1

You might’ve seen discussions in the news and around the web about AI Chatbots like Open AI’s ChatGPT recently. These tools use a variety of inputs to help answer questions and complete tasks in response to user prompts.

After a lot of thoughtful consideration, wikiHow has decided to ban the use of ChatGPT and other chatbot / AI tools in the creation of Google-facing article content on wikiHow.

As I’m sure you can imagine, there was a lot of careful evaluation of this topic while we worked to make a decision that’s best for wikiHow and our future. This likely won’t affect the way that you, as wikiHowians, write articles, but it’s important that we share this policy so we are all aware of it and can prepare for the future.

The reasons for this policy are multifold. First, we believe it is imperative that our product is the result of independent research and original prose from our amazing editors. We also believe that having a human-first approach to content will help differentiate us from the many companies that will adopt AI tools to replace editorial teams. We believe content from real writers and researchers provides a deeper value to our readers and is part of who we are as a company.

Additionally, Google has confirmed that they consider any auto-generated content as spam, and have been clear that they consider its use a violation of Google’s Webmaster Guidelines. We have already seen evidence of Google working to detect and demote AI-created content, and penalizing corresponding domains in search results. Manual penalties from Google often reduce traffic by over 90% for indefinite periods of time. As you can imagine, this would be catastrophic for wikiHow.

The team is currently working on integrating AI detection tools and is in conversation with OpenAI about additional tools to help us identify auto-generated content automatically. For now, if you’re aware of any existing article content on wikiHow that was created by AI or ChatGPT, or are concerned that a particular article may have been created this way, please email me on jayne@wikihow.com and we can look into it.

You may see some updated wording around acceptable content in our wikiHow guidelines coming up, and if this type of content becomes an issue for us as a community, we can look at further policy updates to handle it moving forward.

Feel free to leave a comment or let me know if you have any questions or concerns about this new policy. Thanks!

16 Likes
3

Adding to this, I recently used AI to create a story for fun. Even the best one I have found, NovelAI, has issues. They use nonsense and a story (or article) can switch topics quickly, or name no sense at all. I hope this helps to identify AI articles. (funny I’ve been really into AI generating anime things… lol)

3 Likes
4

@ArtsyBean lol, I see a lot of stuff on Twitter about “type (minecraft/roblox/some random word) and see what the predictive text tool says” and the replies are ridiculous:rofl:

2 Likes
5

I’ve come across only a few TikTok videos referencing some auto-generating software freely available over the web. Though I myself have never used them, I’m still looking to see how to figure out if a page COULD HAVE been created over them, because the tool I’ve been seeing seems to be one of the better ones out there. Just trying to learn what makes one can cause us lots of trouble, and hoping we aren’t involved too heavily at this time.

3 Likes
6

I don’t know much about these chat bots, but if say, a site makes an article with one of these bots, and then someone try’s to write an article here with one of the same bots and it has the same or similar title, wouldn’t that cause the bot to generate the same, or almost the same, content, and thus trigger the copyviocheckbot? I think that might be a good way that we could detect someone trying create an article with one of these chat bots. But I don’t know enough about them to tell if it would. Maybe they don’t always generate the same content when asked the same question.

4 Likes
7

I was listening to a podcast tonight from Offline by Crooked Media called Two Writers Wonder if They’ll Be Replaced by AI. The discussion is generally focused around ChatGPT.

During the discussion, the guest mentions a college student from Princeton has developed an app that can identify content generated through ChatGPT, particularly essays.

Here’s a link to the podcast for anyone who wants to listen: Spotify

Here’s a link to an article about the student’s ChatGPT detection app: College student claims app can detect essays written by chatbot ChatGPT | Artificial intelligence (AI) | The Guardian

It’s great to hear that wikiHow is taking a proactive stance regarding content practices. It will be very interesting to see how AI in general develops over time and becomes more sophisticated.

