starts like this: "In this book we explore a computational approach to learning from interaction. Rather than directly theorizing about how people or animals learn, we primarily explore idealized learning situations and evaluate the effectiveness of various learning methods.1 That is, we adopt the perspective of an artificial intelligence researcher or engineer" Why do they do that? Hm. For no reason, "people can still decide what is the most suitable site". – user82135 Jan 14 at 22:53
• I think I have very good arguments to support my claim. I can find many many more, but honestly I don't have time for this (I'm sure you can find them too). The thing is: Stats SE is more about statistics and is not an AI site, and that's what makes RL more suited for AI SE. If Stats was an AI site, then RL would be perfect for Stats, but it's not. Most people here are probably not interested in intelligence, AGI, superintelligence, AI safety, stuff like that. – user82135 Jan 14 at 22:59
• Statistics is significantly different from AI (that's why I think there should be 2 separate sites) and most research in RL is done in AI and by AI researchers (i.e. people interested in intelligence and AGI, and stuff like that). These are facts that you cannot really deny. – user82135 Jan 14 at 23:05
• @nbro (1) I agree that AI.SE has topics that are not on-topic on stats.SE. (2) AI.SE & Stats.SE have other topics that overlap. (3a) A quotation from a book doesn't change what's on-topic on either site. (3b) This particular quotation doesn't illustrate your argument, because it says that the perspective of that book is that of an AI researcher. A different book could investigate AI from the perspective of stats. (4) An asker can choose which community is best for their question, even if you disagree, because the asker knows best what kinds of answers they want. – Sycorax Jan 15 at 0:19
• @Sycorax What does it mean that an asker knows the best what kinds of answers they want? Do you mean that people that answer questions on Stats SE will give a different answer to the same question than people that answer it on AI SE? As I pointed out in my post, in many cases, people that give answers on AI SE and Stats SE are the same. Moreover, in most cases, there's little to argue what the correct answer to an RL question is. If you know the answer, fine, otherwise, you don't know it. Your argument doesn't make much sense, also because, no, in many cases askers don't know what they want. – user82135 Jan 15 at 10:29
• @nbro I don't want to take up too much of your time, but this is a point I make in my answer. Asking a community of DIY hobbyists about how to build a fence is different than asking a community of lawyers about legal issues for building a fence. With respect to RL, you might be right, that for a specific type of question, it makes little difference if you ask a community of AI specialists or statisticians, but if you are right about that, it would seem to negate large portions of your argument. Alternatively, if the community does make a difference for answers, we should preserve that option – Sycorax Jan 15 at 14:13

Questions can be on-topic on multiple sites. In principle a question should be placed on the site that is most suitable.

1. But this can be difficult to determine. There might be multiple reasons why a person asks a question on CV instead of AI, and we should not overrule this when a question is on-topic.

2. Also, it doesn't hurt if occasionally a question turns up on a different site (it may be even beneficial), or gets cross-posted. Already because of that we should not actively promote to move questions that are on-topic here. The point of multiple SE sites is not (I believe) to have sharp boundaries between them. That is not going to be a benefit. To me this idea of cohesive (but separated) communities sounds like the opposite of collaboration.

Alternatives would be to mention ai.stackexchange more often in the comments (currently this only occurs 44 times). Make more references in tag infos (currently this happens for artificial-intelligence). Be more open to cross-posting. (And probably there are many others.)

• I like this answer (especially the part that "In principle a question should be placed on the site that is most suitable"), except for the last part where you say that cohesive communities sounds the opposite of collaboration. Cohesive only means that communities focus more on certain aspects. However, to achieve this, we need collaboration between the different SE sites. I think you're mixing concepts here. So, what I'm saying is that AI is most suitable for RL questions, that's why RL questions should be asked on AI SE and we should encourage users to ask them there. – user82135 Jan 5 at 23:09
• @nbro your idea/method to create more cohesive communities on specific StackExchange sites is by reducing the attractive forces between different sites and have sharper boundaries. This sort stronger cohesiveness within sites goes along with less cohesion in between sites. – Sextus Empiricus Jan 5 at 23:13
• Yes, sure, I'm suggesting that SE communities focus more on certain aspects so that to avoid the high overlap between certain communities (i.e. to have sharper boundaries). This not only clarifies more what certain communities cover (so users seeking help know better where to go) but also groups better people interested in certain topics. For instance, I'm not interested at all in certain statistical concepts, that's why I don't visit Stats SE often. However, there are some questions asked here that could interest me, such as RL questions, but I often miss them. – user82135 Jan 5 at 23:15
• This would be like having a Slack or Discord group where the administrator divides the group into channels and topics. People interested in RL, should go to that room/channel, and so on. This would only have benefits. Note that now I'm not saying that this should be applied only to RL, but other topics too, and would not be applicable only from Stats SE to AI SE but also vice-versa. We could discuss this and other topics, but I guess that people here on Stats are not very open to these ideas, unfortunately. – user82135 Jan 5 at 23:24
• @nbro I wouldn't say that the division of StackExchange into multiple subsites is like the division of a Slack or Discord group into multiple rooms/channels. That division occurs at a different level and happens within a single group/community. It is a bit like the function of 'tags' on StackExchange which allow a division of the sites. Instead... – Sextus Empiricus Jan 16 at 21:43
• ... Instead a better comparison, using the same analogy, would be like different groups (say one on Slack and one on Discord) that decide to make a division into channels and topics in such a way that they do not have an overlap. The division makes little sense, the Slack group uses the topics A till K and the Discord group uses the topics L to Z. – Sextus Empiricus Jan 16 at 21:43