-54

Up voting and down voting that is free and anonymous might not improve the quality of stack exchange. I would argue that it encourages irresponsible and capricious behavior because of the breadth of the voting community. There are many endless and closed threads about this issue here.

If reasoning is provided for a downvote, direction for improvement is available. Giving a lot of down votes without reasoning to a newbie can be discouraging. You can even get banned from asking questions if you have too many questions that are ignored and simply not up voted.

I would like to see an AI trained to vote. Only the AI would be permitted to down vote. Humans would only be permitted to up vote. The most highly up voted questions would serve as a training set for a recurrent neural net.This is a supervised learning problem.

In a sense, such an AI would represent 'the best of the community', even if only a minority of the community actually voted. Downvotes submitted by the AI might even be accompanied by a comment from it illustrating why it gave the down vote. That would even be entertaining as well as educational.

  • 12
    An AI. Trained to vote. How? By humans? I suggest that would suffer from all the points you've made. It probably needs to be trained by other AIs. Or Aliens. Not humans. – Oded Nov 23 '15 at 21:06
  • 15
    Small wrinkle in your plan is that an AI would need to understand English. And even broken English. As well as code. And be able to determine good code from bad, and code that is inefficient and code that has security issues and I can go on and on... This is not a "typical supervised learning problem". You are vastly underestimating things. – Oded Nov 23 '15 at 21:09
  • 19
    Why on earth you need to use a word like terrorism to describe voting? How are your worthless internet points related to terrorism? Do you really think that is appropriate? – Patrick Hofman Nov 23 '15 at 21:15
  • 4
    I feel like this is a very interesting idea, but crippled by exaggeration, hyperbole, and outrageous wording. Oh and pretty much impossible. But still quite interesting :) – enderland Nov 23 '15 at 21:53
  • Firstly, you're exceedingly idealistic. Secondly, everything you say is wrong. Do you really think SE will implement this? Certainly not to fight "intellectual terrorism". – bjb568 Nov 23 '15 at 21:58
  • 3
    "does not improve the quality of stack exchange" to "intellectual terrorism" is a pretty huge jump. – Undo - Reinstate Monica Nov 23 '15 at 21:59
  • 7
    "I find this objectionable because no reasoning is provided." Then I'm sure you'll love trying to figure out how an RNN decided what to vote. – Jeremy Banks Nov 23 '15 at 21:59
  • 32
    unfounded assertion; passive-aggressive wording; strawman argument; hyperbole; hyberbole; passive-aggressive; Stalwin. Conclusion: downvote (confidence=100%). – Gilles 'SO- stop being evil' Nov 23 '15 at 22:02
  • 13
    If writing an AI is so simple, let's see your github repo. – Michael Hampton Nov 23 '15 at 22:03
  • 1
    Allow me to refer you to this blog post I wrote - "trying to make a machine associate human concepts with other human concepts, by showing it human texts. That's never going to be an easy thing to do." – ArtOfCode Nov 23 '15 at 22:45
  • Among reasons that others have listed, allow me to present another. How do you train this AI? Do you train it to be among the "angriest" users that Stack Overflow has? Or does it like everything? Or, do you train it somewhere in the middle? – Andy Nov 23 '15 at 22:52
  • 1
    ǝpoɔɟoʇɹɐ, your blog post as well as that of Andy are interesting and the github repos appreciated. @Andy I dont see the AI as an entity that can like AND dislike. I was suggesting a partnership - humans can only like (upvote), the AI can only dislike (downvote). The tricky part is getting the AI to know what is bad from only showing it what is good. – aquagremlin Nov 23 '15 at 23:10
  • 12
    If you have an AI capable of this, why would you only let it downvote? Why not upvote as well? And then why do you want humans to vote at all? Insurmountable technical objections aside (!), the reason for doing this is completely nonsensical. In fact, if we've got an AI this sophisticated, we probably don't need humans to even answer anymore. – jscs Nov 23 '15 at 23:23
  • 8
    Your primary mistake here is characterizing downvoting on a post as an inherently negative action. While it does have a negative numerical impact on the score of the post (and, by extension, the reputation of the user), it has a positive impact on the community as a whole. Voting distinguishes good posts from bad ones so that future readers know which ones are more likely to be useful to them, and that requires both upvoting good posts and downvoting bad ones. – murgatroid99 Nov 24 '15 at 1:50
  • 2
    I'll +1 this, but only if I can be the one to train it. – user1228 Nov 24 '15 at 15:10
33

So, do it. Train an AI to up- and down-vote. Train an AI to leave comments explaining its votes. Run it under your own account; there's an API.

Talk is cheap. There are countless people doing this by hand every day, with precious little in the way of automated assistance... Our past efforts to automatically identify good and bad posts have seen limited success. If you think you can do better, then step up and try - and good luck to you.

