12

I've been reviewing a spate of tag wiki additions in the Suggested Edits queue, and I noticed that a couple of authors I'm quite familiar with (Walter Jon Williams and H. Beam Piper) had descriptions that didn't match my recollection. (Which recollection I verified against reliable sources like the SF Encyclopedia.) Running those texts through a chat-bot detector gives a >90% likelihood that the text is machine-generated.

I know we've agreed we don't want chat-bot generated answers; what about chat-bot generated wiki text? Is this acceptable? I don't know what to do about this, since there's no way to flag it.

The following examples are all more than 90% likely to be bot-generated, but this is not an exhaustive list:

Note that this is problematic for the same reason that bot-generated answers are; for the wikis I have sufficient expertise to judge they have inaccuracies. (Emphasis added because it seems a couple of readers have somehow managed to miss the point. If 100% of chatbot descriptions for topics I know well enough to review for content contain errors, then they likely all do.)

8
  • Just a note: It is most likely an AI chatbot if the text sounds too perfect. Humans make mistakes in grammar and ponctuation.
    – Silvermidnight
    Mar 15 at 3:13
  • @Silvermidnight You mean like Agent Smith's flawless grammar?
    – Clockwork
    Mar 15 at 7:56
  • 1
    @Silvermidnight and in spelling
    – user13267
    Mar 15 at 8:02
  • Your emphasis just says AI generated text should be banned in tag wikis, just for the same reasons they are banned on answers. That doesn't justify anything. I'd say they should not be banned in answers either, for the same reasons I outline in my answer below. If you read the answer to your linked question or the highest voted answer here, even they agree that AI use to help generate answers are probably fine, just use of AI to post any random wrong information is what should be not allowed, exactly what I say in my answer.
    – user13267
    Mar 16 at 3:43
  • I have seen a lot of people throughout stackexchange jump on the 'ban AI' bandwagon, just because it seems to be the "general consensus". I wonder if majority of the bandwagon jumpers even know what these AI based text generators are, or what they can and cannot do. A lot of people just seem to have a mindset of "AI bad. need ban AI" and go around downvoting any post that vaguely suggests not banning AI. And it's always users arguing with other users too. For what? Protecting the stackexchange ecosystem? I am pretty sure once these AI chatbots become cheap and ubiquitous enough, .....
    – user13267
    Mar 16 at 3:43
  • ... stackexchange themselves will find a way to integrate it into their service, much like how Microsoft is trying to integrate chatGPT into their web-search. I have also tried generating answers for scifi stackexchange using an AI chatbot, and I have also found majority of them to be wrong. What ended up happening? They just didn't get posted. It just meant the chatbot was not good enough, or I hadn't understood how to give it the proper inputs yet. I didn't come to the conclusion that use of AI to generate answers itself was the bad thing to do.
    – user13267
    Mar 16 at 3:43
  • I think rather than claiming that any user who disagrees with your reasoning has somehow "managed to miss the point", a bit of retrospection is needed here. May be I myself came to the wrong conclusion?
    – user13267
    Mar 16 at 3:43
  • 1
    @Clockwork I mean that AI had difficulty distinguishing between, let's say, kid writing, 8th-grade writing, 10th-grade writing, and college term paper writing. Usually, AI writes somewhere between the 10th and college writing styles, sometimes with fancy words such as "hackmatack" that most people don't really use on a day-to-day basis. Of course, someone could fix the text after it is generated to make it sound worse + normal (also known as a lot of work).
    – Silvermidnight
    Mar 16 at 23:54

3 Answers 3

Reset to default
14

We should ban AI generated content in tag wikis

To be honest I think the reasoning given in the linked post about banning AI generated posts holds true for tag wikis as well. It's not quite as important in the sense that tag wikis are more hidden but they should be implicitly a more trust worthy source than someone's handwavy answer. The information within a tag wiki should be 100% accurate and trustworthy.

So, for the same reasons that the content is not accurate but looks like it is it should not be allowed in a tag wiki. Lets not also forget that copying without attribution from a source is also plagiarism and that is one of the default reasons to reject a tag wiki edit.

Again similar to the other answer, tag wikis which use an AI tool to assist in generating content should, for the time being, be fine. But not content that is 100% copy pasted from an AI.

Ona side note I've spent a lot of time creating and editing tag wikis and one of the things I've done in the past is clear out those that were just copy and pasted content from Wikipedia. We should not have copy pasted content and certainly so if the content isn't even correct because it's been created by an AI chatbot.

-4

We should not ban content just on the grounds that it was generated by AI.

As the general policy on any content here is "moderate based on content quality and nothing else" (eg, you're not supposed to upvote any random posts from a user as a reward just because you liked one of their posts, as is mentioned on the help page for serial voting, upvotes should be based on post not the user), I think the same standard should be upheld for posts based on AI. Even OP says he found out it was AI generated because he ran it through a detector tool, and in the end it didn't matter anyway because OP knew the posted information was false. This, I think, should ultimately be the standard we accept or reject posts on; quality of the content, not the (assumed) source. AI is just a tool, I am sure we will not be banning posts just because someone used a sentence completion tool or a grammar/spell checker or a thesaurus. I think we should consider this the same or at least a similar situation. These types of AI based text generators are going to become more common place whether we like it or not, its better to start learning how to deal with them rather than ignoring their existence.

