What niche does this AI proposal aim to fill?

Question

Proposal: Artificial Intelligence

"AI" questions are already on-topic on several sites. Engineering questions about tools like TensorFlow, Torch, etc. are on topic on Stack Overflow. In fact Stack Overflow is the site that the creators of many of those tools direct people to ask their questions.

There is Cross Validated which covers Statistical Machine Learning. Anything related to statistics is covered there. There is Computer Science which includes all scientific subfields of Computer Science related to "AI". There is now Data Sciences which covers applications of machine learning to data analysis. Computational Linguistic and Natural Language Processing are also covered by Linguistics. Philosophical questions are covered by Philosophy.

The recent increased interest is focused completely on Statistical Machine Learning (which includes topics like Neural Networks, Graphical Models, Deep Learning, Reinforcement Learning, ...) and its applications, which are already covered by these existing sites.

What is not covered already that would justify a new AI proposal?

There has been two AI proposals that went to beta and then failed. I supported one. There is a reason that they fail. Let me direct people to Robert answer from last year. These AI proposals don't come with the backing of a group of "AI" experts. The experts in "AI" are happy using the existing sites like Stack Overflow and Cross Validated. These AI proposals come from people who are not experts in "AI" but enthusiasts who might have taken a machine learning course or read a popsci or scifi book related to "AI", they often lack familiarity with the fields related to "AI" and are unaware of how these fields are actually organized. Without the backing of a group of experts in "AI" these sites will continue to fail.

I think this proposal will definitely fail and is a waste of people's time unless it is backed by a significant number of experts and researchers in "AI". Therefore my opinion is that this proposal should not go ahead unless we see a good number of experts and researchers in "AI" have committed to it.

If you can convince at least 50 experts and researchers in these fields to commit then sure, let's give this another try. Otherwise don't waste time on a proposal that even if it gets to beta will be shut down pretty quickly.

PS: By experts and researchers we mean experts and researchers in Machine Learning, Computer Vision, Computational Linguistics, Knowledge Representation and Reasoning, Computational Learning Theory, etc. If you are a "Professional or Expert" or "Academic or Research-Level Student" in another field but not in these fields you should not commit to this proposal claiming to be "Professional or Expert" or "Academic or Research-Level Student".

PPS: Here are some of previous failed "AI" sites: AI, ML, ML. None survived even a month.

Answer 1

The whole point of the proposal is to build up a knowledge base on AI, and cover the field from its inception to current advances. Your question points out critical issues to address (cheers), but I see a few good reasons to keep going:

History

The vague field of AI does have a history on its own. A few of us know about it, and I believe that quite a few would like to know about it too. This is usually not discussed in the sites linked from the question. Just a few keywords (no particular order, just RAM): Expert systems, Logo, Situation Calculus, artificial mouse and mazes, a Brick World, Deduction/Induction, Case-Based Reasoning, the Chinese room, the Subsumption Architecture. Sure we have the amazing Wikipedia and its pointers, but there may be quite a few questions around related to this history and what it can teach us for the future. Some understand that NN are not the final answer to any ultimate question.

2016

Magazines and distinguished newspapers around me pounder that this year is the "year of AI" (whatever it means; and by the way the year of IoT too). This context is a significant difference with the two previous attempts. The environment in which we are evolving is changing fast, and there may be ground now to have an AI Q&A somewhere. As K. Kelly wrote "the business plans of the next 10,000 startups are easy to forecast: Take X and add AI".

Point of view

A lot of exciting questions today pertain to NN, CNN, RNN, and co., and they can be addressed on some existing SE sites. But do these sites cover all aspects of these techniques, technologies, and perspectives? It's a hard bet, even if we can (rightly) argue it may be all that matters. However let's assume that someone is trying to understand RNNs, she would probably hit Wikipedia first, and then some online courses and blogs. But then she may have "naive" or "beginner's" questions that are not going deep into Cross Validated or Data Sciences.

Another example that has happened to me three times in less than a year: A young & dynamic person has some business idea; she's not technical and talks with engineers; she's adviced to use RNNs; she wants to grasp a bit what it is about. Again Wikipedia can give a hand, but when you tailor a technology to a business idea, you usually need answers to your questions.

This "point of view" aspect cannot be discarded easily, in the light of the so many SE sites that exist already: SO and Programmers, Android Enthusiasts and SO (Java?), etc. Several non-SO sites offer an interesting perspective on software.

