7
$\begingroup$

We've been informally allowing ML implementation questions, software & hardware evaluation questions, and the scope of the humanities side of the field has expanded also...

My sense is, the expansion of scope has been helpful and, in aggregate, welcomed.

  • We're the general AI site, so I feel like pretty much anything we have a tag for, when it's related to AI, is within scope

For example:

terminology should definitely be on-topic and mentioned

hardware evaluation and software evaluation (libraries, frameworks, etc.) questions can be answered objectively and provide valuable information

game theory and extensions I'd personally like to see mentioned

logic seems to me to be fundamental, as does probability

The caveat is that we do want to work in conjunction with the communities with which we have overlap, and support those communities.

We feel firmly that there needs to be a Stack:AI, but we're still in the process of figuring out how to make that permanent, and so we also depend on the support of these related communities.

-----------------------

Because we also deal with the humanities, I'd want to have an explanation of what constitutes a good "soft question".

These are cases where there is not an objective answer, but answers that are sufficiently supported, ideally with citations, are legitimate.

These types of questions are a great opportunity to introduce OP's to fundamental concepts.

| |
$\endgroup$
  • $\begingroup$ I would like to revive this question: ai.meta.stackexchange.com/questions/1320/…. I made some comments to your answers but they were entirely ignored. $\endgroup$ – JahKnows Jun 1 '18 at 9:11
  • $\begingroup$ On the front page of the site today, every single question falls outside the scope of this site. This is normal, most people have questions regarding the implementation of these methods and algorithms. Whereas ethical, moral questions are much fewer and require much more attention when answering. I do not see why these questions of ethics, morality cannot be a part of site which also serves to answer all questions pertaining to Data Science. I would suggest a name such as Data Science & AI. $\endgroup$ – JahKnows Jun 1 '18 at 9:14
  • 1
    $\begingroup$ @JahKnows I'd venture a guess: AI is an established and serious scientific field. Data Science borrows many of its techniques, and so many Data Science-ish questions are on topic here. However, Ethics, Morality, and Philosophy are not part of science, and only part of AI in a fringy sense, despite the popular media attention. I think including them undermines the credibility and usefulness of our site, because they tend to elicit low quality and subjective answers. $\endgroup$ – John Doucette Aug 29 '18 at 11:56
  • $\begingroup$ @JohnDoucette It does seem that this stack has been evolving more towards the applied side of the field, and I think that's positive because it provides utility and is in line with the origin of Stack. Our new math formatting capabilities are clearly reinforcing our capability in this regard. That said, there has been a significant expansion Humanities Stacks, and I do think social impacts of AI is an important topic. The philosophical questions I view more as "fun topics", though the question there may be does it undermine the perceived seriousness/utility of the Stack. $\endgroup$ – DukeZhou Aug 29 '18 at 16:51
  • 1
    $\begingroup$ @DukeZhou That makes sense. I think that with some careful moderation the soft questions can have a lot of utility, but because the ones that are further out tend to attract a lot more attention, we ought to be pretty careful about them. $\endgroup$ – John Doucette Aug 29 '18 at 20:25
4
$\begingroup$

Similar questions come back on Meta, but no convergence.

I am a proponent of technical questions since before the exchange creation. The hairy issue is to clearly define the boundary.

Any kind of technical question will lead to an overflow of simple programming questions on how to do something with Tensorflow or Pytorch. Such questions are (in my opinion) better answered on StackOverflow. These frameworks are still complex enough so as many questions are really about syntax and framework-specific understanding (e.g. I concieve it is hard to use TensorFlow if you have never used graphs or data flows).

Technical questions like "how many layers to do something?", "what architecture is best for mushroom recognition?", or "why SVM here and ANN there?" seem fine to me.

All in all, I expect the community manages to still attract questions about consciousness, AGI, ethics, etc. A tsunami of small technical questions is good for traffic, but causes a low signal/noise ratio.