If someone is consistently posting AI generated carp that is blatantly wrong then sure, delete the post or ban the user. I think this how we currently deal with spam anyway. But again the moderation should be based on the harmful actions, not because the poster used an AI or any other tool they like.

8
  • 4
    No, you misread. I ran it through a detector in the first place because it was inaccurate. That is why chatbot-generated text is a problem.
    – DavidW
    Mar 15 at 1:44
  • 4
    Well the problem is it was inaccurate, not that it was generated by an AI
    – user13267
    Mar 15 at 1:54
  • 7
    Again, no. The problem is that any chatbot text needs content review by someone knowledgeable enough to do it. (Or with the time to do the research.) Meanwhile someone can spam a dozen inaccurate wiki entries in minutes, leaving many times that much work for others to do in cleanup.
    – DavidW
    Mar 15 at 2:06
  • 3
    @DavidW Then like I said, deal with it like you would with any spammer, based on the content of the post and the vandalism, not because some user posted something that happened to be generated by an AI
    – user13267
    Mar 15 at 2:08
  • 1
    @DavidW "The problem is that any chatbot text needs content review by someone knowledgeable enough to do it." That's true of any content posted here, not necessarily just those generated by an AI
    – user13267
    Mar 15 at 2:10
  • 1
    This argument fails because it presumes that any random person can produce a garbage wiki capable of fooling anyone but an expert, and that they would bother. Neither of those are the case; it is only the chatbot tool that enables this spam. Further, you are completely ignoring the attribution problem, where people are posting work that is not theirs as their own.
    – DavidW
    Mar 15 at 3:52
  • 1
    @DavidW "it is only the chatbot tool that enables this spam" I disagree with this premise. People might have any number of reasons to post authentic looking wrong content (trolling, spamming, just bored, whatever). It's not anything specific to do with AI. "Further, you are completely ignoring the attribution problem, where people are posting work that is not theirs as their own" That's a completely different problem and nothing to do with AI either. I am sure a lot of people have, before the advent of AI chatbots, copied text from wikipedia/wherever and posted here as suggested tag wiki...
    – user13267
    Mar 15 at 3:56
  • 1
    ...and there are tools already in place to deal with all of the problems mentioned above. Yes may be AI makes this kind of spamming easier, but the "spamming" part is what is the problem and should be dealt with, not the fact that the tool for spamming in this case happened to be some popular AI text generator.
    – user13267
    Mar 15 at 3:57
-9

There are several arguments for allowing chatbot-based tag wiki descriptions in SciFi Stack Exchange:

  1. Efficiency: Chatbots can provide quick and efficient responses to user queries, which can save time for both users and moderators. Instead of waiting for a human to write a tag wiki, a chatbot can provide a preliminary description that can be edited and improved upon by human moderators as needed.
  2. Consistency: Chatbots can ensure consistency in the tag wiki descriptions across the site, avoiding discrepancies in formatting or information provided. This can help users find what they're looking for and understand the content of the site more easily.
  3. Accessibility: Chatbots can make tag wiki descriptions more accessible to users who may not be familiar with the site or the content, by providing clear and concise explanations that are easy to understand. This can also be helpful for non-native speakers of the language used on the site.
  4. Scalability: As the site grows and the number of tags increases, it may become increasingly difficult for human moderators to keep up with the demand for tag wiki descriptions. Chatbots can help fill this gap by providing initial descriptions that can be refined over time.
  5. Innovation: Chatbots can be programmed to use natural language processing techniques and machine learning algorithms to provide more personalized and accurate tag wiki descriptions based on user input and behavior. This can help enhance the user experience and make the site more engaging for users.
5
  • 1
    Generated by a popular chatbot for "Give arguments for allowing chatbot based tag wiki descriptions in scifi stackexchange"
    – user13267
    Mar 15 at 1:37
  • Have you tried asking it "Give arguments against allowing chatbox based tag wiki descriptions in scifi stackexchange"? I'm just curious.
    – Clockwork
    Mar 15 at 7:55
  • @Clockwork I haven't but someone in the stackoverflow version of this same question did and posted it in their answer. It just gives out generic advice on why not to use chatbots for things like these, in points, much like this answer
    – user13267
    Mar 15 at 8:01
  • 2, 3, 4 and 5 are all incorrect – and 1 is, too, depending on your metric for efficiency. This is a good demonstration.
    – wizzwizz4
    Mar 22 at 21:32
  • @wizzwizz4 feel free to take it up with chatgpt, as that is what was used to generate this
    – user13267
    Mar 23 at 0:43

You must log in to answer this question.

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