Wikibooks:Artificial Intelligence: Difference between revisions

This page contains a draft proposal for a Wikibooks policy or guideline. Discuss changes to this draft at the discussion page. Through consensus, this draft could become an official Wikibooks policy or guideline.

The following draft policy outlines the Wikibooks community's perspective on the use of artificial intelligence-generated content on this site.

Large language models (LLMs), often referred to as "AI chatbots" in certain contexts, can be beneficial. However, like human-generated text, machine-generated text can also contain errors, flaws, or even be entirely useless. In particular, requesting a language model to write a book or an essay can sometimes cause the production of complete fabrications, including fictitious references. The output may be biased, may libel living people, or may infringe copyrights. As such, LLMs may not be used to generate original material at Wikibooks, and their sources should not be blindly trusted.

LLMs may be used to assist with editing, adhering to the following guidelines:

1. You may use LLMs as writing advisors, such as asking for outlines, asking how to improve a paragraph, asking for criticism for the text, etc. However, you should be aware that the information provided by the LLM can be unreliable or entirely incorrect. Exercise due diligence and common sense when deciding whether to incorporate the LLM's suggestions.
2. Use due diligence when crafting prompts for LLMs. Prompts designed for Wikibooks should use natural sentences, be as descriptive as possible, and include keywords such as "encyclopedic", "keep the meaning intact", etc., to minimize the AI's addition of original content.
3. You are responsible for ensuring that using an LLM will not disrupt Wikibooks. Therefore, you must document your use of an LLM and its purpose in the edit summary or on the talk page.
4. LLMs should not be used for tasks with which the editor does not have substantial familiarity, and their outputs should be rigorously scrutinized for both quality and compliance with all applicable policies. Editors who are not fully aware of the risks associated with LLMs and who are not able to overcome the limitations of these tools should not use them to write books.

AI tools can often generate reasonably accurate translations from languages that have large bodies of written work. Translations from another language into English Wikibooks can be easily vetted by someone who is literate in English; however, if you do not know the original language, be aware that the output can be grammatically correct but still a flawed or even inaccurate translation from the original. Similarly, if you are translating from English but are not literate in the target language, you cannot assume that the output is a valid or reliable translation. Consequently, translated works from these LLMs should not be incorporated into Wikibooks without a literate native of the original text who can assess the quality of the translation into English. It is also best practice to assume that translations from the English Wikibooks should not be copied and pasted wholesale into other language editions without understanding local policies and having local editors review the text before publication.

Most AI tools have the capacity to create media, particularly images, from prompts. If you are interested in uploading this media, please be aware of relevant policies on licensing at our sister project Wikimedia Commons or our local policy on images, depending on whether you are uploading it here or there.

All content generated using LLMs must be explicitly marked as such in both the edit summary and the page's discussion page. The following information must be provided:

• The date of generation/addition
• The tool used (e.g. Google translate, Bard, ChatGPT, Midjourney)
• The prompt(s) fed into the tool (e.g. "suggest how to improve this text")

This applies to every instance of using AI content. If you create new prompts and incorporate them into a page multiple times, each instance must be documented, including on the talk page. Any user incorporating artificial intelligence content who fails to include this explicit notice is subject to warnings and subsequent blocking.

The following templates may be used:

Template	Output
{{Ai-generated}}	This page incorporates text and/or recommendations from a large language model like GPT. Generated information may be false, and references may be invalid. The page should be thoroughly reviewed by a human to remove any incorrect information or copyright violations.