| |
$\endgroup$
  • 1
    $\begingroup$ Eric, thank you for weighing in. These are valuable insights. $\endgroup$ – DukeZhou May 30 '18 at 16:25
  • 1
    $\begingroup$ "Technical questions like "how many layers to do something?", "what architecture is best for mushroom recognition?", or "why SVM here and ANN there?" seem fine to me.". Wouldn't the Data Science site be way better suited for such a question? Choosing such parameters is a matter of the data at hand, not any aspect of intelligence. Artificial Intelligence is a misnomer. I believe that this problem can be mended by combining the Data Science and Artificial Intelligence sites. Both communities would benefit in this regard. $\endgroup$ – JahKnows Jun 1 '18 at 9:09
  • 1
    $\begingroup$ I agree with Eric. There's always going to be the problem of newbies coming to AI to ask, and these more general questions can be used to bring Data Science to their attention. Part of the reason I feel this way, speaking as a generalist, wading into the hard science stacks can be not only intimidating, but all too often fruitless for a newbs. I think there's probably value in AI as a friendly intro site. $\endgroup$ – DukeZhou Jun 2 '18 at 2:49
  • 1
    $\begingroup$ I think it's also valuable for the general AI enthusiast public to see some DS and CV and SO and CS and Math questions over here, so they can understand what constitutes the field of AI. $\endgroup$ – DukeZhou Jun 2 '18 at 2:52
  • $\begingroup$ @JahKnows I see what you say as valid. There is a fine line how questions get asked, though, and this becomes case by case. If the question is really about the data, then I would vote for migrating to DS. If not, probably keep it and answer. A bit like a distinction between DS and "Data Engineering". I am half-saying that DE should be here, like in "how to make a brain". $\endgroup$ – Eric Platon Jun 2 '18 at 13:16
  • 1
    $\begingroup$ @DukeZhou I start really liking the idea of AI.SE as a hub site, with AI-specific questions, and proper forwarding when more specialization is due. Fine line on some cases, but it looks feasible and valuable to me as for now. $\endgroup$ – Eric Platon Jun 2 '18 at 13:17
3
$\begingroup$

This site is about Artificial Intelligence (AI) which generalizes Machine Learning and Deep Learning:

enter image description here

Hence, I think, the site should embrace and be the home questions about any of those. Both practical and theoretical, science and engineering.
In order to do so and bring this great audience we should:

  1. Change the name of the community into Artificial Intelligence and Machine Learning.
  2. Write explicitly in the site description that it deals with those subjects and welcome questions about them.

Doing so, I believe, will fill the void in the SE communities which doesn't dedicate any community to gather people which are experts on those.

Remark
Image taken from the book Francois Chollet - Deep Learning with Python.

| |
$\endgroup$
  • 1
    $\begingroup$ I'd also mention that incorporating Machine Learning into the Stack name might increase our traffic, but changing that is like turning a ship--requires a great deal of consensus. That said, I do think we need to revise the guidelines to at least reflect the current scope of the Stack. $\endgroup$ – DukeZhou May 12 '18 at 2:47
  • $\begingroup$ Definitely agree of changing the site description, more importantly the scope definition. As for the title, I cannot understand the suggestion. The image shows inclusion, and you suggest putting them both in the title. (1) long title, less impact, (2) deep learning is really what people talk about (letting aside whether they understand), (3) I hope the community is really about AI at large---not only ML, (4) ML in the title confuses when considering Data Science, Cross-Validated, and probably some other exchanges. $\endgroup$ – Eric Platon May 29 '18 at 9:48
  • $\begingroup$ @EricPlaton, The image shows the Hierarchy between the subjects. The title is about making the community attractive to more people. The title isn't long in my opinion. Yet the inclusion of Machine Learning will do miracles to the number of visitors. $\endgroup$ – Royi May 29 '18 at 10:15
2
$\begingroup$

I'd personally like to expand the guidelines to formally include:

  • a specific AI programming problem, or
  • an AI software algorithm, or
  • AI software tools commonly used by programmers; and is
  • a practical, answerable problem that is unique to AI software development

which is basically Overflow with "AI" added to each line.

WHY?

My main competency is in the humanities side of the AI equation, but I don't think it's possible we'd going to be able to sustain the level of activity to graduate from Beta on philosophical and conceptual questions alone. And, I'm inclined to believe that AI is a field where the humanities and sciences intersect.