4 Likes
8

Can this maybe be added as its own dedicated policy page? I absolutely agree that GPT, DALL-E, ChatGPT, and other AI-generated content usually does not meet the editorial standards needed to have a wikiHow article. Even if they did, our goal is to present original information, so using an AI trained on unoriginal content is counterproductive.

5 Likes
10

lol me too!Also I noticed a thing “NFD” (Nominated for deletion) and I think maybe this is a reason to be added to that?

2 Likes
11

Amazing initiative, Jayne. As a cybersecurity writer, I have come across experts bringing forward how ChatGPT account details are found easily by threat actors without much effort.
However, to my surprise, I have also come across reports by Microsoft that they may integrate it for optimizing search results.
Either ways, security of users and authenticity of content is foremost!

7 Likes
13

It has been a few days and I want to workshop on how to word this policy. I don’t think it will need to go through Policy Proposals, but I think it would be helpful to send it through after this is all done, since I do have a few proposed additions. I have been made aware some university professors are using AI tools like GPT-3 and ChatGPT to help make engaging presentations.

I’ve started working on a policy proposal at wikiHow:Policy Proposals/AI Policy - wikiHow . I look forward to more input from community members and staff alike.

14

Thanks for your work on this, Aasim! I think it’s ok to have the policy page, but from the staff perspective we were just planning on the smaller updates that I’ve made to the Writer’s Guide and How to Write a New Article on wikiHow .

2 Likes
17

That’s very true. (I’ve also used similar apps/sites like DreamAI and AI Dungeon for fun).Of course, I’m pretty sure Novel AI in particular was designed to be used as a tool for authors to generate a starting point for their ideas, after which, they could take the output and add to or remove from it to fit their needs.

I’m curious as to what wikiHow’s stance will be towards articles that were initially written by an AI, but then manually edited and overhauled by a human before publishing.

1 Like
18

Just a heads up: As Google seems to be bowing out of utilizing AI content, Bing has entered the ring and added AI functionality to their search engine for some users (I’m part of the beta group). It doesn’t construct paragraphs and long essays, but have it build a personalized answer - and you’ll have AI power your search query.

(I tried once or twice this afternoon to see what it would do, and upon a search for constructing a paragraph about a person throwing a baseball bat high into the air (a most-bizarre thing), it came up with about 5 places to search for information by which one of our articles was its starting point for obtaining information.)

6 Likes
19

Google is also working on its own chat like AI for Google search. I think the move away from using AI generated content is more towards allowing AI generated answers to be used as a starting point for visiting other links by humans, rather than having complete AI entropy.

(side note: I learned from TikTok that ai is love in Japanese. I guess we will have to love AI:rofl:)

3 Likes
21

Regarding this–if we were to find an article that was written by AI, would it make sense to add a new reason to the deletion policy (e.g., nfd|ai) should this become a bigger problem in the future?

22

Just curious — how would you be certain it was written using an AI tool?

23

There are quite a few AI detectors out there (I think OpenAI even made one themselves to check for plagiarized content). You can run text through one of the detectors.

1 Like
24

Because even AI detectors can’t detect with 100% certainty whether content was written using another AI tool, I think we may need to put more thought into this. It could result in erroneously deleting legitimate content.

Curious if @JayneG can shed any light on how staff intends to verify AI-generated content isn’t promoted.

1 Like
26

Yeah, I think the tough thing is that everything is changing so quickly in this space right now. Perhaps long term we can look at adding a new reason to the deletion policy, but I think for now the goal is still just to monitor and see if we need further action.

Currently, we are using external AI detectors. If anything comes up from community-written articles, I reach out to check. I’ve found about 99% accuracy in the detection from the replies I get from authors:wink:If you’re a booster and have your suspicions, you’re welcome to let me know, or pop it in an external AI detector like https://openai-openai-detector.hf.space/ .

The note I leave on an author’s page is simple and straight to the point: Hi there, I just wanted to let you know that we don’t allow AI-written content on wikiHow, so we will likely need to delete the article you started. Please let me know if you think there’s been a mistake.

5 Likes