3
$\begingroup$

Because of a comment added to someone else's question about it being off topic, I looked up the definition of what is on topic for the AI beta.1

I was startled to find that the current Artificial Intelligence SE site description is this.

Q&A for people interested in conceptual questions about life and challenges in a world where "cognitive" functions can be mimicked in purely digital environment

What I would have guessed it to be based on actual Q&A content is this.

Q&A for people interested in conceptual, mathematical, design, and approach questions related to the creation, use, and cultural impact of artificial intelligence and cybernetics

Below are reasons why that second phrase is more descriptive of what is actually discussed and accepted as Q&A. The evidence for the below reasons is clearly evident not only in the titles and bodies of the most popular questions and answers but also in tag usage, the top ten being these.

  • Neural networks
  • Machine learning
  • Deep learning
  • CNNs
  • Reinforcement
  • AI design
  • Image recognition
  • Algorithm
  • Classification
  • Training

Less than 1% of the content is about life and challenges in world where cognitive automation is emerging. It is likely that many of our members couldn't distinguish a cognitive function from either first order predicate logic or learned routine. The later two can be simulated in many respects with current digital systems, but it is unclear whether, in this century, cognition will be simulated. I know that's not the popular conception, but those with the greater experience that have been watching AI and working in the field for decades know it to be the truth.

Here are my reasons why I think the current AI meta site definition is not consistent with either expectation, actual content, or the real interests of the membership.

  • A large proportion of the most active and interesting questions are about what can and has been done in the area of machine learning.
  • Robotics appears and should appear in the content for the AI beta. To begin with, the field of AI was as much born out of the need for control systems faster than humans to defend against attacking aircraft and intercontinental ballistic missiles than as a way to do proofs using formal logic. More importantly, the long awaited for emergence of autonomous vehicles, automated vacuum cleaners, and a host of other non-military intelligent control systems is no longer held back by the prices of control system components (CPU, motor drivers, memory, operating system) and the applicable machine learning strategies are now on GitHub and packaged in Python and Java libraries.
  • Many of the current AI beta Q&A are lacking in scientific rigor even though the AI beta is in the Science SE category. The use of mathematics is a quality factor in a science site as much as inclusion of academic references or narrowness of the problems set forth in the questions. I think it is correct to assemble AI under Science and not Technology because the technology side is covered under SE sites such as Arduino and Data Science, which are properly placed in the Technology category.
  • I don't see a membership-wide interest in, "Questions about life and challenges in a world where "cognitive" functions can be mimicked in purely digital environment." People are aware that some job functions get replaced in the human job market by machines, and we've been, as a culture, adapting to it since the late 19th century. Industrialization saw the emergence automated cotton picking, textile manufacture, type setting, and electronics assembly. Most people, myself included, will actually be happy when office worker and programming is automated. It's boring and no more psychologically healthy than coal mining was healthy for breathing. Our members are not here not to ask questions about how to live or face the changes but to engage in right-now-present-day adaptation to what is so obviously the nearing of another big job market swing.
  • For reasons given above, "Cognition," is not the best choice of words to describe the relevant AI research and development under way today. Notice that the word cognition or questions surrounding it are rare in tagging, titling, and discussion. What is currently being synthesized is the lower mammalian functions that do not take place in the cerebral cortex and have no relationship with comprehension, discernment, or insight. AI beta Q&A does not fit into the cognitive science definition of cognition or the dictionary definition of it. For instance, character recognition, visual collision detection, identity recognition, and such are not cognitive functions. Artificial analysis of large data sets, not for training but for feature extraction, is not cognitive either.

There's much more pointing to the inadequacy of the current description but I'll stop there. Back to the central question.

Is the current out-facing description of the AI meta descriptive of what it is?

The co-question is this.

Is discussion about life in a changing world really what is relevant to most people who would search for Artificial Intelligence in the search field of SE? And if not, shouldn't we adapt to the real interests of our membership?

There was an older AI beta that had a great description for what the current AI beta does, which failed as an SE beta probably only because it was before its time.