When I first came on as mod, there was a flood of Python question related to AI development. It seemed clear that these endeavors constitute a relatively new sub-field. So while I'd point someone with a general Python question to Overflow of Computer Science, if that question relates to AI, I think it belongs here. That's just one example.

| |
$\endgroup$
  • 1
    $\begingroup$ Thank you for the perspective. That goes in an interesting direction to me. I guess a future question would be if a question is about Python/Tensorflow, is it an "AI" question because the question is about TF ? Perhaps questions on how to use the API go to SO, and how to implement an AI algo can go to AI.SE. $\endgroup$ – Eric Platon Jun 2 '18 at 12:57
  • $\begingroup$ @EricPlaton My feeling is that if the Python or Tensorflow question is related to an AI implementation, we want to address it here. (There were a flood of python questions when I first started modding, no doubt due to AlphaGo, but my feeling is there is enough activity in AI specific software development to support specialization re: AI-specific implementation question and problems.) $\endgroup$ – DukeZhou Jun 4 '18 at 20:37
  • $\begingroup$ Difficult and hairy. A question about how to do something in TF, because TF is the assembly language for ML would be better on SO. Agreed that a more abstract question with code would be terrific here. $\endgroup$ – Eric Platon Jun 9 '18 at 5:47
1
$\begingroup$

I preemptively modified the guidelines just now to make it clear that reference requests are on-topic. (We have a tag for it, and reference requests have utility and traffic-drawing value.) The idea is that experienced contributors can suggest reference materials with some vetting and, ideally, context and synopsis.

We also have software evaluation and hardware evaluation tags, and I'd like to add these officially as well because here there can be a great deal of objectivity. (i.e. processor performance can be precisely quantified, and functions related to AI development explained. Likewise, with software utilities, functions and capabilities can be accurately listed and broken down.)

AI Career Advice I strongly feel this should be on-topic. While it's typically the type of thing one undertakes on chat, most chat participation is low, and good luck finding someone who can give you advice in any given span. But AI has never been more burgeoning as a field, with opportunities for the average programmer in addition to PhD's. A lot of people want to get into the field, and advice from professionals and scholars would be salient, beneficial, and potentially boost activity/engagement with answerable questions.

| |
$\endgroup$
  • 1
    $\begingroup$ Reference requests... Hairy topic, but I also see it working. $\endgroup$ – Eric Platon Jun 2 '18 at 13:05
  • $\begingroup$ Career advice? If this is typical of chats, perhaps keeping them there is better (and traffic will increase as chats). I am currently on an opposing stance, but out of curiosity, is there any career advice on SO ? I've never realize, and I think that is better this way (there are so many ways to make a career). $\endgroup$ – Eric Platon Jun 2 '18 at 13:08
  • 1
    $\begingroup$ @EricPlaton Re: career advice on other stacks, that might be worth investigating. My feeling is expanding to allow this type of "soft question" could boost traffic and engagement if questions were allowed, and would have utility value if the advice came from experienced parties. (With everyone and their brother trying to capitalize on the renewed interest in the AI field, there are all kinds of pitfalls. I know someone who spent ~15K on a "bootcamp" for AI, and was only able to find a job as a SQL admin.) $\endgroup$ – DukeZhou Jun 4 '18 at 20:44
  • $\begingroup$ My problem with a traffic driven tactic is allowing now is like allowing forever. Hard to rollback. I’ll let more people weigh in, as I tend to exclude it. $\endgroup$ – Eric Platon Jun 9 '18 at 5:50
  • $\begingroup$ @EricPlaton very true. $\endgroup$ – DukeZhou Jun 11 '18 at 19:42
1
$\begingroup$

I don't see a discussion of what constitutes a good "Soft Question", but @DukeZhou's suggestions make sense:

  • Questions should be rooted in existing AI research, or research by serious philosophers on AI related topics, not in popular non-fiction books. (i.e. favour Moshe Verdi or Nick Bostrom over Ray Kurzweil).
    • Rationale: Popular non-fiction tends to exaggerate AI's capabilities, and tends to be written by people with little actual knowledge of the field, despite reaching a broad audience. Questions rooted in this material will tend to elicit wildly speculative answers, or to be unanswerable.
  • Soft questions and their answers should include supporting citations to scholarly works, and should be rooted in empirically supported facts whenever possible.
    • Rationale: A good example was a recent question on automation. It's easy to speculate, but there's actually lots of good data, both about what financial markets think will be automated, and what AI experts as a whole think can be automated. These estimates are likely to be far more reliable than an individual user's opinions, or even any philosopher's opinions.
| |
$\endgroup$
  • $\begingroup$ John, thanks for weighing in. I agree wholeheartedly. (I'd noticed that many hard science stacks have the "soft question" tag , and our guidelines should reflect the guidelines on those communities.) $\endgroup$ – DukeZhou Aug 29 '18 at 17:02
  • $\begingroup$ I also created a "mythology-of-ai" tag to identify the fringey questions, including the singularity, but it hasn't gotten much use. My thought there is that the Carl Sagan/Cosmos initiative to popularize physics and astronomy probably had benefits for the field. These types of questions surely help with traffic, but is it the right kind of traffic? That said, there may be value in demystification of these topics, separating reality from speculation. Another way to look at it may be "Was Asimov useful?" $\endgroup$ – DukeZhou Aug 29 '18 at 17:09
  • 1
    $\begingroup$ @DukeZhou That makes a lot of sense to me. I like that tag a lot, and will be sure to suggest it in future when I see questions like it. $\endgroup$ – John Doucette Aug 29 '18 at 20:23
1
$\begingroup$

