4
$\begingroup$

This is in relation to comments on: Differentiable activation function

I made the point that the question seems to fit into the "conceptual aspects of AI" covered by this stack, but T.C. countered that Machine Learning questions, in particular, are already quite fractured across several sites.

How can we reconcile this so that the related Stacks support and add value to each other?

I personally would welcome guidance from trusted contributors and mods on the related Stacks.


As an analogy,there is a relationship between the Humanities Stacks Mythology, Literature, Latin and Philosophy (in addition to others such as History.). Different aspects of a single topic are best addressed in the forums where contributors have the relevant strengths. My point is these are subjects where a fuller understanding requires many fields.

I see this as one of the main strengths of Stack in an information explosion era with so many fields and subfields. Specifically that we can, and should, be walking "across the hall" to take advantage of the breadth of competencies Stack offers.

Part of my inclination may derive from having been in an interdisciplinary studies program as an undergraduate. In that program, we did not learn Science independently of History, Philosophy, Psychology, Art and Literature. Rather, these subjects were taught in tandem.

| |
$\endgroup$
  • 1
    $\begingroup$ I get the sense that this site may have moved too soon to close questions and define a strict scope early in the private beta. Something that SE should do a better job of communicating during the private beta is that while you can always close questions if they turn out to be a problem down the line, it's harder to recover from pushing certain topics away that later turn out to be very useful. $\endgroup$ – user6698 Jul 30 '17 at 22:26
  • 2
    $\begingroup$ I also get the sense that if AI has problems, it's biggest problem is a lack of expertise. But of course, that lack of expertise could be caused by a scope that pushes away experts or draws people lacking experience. $\endgroup$ – user6698 Jul 30 '17 at 22:28
  • $\begingroup$ @Hamlet Definitely an imperative of this stack should be to attract contributors with experience and expertise. This meta is partly in response to earlier critiques in the same vein. $\endgroup$ – DukeZhou Jul 31 '17 at 2:40
4
$\begingroup$

I made the point that the question seems to fit into the "conceptual aspects of AI" covered by this stack, but T.C. countered that Machine Learning questions, in particular, are already quite fractured across several sites.

I believe most ML questions are on CV. Then DS got created, which has a huge overlap with CV, and a more trendy name. So one way to avoid fracture is not creating new Stacks with huge overlaps (Are all questions asked on stats and data science SE also on topic here?).

How can we reconcile this so that the related Stacks support and add value to each other?

Build and strengthen the Stack Exchange community with "crossover questions" between sites

Part of my inclination may derive from having been in an interdisciplinary studies program as an undergraduate. In that program, we did not learn Science independently of History, Philosophy, Psychology, Art and Literature. Rather, these subjects were taught in tandem.

In practice, the development of AI models doesn't care much about History, Philosophy, Art and Literature. Most AI experts focus on the models, which tend to be statistical, therefore on-topic on CV.

| |
$\endgroup$
  • $\begingroup$ Thanks for this enlightening answer. I am aware that the humanities don't bleed into AI unless we're talking AGI, but it seems to me combinatorics, game theory, and several fields of mathematics relate to AI development in general. What I was really trying to get at is that the separate AI stacks should be more related to promote cross-stack participation. $\endgroup$ – DukeZhou Aug 14 '17 at 20:18
1
$\begingroup$

The condition is related to the beta process definition and incentives built into the back end rules and user interface rules. These are created and maintained based on the analysis of trends and the projections of that analysis by the owners of the system upon which the domains stackexchange.com and stackoverflow.com sit.

Members, especially moderators and even more so diamond moderators, can mitigate the inevitable chaos that forms in any large account based network by choosing names and definitions that are likely to disambiguate options that users have. People can also request features and enhancements that may modify incentives in positive ways.

To meaningfully do any of these things it is important to understand that knowledge is segmented in some ways and homogeneous in other ways. Forcing questions into clean compartments is not even done at universities with curricula. In fact, trends toward interdisciplinary work are usually found at the most progressive universities and offered to the highest performing students.

The natural overlap of human discovery and achievement cannot be changed by any web site incentives system. Even the extreme measures of totalitarianism, jihad, or martial law are unlikely to bring about compartmentalized knowledge, mostly because smart people won't put up with it and will literally shoot back if pushed too far.

Artificial intelligence was born of interdisciplinary thinking, and is bound only by two things.

  • It concerns primarily what can be artificially created
  • It concerns primarily how to make choices that produce better results than arbitrary selection

Some may argue this. I won't because I've heard all the arguments otherwise, and they lack merit to the degree that further response is ... .

Regarding the current machine learning trend, it is primarily social and economic phenomenon that may or may not sustain. Recognizing that what goes up often comes down is another key to making choices today that we don't regret later.

At one time, stone work was a technology that bled into every topic. In 100 years, one may not be able to find the phrase, "Machine learning," in a recent piece of media. Perhaps nanotech-genetic portals might have become the craze, where people are id-based swallowed by their cars and homes instead of unlocking doors and keying alarm codes. Or not.

It could go the other way where people write ML algorithms that write poems instead of writing poems. People might go to art museums and plug their mind into the Salvador Dali machine and their friends might laugh at their Dali-ized creative thoughts seen on a 4 dimensional canvas.

In today's SE/SO reality, the best we can do is to consider naming and defining sites and tags based on a balance between currently common use of terms, the literal meaning of the words that comprise the term, and the overarching pattern of academics, publication, and terminology in those two places and on the web.

My gut feel is that excessive control will do the exact opposite of balance and push everyone with a brain away from the entire SE/SO engagement model and other sites with more incentive and less control will capture those emigrants.

| |
$\endgroup$

You must log in to answer this question.

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