Q&A site for theorists, system architects and analysts of intelligent machines and software


Footnotes

[1] I had to log off to find the AI beta description, which is a bug report I might make, but not directly related to this question. It can be seen when one acts as a non-member and looks up "Artificial" in the SE site list using the search field.

$\endgroup$
  • $\begingroup$ This stack itself is very problematic with a few core group of regular users...so there is no actual "interest of members"....somehow if you look at the stacks older questions you can see a lot of participation from well reputed users...for some reason they stay away from this stack now...otherwise you would have got a lot of answers to your theoretical posts $\endgroup$ – DuttaA Aug 4 '18 at 14:07
  • $\begingroup$ @DuttaA I suspect there actually is a significant amount of interest in a place like this for questions that are slightly on the technical side of AI, but dont have enough focus on programming for stackoverflow or enough focus on stats/math for crossvalidate.se or math.se. Questions like "how does this algorithm work in this particular case?", "does this algorithm suffer from problem X or Y?", etc. $\endgroup$ – Dennis Soemers Aug 5 '18 at 11:26
  • $\begingroup$ I have personally answered many questions like that over on StackOverflow (even though they tend to officially be off-topic there), and seen many others get closed. I don't think the problem is a lack of interest, the problem is that most of such users find their way to StackOverflow first, and don't get redirected here. It doesn't help that AI.se is not one of the options to select to migrate to when flagging questions as off-topic on StackOverflow. $\endgroup$ – Dennis Soemers Aug 5 '18 at 11:27
  • $\begingroup$ @DennisSoemers have you notified to the stack admins of this problem? $\endgroup$ – DuttaA Aug 5 '18 at 11:42
  • $\begingroup$ @DuttaA No I haven't. I'm not sure if I'm necessarily the person who should? I can try to compile a good list of examples questions on StackOverflow for which I feel they'd be a better fit on AI if that's helpful though. $\endgroup$ – Dennis Soemers Aug 5 '18 at 11:45
2
$\begingroup$

The evidence for the below reasons is clearly evident not only in the titles and bodies of the most popular questions and answers but also in tag usage, the top ten being these.

  • Neural networks
  • Machine learning
  • Deep learning
  • CNNs
  • Reinforcement
  • AI design
  • Image recognition
  • Algorithm
  • Classification
  • Training

All these topics are on-topic on http://stats.stackexchange.com and https://datascience.stackexchange.com. I don't see any point in having https://ai.stackexchange.com covering them as well.

$\endgroup$
  • 1
    $\begingroup$ @DouglasDaseeco I am okay with slight overlap. But as the list of top 10 tags you have listed show, the vast majority of questions (>80%) on ai.stackexchange.com are very much on topic on stats.stackexchange.com and datascience.stackexchange.com. I believe this is why the scope on meta was worded to try to avoid so much overlap. $\endgroup$ – Franck Dernoncourt Aug 9 '18 at 1:22
  • $\begingroup$ @FranckDernoncourt I personally feel like there's value in... "findability". For people who do not already happen to know that there exists a site named "CrossValidated" or "stats.se", I suspect many people interested in these topics are going to have an easier time finding a site named "Artificial Intelligence". I don't know if that's an "officially acceptable" reason for a new site though, just something that I personally think has value. $\endgroup$ – Dennis Soemers Aug 9 '18 at 10:55
  • $\begingroup$ Either way, on the more technical side of questions, there are also still topics in AI that don't fit into either stats or datascience. For example, search, planning... maybe genetic algorithms / evolutionary stuff, depending on who you ask? Not sure. $\endgroup$ – Dennis Soemers Aug 9 '18 at 10:56
  • 1
    $\begingroup$ @DennisSoemers true, there are some questions on AI that aren't stats or DS but in practice, most questions here are about stats or DS. Evolutionary computation is definitely on topic on stats SE. I agree the name CV sucks. $\endgroup$ – Franck Dernoncourt Aug 9 '18 at 20:10
  • 1
    $\begingroup$ @FranckDernoncourt, I'm not sure that you're aware that the proposal and selection of a new AI site description has been resurrected by Ben and others at ai.meta.stackexchange.com/questions/1430/…. It would probably be helpful to see your view in the comments there. For instance, it may be valuable to see your point about the overlap with Stats and DataScience sites, not well represented in that Q&A. Some involved have probably not read this Q. Also, if you have a proposed wording, I for one would like to read it. $\endgroup$ – FauChristian Sep 29 '18 at 22:11
  • 1
    $\begingroup$ @FauChristian thanks, I added an answer but the main issue is that since we already have CV and DS it doesn't make much sense to me to add one more AI-focused SE. $\endgroup$ – Franck Dernoncourt Sep 29 '18 at 22:58
  • $\begingroup$ Your point is clear and astute. CV & DS do cover data center AI, reflecting Google's huge success with mining, ranking, & indexing. Yet, data intelligence is subset of smaller proportion than public media represents. The foundation of AI is not data-centrism but rather cybernetics, kick started by cold war funding of ICBM countermeasures under Reagan. Concurrent with data-centrism is the serious investment into real time embedded AI for transportation, factory defect reduction, and smart home robotics. Automated mathematics (Leibniz) is also well funded & lacks data in the conventional sense. $\endgroup$ – FauChristian Sep 29 '18 at 23:57
  • $\begingroup$ Even so, I think your concern needs to be heard clearly, thus my earlier comment. $\endgroup$ – FauChristian Sep 29 '18 at 23:58
1
$\begingroup$