There is no stupid question

If there are too few "experts" or "professionals", the site can fail, yes. Trial is what it takes to build up a community. If there is significant activity from "non-expert people", experts are likely to come and share. This is how SO was born, and that is how Quora seems to thrive these days. I got in touch with ex-colleagues, students, OpenAI, experts who hardly speak English. They do express interest, but they can't contribute without trying. I am not saying they will move if there is a Q&A, though, the thing needs be attractive, meaningful---that's why SE. I agree this point is debatable, and its validity is speculative, based on the History and 2016 points above.

---

All this is a bet. I believe we have to do it to know what is going to happen. Is it a waste of time? Well, one assumption is that people spend their time freely. We have stated several times that two attempts have failed a few years ago. We can assume that people know and can choose for themselves.

One final note on where the AI proposals come from. I did not look at the profiles of the previous instigators to check whether they match the 'people who are not experts in "AI" but enthusiasts who might have taken a machine learning course or read a popsci or scifi book related to "AI" [...]'. If that can serve---I doubt it---I do not match that profile, by something like 20 years (a bit less).

To end the final note, I would like to thank Kaveh for elaborating on what is basically the frame of this AI proposal. Thanks also for reminding people to make thoughtful commitments. That is the most important message to me: If we think it is worth it, the rest just follows.

---

As for your important point on: "In fact Stack Overflow is the site that the creators of many of those tools direct people to ask their questions.", I am not sure about the implicit conclusions of the post. SO is by far the StackExchange site for anything related to software. The tools (TensorFlow, Torch, etc?) are software, and the creators want to hit the largest software community first. That's why they go to SO, and not Cross Validated or Data Sciences. Note that usual questions are not really about "AI", but about how to use the tools' APIs... After all, you can draw a Mandelbrot Set with TensorFlow.

Answer 2

You're right that we already have StackExchange websites that can handle all these existing questions. I'd even go further and state that those who are wondering about "Friendly AI" and the Singularity can also go to Worldbuilding, claiming that they are writing a science fiction story and already have a fictional "world" in mind that just need some fleshing out. If the What If? goes into beta, questions about "Friendly AI"/Singularity concepts might also appear over there.

However, you now have 6 to 8 StackExchange websites that an AI programmer has to visit to get the help he needs. One of the highest ranked example questions for this site is "How emotional intelligence could be implemented in AI?" Now, you can get that question answered right here on Stack Exchange! All you have to do is break it down into smaller questions, each appropriately scoped...

• Oh, you want a definition of Emotional Intelligence so you can figure out how to program it into an bot? Go to Philosophy.
• Okay, you now got that definition? Good, now you need to acquire a dataset and analyze it before you feed it over to a machine learning algo like Torch so that it can learn Emotional Intelligence...head over to Cross Validated.
• And now you have a problem with using Torch? Stack Overflow is your man.
• Wait, wait, wait...you don't actually need machine learning for AI? There's other forms of AI out there that you could use instead that are just as good at Emotional Intelligence, but is much faster and more efficient? Let's head on over to Computer Science then...

And so on and so forth. The AI programmer is stuck traveling from one site to another (though Google is helping out), asking questions on each individual site, to try and figure out how to accomplish his single goal. This AI programmer is a migratory beast, out to stitch together random answers on Stack Exchange into a coherent 'whole' that he can then use. Not only is this fate a rather big inconvenience for the programmer in question, but future people will probably have to retrace the same steps as our original AI programmer if they ever wanted to implement Emotional Intelligence in their bots as well (again, Google can help out here...but you can't use the existence of an search engine algorithm to justify keeping an AI community fragmented).

In addition, each of the 6 to 8 sites each have their own community, with their unique experts, ideas, and cultural traditions. The "time until answer" will vary, as well as the quality of the final answers, and their tolerance levels towards these migratory AI programmers and their various questions that kinda, sorta apply to the scope of the site.

Is it sensible for AI programmers to be dependent on 6 to 8 different sites, at any given time? Is it sensible for these 6 to 8 sites to pander to these AI programmers instead of answering other questions? Or is it better to instead consolidate AI-specific questions from these 6 to 8 different sites, onto a single site that is easy to browse and look up? Some overlap between the communities may be inevitable, but it's better than the current status quo.

---

As of the time of this answer (12:54 AM GMT, Saturday, May 7th), we have 15 self-proclaimed experts/academics. So we are 30% of the way towards your proposed target of 50 experts/academics.