On-topic and off-topic pages need to be clarified

Question

I think this website is good for:

1. Clarifying/explaining/discussing theoretical AI concepts (including concepts described in AI research papers, books, etc.), notation and terminology

2. Discussing philosophical issues related to AI (risks, safety, AGI, super-intelligence, etc)

3. Discussing the history (e.g. AI winters) and the future of AI and how it relates to other fields

Based on all topics described in the box on the right side of the AI Wikipedia page, theoretical AI concepts/goals comprise:

• Knowledge reasoning
• Machine learning
  • Reinforcement learning
  • Supervised learning
  • Unsupervised learning
  • Online learning
  • Continual, lifelong or incremental learning
  • Active learning
  • ...
• Planning
• Natural language processing
• Computer vision
• Robotics
• AGI

The approaches are

• Symbolic (GOFAI)
• Deep learning
• Bayesian networks
• Causal inference
• Evolutionary algorithms
  • genetic algorithms
• Swarm intelligence
  • Ant colony optimization algorithms
  • Artificial bee colony algorithm
  • Particle swarm optimization
• ...

Some philosophical and social issues include:

• Ethics
• Existential risk
• AI tests
• Definitions of AI
• Chinese room
• Weak vs strong AI
• Super-intelligence
• Friendly AI
• Emotional AI
• Explainable AI
• ...

Currently, the Help Center does not explicitly state that these topics are suited for this website, but I think it should. I think we should clarify which topics are on-topic here. In general, we can use the linked Wikipedia page to help us clarify which topics are suited for the website.

Furthermore, I would say that every implementation-related question should always be considered off-topic here, given that there's already Stack Overflow (and Data Science SE) for this. Which other topics are off-topic here? Should we also be more strict regarding primarily opinion-based questions? I think so, but given that philosophical questions are allowed here, we need to be careful when defining the borderline.

What about hardware questions related to neuromorphic chips? We actually have a neuromorphic-engineering tag. If they are about theoretical properties and not implementation issues, can they be considered on-topic?

Furthermore, it would be useful if every new user was "forced" to read this on and off-topic pages (before posting a new question), to avoid them to post off-topic questions. It would also be useful to have an automatic way to guide them to the more appropriate website, in those cases. Is this possible to do?

We should spend a few paragraphs to describe our community to new users and how it is different from (or similar to) other communities (in particular, Data Science SE, Cross Validated SE and Stack Overflow).

Several related questions have been asked in the past

• What topics can I ask about here?
• Is it time to modify our site guidelines?
• How to distinguish between 'programming' and 'conceptual' questions?
• Technical questions are not getting closed
• Why aren't implementation based questions welcome on this stack?
• What is in scope under the "implementation of machine learning" exclusion?
• What kind of implementation questions should be off-topic?
• What should the AI.SE Site Description be?

Answer 1

I agree strongly and think this is necessary, especially now that we're getting a good number of questions, and need to raise out stats on having multiple answers.

2 suggested additions:

• Clear guidance on when to ask on Data Science and Overflow.

• "Social impacts" as a separate sub-category

re: social impacts, I think it's distinct from philosophy, because it deals with tangible effects. It would also cover "mythology of AI" (singularity, robot takeover, etc.) and "AI journalism", which I think is an increasingly important topic--policing misleading reporting.

Answer 2

I mostly agree with the guidelines for on-topic topics proposed in the OP. The only possibly gray area for me is in this:

Furthermore, I would say that every implementation-related question should always be considered off-topic here, given that there's already Stack Overflow (and Data Science SE) for this.

Now in most cases I do agree implementation-only questions are better suited for StackOverflow (especially Tensorflow ones, since Stackoverflow is the official place for Tensorflow questions). However, a recent implementation question that I feel like may be better suited here was this one:

Expressing Arbitrary Reward Functions as Potential-Based Advice (PBA)

It technically is just an implementation / bugfixing question, which I'd usually feel like should be off-topic... but it is about a rather specific, non-trivial, relatively recent AI publication. I was personally already familiar with the paper (maybe because I spent a couple of years working in the same lab that these publications came from earlier, with some of the same people), but I think very few people on StackOverflow would be familiar with the paper or feel like reading it just to make sure they'd be able to answer the question correctly.

Would such implementation questions about very specific, relatively uncommon approaches still be considered on-topic here? I'm not talking about common stuff, like implementing "Neural Networks" or "an image classifier", plenty of people on SO know about that too.