Is the current out-facing description of the AI meta descriptive of what it is?

I think the answer to this is a fairly obvious "no" at this point in time.

The co-question is this.

Is discussion about life in a changing world really what is relevant to most people who would search for Artificial Intelligence in the search field of SE? And if not, shouldn't we adapt to the real interests of our membership?

This is in my opinion the much more important question. Again, I think the answer is "no". Those topics may be interesting and relevant for some, and it's fine to allow them, but I suspect that much more detailed questions about specific little things in AI are more relevant to more people that happen to find their way onto this site. In my opinion, the description should indeed be adapted to allow more "technical" questions... basically, allow the kinds of questions we see many of. Not necessarily technical in the sense of "why doesn't this snippet of code work", but technical in the sense of "how/why/in what cases should this part of an algorithm work?"


A few minor nitpickings from me:

Many of the current AI beta Q&A are lacking in scientific rigor even though the AI beta is in the Science SE category. The use of mathematics is a quality factor in a science site as much as inclusion of academic references or narrowness of the problems set forth in the questions.

I don't agree that this is a problem. StackExchange as a whole (all sites across the entire network) tends to be primarily about "quick" questions and answers, about building a site that people can easily reach through google searches, quickly see a question relevant to their search terms, and quickly find an answer that addresses their needs.

Most questions really don't need answers with a thorough literature review, like a scientific paper would. Some do, sometimes there'll a really great question that is best addressed with a great answer containing interesting references to literature, etc... kind of like how, on StackOverflow, you'll sometimes find great answers with lots of different possible solutions, a lot of work put into timing the different implementations, explaining observed performance differences, etc.

That's certainly not necessary for the majority of questions though. Many more questions are asked by non-experts, or first-year or second-year students for example. They might use slightly incorrect terminology, not be aware of all kinds of other potential solutions, etc. But when they have a clear question about an algorithm they're learning about, they just need an answer to that, they don't need a thorough literature review.

I think it is correct to assemble AI under Science and not Technology because the technology side is covered under SE sites such as Arduino and Data Science, which are properly placed in the Technology category.

I don't agree with the bolded part there. I've personally never heard of Arduino, but a quick google search does not tell me how that covers a major part of AI at all, it seems really specific and niche. AI is also much much more than just Data Science. AI includes things like search algorithms, planning, pathfinding, and probably much more stuff that is not Data Science. People need a place to ask questions about all that, and it's not covered by any other StackExchange site.

$\endgroup$
  • 1
    $\begingroup$ @DouglasDaseeco That is a great question... to which I unfortunately don't know the answer $\endgroup$ – Dennis Soemers Aug 5 '18 at 11:28

You must log in to answer this question.

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