Stackoverflow has also an Artificial Intelligence tag. But posting a question there is often not the best choice, because AI-related questions are too complex for Stackoverflow. I would guess that it is question of which questions fits to Stackoverflow and which to AI.SE. At the end, one of both websites is always right. If somebody has an issue with Python, Neural Networks or C++, it is only a question of which of both forums is the right choice. It is correct, the machine learning implementation problems with Tensorflow and Co are exactly on the border. And a framework like OpenAI gym can be discussed on SE.AI but also on Stackoverflow. I would say, that AI.SE can allow any question and this one with a low academic impact should be migrated to Stackoverflow.

SO vs. AI.SE

One remark to Datascience

I see the point, that Datascience has a tendency to overwhelm AI.SE. If someone is using a three layer peceptron for creating a mathematical model in Matlab he is forwmost not interested in Artificial Intelligence but it is plain old statistics question. On the other hand, neural networks were clearly invented in the AI domain, so it is wrong to answer this question with: “You're in the wrong forum”. In contrast to the majority here in the forum, I do not believe that we should migrate such questions to Datascience, because it allows us to explain the AI-related point of view to neural networks and how this technology can be used to implement intelligent machines.

| |
$\endgroup$
  • $\begingroup$ I think we're sort of in accord in that context is important. I like overlap questions on SE:AI because I think it's probably useful, especially for people new to the field, or without an academic grounding, to see "how the sausage is made" (all of the subfields that, together, comprise the incredibly broad, continually evolving scope of artificial intelligence.) $\endgroup$ – DukeZhou Sep 4 '18 at 19:46
0
$\begingroup$

The Data Science site already covers such topics. It is in my opinion that the Artificial Intelligence site and the Data Science site should be merged where the scope would include

  • The humanities of artificial intelligence (ethics, morality, etc.)
  • The humanities of data collection and privacy (ethics, morality, etc.)
  • The discussion of state-of-the art research in the field of artificial intelligence, machine learning and data science.
  • Questions pertaining to the implementation of techniques and methods that can be used to achieve artificial intelligence (there are very few of these).
  • Questions pertaining to the implementation of machine learning techniques and methods (Bayesian models, trees, neural networks, deep learning, etc.).

A site which combines both Artificial Intelligence and Data Science would have many benefits:

  • A wider audience of potential answerers such that individuals may have a higher probability of find resolutions to their queries. For example a deep learning question asked on either of the sites only, will not reach as many answerers, this hurts the questioner's chances of getting the best possible answer.

  • The possibility of people with a strong implementation background whom are more likely to peruse Data Science, to also be involved in discussions regarding the ethics and morality of artificial intelligence.

  • The possibility of those more interested in the humanities of artificial intelligence to see the kinds of problems that machine learning algorithms are capable of solving and forging stronger arguments about the ethical use and morality of artificial intelligence.


In my opinion, artificial intelligence does not yet exist, very fancy computational models which are essentially hyper-plane separators are not intelligent. However, due to the misnomer used in the medias for machine learning, artificial intelligence is used to describe these techniques.

As a result, many questions on the Artificial Intelligence site do not match the intended guidelines of the site. Most questions on any particular day do not belong on this site and should be migrated to Data Science. I propose the sites be merged into a single site.

I really do like the questions asked on the Artificial Intelligence site and I would love to partake in them. However reading through Stack Overflow and Data Science usually occupies most of the time I want to spend on my couch. Furthermore, I often see questions in Artificial Intelligence that are almost mirrors of those that have already been answered in great lengths in Data Science. Specifically those relating to neural networks, backpropagtion or gradient descent.