You might find these projects helpful in your efforts: https://github.com/Charcoal-SE

  • "Train an AI to leave comments explaining its votes." Why? Commenting votes is not required and the system actively discourages such comments ("Comments cannot contain that content"). – CodesInChaos Nov 26 '15 at 16:54
  • 1
    This was the feature the asker requested, @codes. – Shog9 Nov 26 '15 at 16:55
  • @Shog9 consider setting up a competition on Kaggle or something? It would be good faith on part of SO to encourage this kind of research, and should someone significantly crack this problem of course that would be of interest to you. – djechlin Nov 27 '16 at 4:00
  • We did just that a few years back, @djechlin. Didn't really go anywhere; this is, as I said, hard. – Shog9 Nov 27 '16 at 16:52
23

Up voting and down voting that is free and anonymous does not improve the quality of stack exchange.

Yes it does. It lets visitors see what the rest of the community think, and prevents wars between users.

I find this objectionable because no reasoning is provided.

It lets visitors see what the rest of the community think, and prevents wars between users.

This is especially vexing when you get a lot of down votes as a newbie and you don't have a clue.

Those newbies you're talking about don't look at the help pages before posting. All they do is post a question, generally containing broken code, and that's it. On almost all SE sites, questions need to contain the problem, and surrounding detail, and what you have tried.

You can even get banned from asking questions if you have too many questions that are ignored and simply not up voted

That's only if you've got a lot of deleted/closed/downvoted questions. And that can be prevented by reading the /help section of every site you post on.

I would like to see an AI trained to vote. Only the AI would be permitted to down vote. Humans would only be permitted to up vote. The most highly up voted questions would serve as a training set for a recurrent neural net. This is a typical supervised learning problem. Downvotes submitted by the AI might even be accompanied by a comment from it illustrating why it gave the down vote.

That, I must say, would be the most sophisticated AI in the entire world. Or even the Universe!

That would even be entertaining as well as educational.

The only thing I agree with you on... it would be entertaining...


Suggested Reading

  • Wow, already 10 downvotes in a half hour. Just for suggesting an AI. When our robot overlords read this they will hunt you down and downvote every one of your posts into oblivion. Seriously though, this is not a difficult machine learning problem. I disagree with Oded that an AI would need to understand English. And even broken English. As well as code. And be able to determine good code from bad, and code that is inefficient and code that has security issues and I can go on and on. Most neural nets have no understanding as we define it. Yet they are very successful at many tasks. – aquagremlin Nov 23 '15 at 21:49
  • 8
    @aquagremlin Why do you disagree? One needs to understand English in order to read, no? And one needs to read a post to understand it, and one needs to understand a post to vote on it. – HDE 226868 Nov 23 '15 at 21:53
  • @HDE 226868 No, my computer can do OCR very well -and then speak it back to me; is this not reading? Does a compiler not understand source code yet it parses into machine code and also point out human error? As for understanding in the human sense, is this nothing more than a mapping of sets of ideas to different sets of ideas previously abstracted from experiences through a sequence of 'hidden layers'? I am not simply describing a 'thesaurus AI' - but a complex 'graph'. The value of the AI that I described is imparted by the human upvotes. In that way it simply represents the best of us. – aquagremlin Nov 23 '15 at 22:02
  • 12
    @aquagremlin Reading? Yes. Understanding what those letters mean? No. A compiler can read loads of fancy code, but code doesn't make an answer. On most Stack Exchange posts, code is not the most important thing. So this hypothetical AI is useless most of the time. Going to your second point on understanding, I quote Michael Hampton: "If writing an AI is so simple, let's see your github repo." – HDE 226868 Nov 23 '15 at 22:09
5

Up voting and down voting that is free and anonymous might not improve the quality of Stack Exchange.

Anonymous = Yes. Quality Improvement = Nope Yes. Why? Because most of the time, downvotes are used to signify that the post is low-quality and should be improved on or be deleted by the poster. Questions that have multiple downvotes tend to be closed and/or deleted quickly by the community. Users also comment to tell the poster why there were downvotes.

If reasoning is provided for a downvote, direction for improvement is available. Giving a lot of down votes without reasoning to a newbie can be discouraging. You can even get banned from asking questions if you have too many questions that are ignored and simply not up voted.

Reasoning is already clear as per downvote, quality is not good, poor effort, etc. Users also comment why they downvote sometimes to help the newbies. There is an algorithm used to determine your ban, which is NOT based on the fact the question has low views or no up-votes. There is actually a badge for having no votes and little views on a question.

Only the AI would be permitted to down vote. Humans would only be permitted to up vote.

AI's are not humans and vice versa. We can not simply not be allow to downvote. The AI would literally have the perfect brain, which is impossible. Voting is in our opinion, and AI's have a lot of errors than we do. We can even retract our vote and change it when we see our mistakes!

In a sense, such an AI would represent 'the best of the community', even if only a minority of the community actually voted.

What if the community disagrees? The AI would be a dictator and our voting rights and leverage would be immediately removed. For example, if everyone upvoted this post and the AI downvoted, should anyone trust a crazy robot over thousands of people? No.

Downvotes submitted by the AI might even be accompanied by a comment from it illustrating why it gave the down vote. That would even be entertaining as well as educational.

We are encouraged to give comments why we downvoted in the first place. If the AI might comment, it is worse than nothing here. Entertaining, maybe. Educational? Nah, just do some studies there and there and we have some data.

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