I would ask kindly for the mods of this site to consider that in unity we are all stronger, in division we fall.

| |
$\endgroup$
  • 1
    $\begingroup$ Thanks for contributing. I can tell you there's no way DS would merge with us, and the main problem there is that what is covered by DS is just a fraction of the scope of AI. (We don't actually get so many DSish questions compared to Overflow or Cross Validated.) But the scope of AI also includes social and ethical issues, history and mythology of AI, terminology and concepts, in addition to the technical stuff. $\endgroup$ – DukeZhou Jun 1 '18 at 19:41
  • $\begingroup$ @DukeZhou, Actually when I proposed this idea on the Data Science meta people were very much open to it for the above reasons. I do not see why Data Science couldn't also include questions regarding ethics, history and mythology of AI. $\endgroup$ – JahKnows Jun 2 '18 at 0:59
  • 1
    $\begingroup$ @DukeZhou, "(We don't actually get so many DSish questions compared to Overflow or Cross Validated.)". Right now the front page is littered with questions that are outside the scope of this site. The only reason people choose to post here is due to the misnaming of machine learning in the media as artificial intelligence. Even though everyone in the field agrees that it is hardly artificial intelligence to fit parameters using an optimization function. $\endgroup$ – JahKnows Jun 2 '18 at 1:02
  • $\begingroup$ It's interesting because back in the 80's, they were calling what Id was doing with 3D games virtual reality. Now we'd laugh at the idea, but it still constitutes a virtual reality. Machine Learning is the hot field due to recent milestones, and is almost certainly the future of AI, but it didn't arise out of nowhere. There are an array of semantic that have to be navigated to say "only Machine Learning is AI". $\endgroup$ – DukeZhou Jun 2 '18 at 1:06
  • $\begingroup$ The other thing is, Data Science is very specific. It's the statistical part of the equation that ties in machine learning. There are certainly social issues that arise out of statistical modeling, and AI behavior {See: Unchained: A story of love, loss, and blockchain} but I don't see any reference to that in the DS guidelines. $\endgroup$ – DukeZhou Jun 2 '18 at 1:11
  • $\begingroup$ @DukeZhou, I am not claiming machine learning to be AI. On the contrary, machine learning is hardly AI. A perceptron in the 60s was never considered to be intelligent, it was just a mathematical function that gave an output from a set of inputs. Simply stacking these functions does not generate some intelligence, it just adds layers of complexity to the function. The inventors of backpropagation openly admit that they regret the high performance of their algorithm because it will stunt the development of real AI. $\endgroup$ – JahKnows Jun 2 '18 at 1:50
  • $\begingroup$ @DukeZhou, again, I do not see why the Data Science site and the AI site cannot exist as one, where the ethical, mythological, etc... be added to the technical. This would make it so much better. We'd have a wider audience answering technical questions for all the noobs who think AI is the technical term to describe a neural network. And we'd get interesting insight of the potential risks associated with this technology from a philosophical stand point. $\endgroup$ – JahKnows Jun 2 '18 at 1:52
  • $\begingroup$ "Data Science is very specific. It's the statistical part of the equation that ties in machine learning.". I do not think this is a good description of data science in the least. Data science is the larger umbrella which contains, data generation, data storing, and the utilizing of data to solve problems (machine learning, statistical analysis, etc.). Consider what the job of a Data Scientist entails. $\endgroup$ – JahKnows Jun 2 '18 at 1:54
  • $\begingroup$ @DukeZhou, we can really do much better as a community if we work together then if we keep separated and continue discussing semantics. Objectively speaking, the AI site right now has very few questions that match the scope, most should be migrated to the Data Science site. The questions regarding the ethics, mythology, etc... are very interesting and I do not see how they cannot be integrated into a technical site. $\endgroup$ – JahKnows Jun 2 '18 at 1:56
  • $\begingroup$ I agree with you 100%. I strongly believe SE:AI must be a feeder site to DS because we're always going to be the first place many come to ask. While I'd like to see basic questions answered here, with an emphasis on the concepts, those answers should point to DS for more advanced followup. Most importantly, we do need to start migrating DS questions to DS! I just opened a meta for a DS migration campaign $\endgroup$ – DukeZhou Jun 2 '18 at 2:41
  • $\begingroup$ Please migrate! I'll vote and flag on that vein. $\endgroup$ – Eric Platon Jun 2 '18 at 13:02
  • 1
    $\begingroup$ I do not think DS and AI should merge, so each can keep their focus. Ethics on DS is terrific, given all the weapon of Math destruction that come online. However, mythology, history, philosphy, and topics like conciousness would be noise on DS (with rare exceptions). May AI be the home to these questions. That was the intent in creating it. $\endgroup$ – Eric Platon Jun 2 '18 at 13:03

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .