Wikipedia: ChatGPTorNot? This is a question

Wikipedia excels in its clear knowledge structure, ChatGPT excels in specific issues, and both sides may not necessarily be irreconcilable.

Author | Moonshot

edit|Jingyu

22 years ago, in early 2001, Wikipedia emerged and became an important platform for people to acquire knowledge in the subsequent Internet era.

editwikiality

Later, with a survey by Nature magazine, it was found that the accuracy of Wikipedia was close to that of Encyclopedia Britannica. Google began placing Wikipedia at the top of search results, and the Wikipedia community and content contributors continued to grow. It took many years for Wikipedia to finally gain public trust.

At the beginning of its birth, it was questioned, and more and more people participated in improving it. Later, it received public recognition and became a daily tool. This process, which was only half a year old, ChatGPT is experiencing, and not only has it become a challenger to Wikipedia.

Not long ago, the Wikimedia Foundation held a conference call for the 2023-2024 annual plan, which mentioned 35 AI sessions and discussed the challenges brought by ChatGPT.

But the challenge that Wikipedia is worried about is not being replaced by ChatGPT. But a deeper test: Will the future of Wikipedia be written by ChatGPT?

01

Where does the content of Wikipedia come from?

To know if ChatGPT can write Wikipedia, one must first know where the current content of Wikipedia comes from.

edit

As of the beginning of 2021, the number of Wikipedia entries in all languages reached 55 million. How to ensure the accuracy of content depends entirely on the screening of Wikipedia community volunteers.

ChatGPT AI edit

2016 Aaron Halfaker edit2020MIT AI edit

Wikipedia About ChatGPT Page | Wikipedia

And as stated by the Wikipedia community,Artificial intelligence is very good at summarizing a long technical article into a version that children can understand, and it works well for AI to generate a children's version of Wikipedia.

Translating, checking, summarizing, and simplifying existing content, Wikipedia's application of AI has been limited to this until the emergence of the large language model ChatGPT.

At present, ChatGPT, which mainly interacts through text, can not only answer users' direct questions, but also be used for very complex language tasks, including automatic text generation, automatic question answering, automatic summarization, and so on.

ChatGPT can write articles similar to real people and provide detailed and clear answers in many areas of knowledge. Even if the factual accuracy of the content generated by ChatGPT still requires manual secondary review, the problem faced by Wikipedia is already clear: can volunteers use ChatGPT to write Wikipedia entries?

02

Can we talk about whether we want to or not

Richard Knipel, a veteran Wikipedia member of the Wikimedia chapter in New York City, used ChatGPT to draft a new entry on Wikipedia called "Art Title." Knipel stated that the version provided by ChatGPT was general but syntactically correct, defining the concept of art title and providing examples from ancient times to the present. He only made minor modifications based on the draft.

edit Knipel ChatGPT edit

ChatGPT edit AI ChatGPT

ChatGPT editWikipedia

Version generated after extensive manual intervention | Wikipedia

But another faction also refutes this claim, as Knipel believes that modifying and enriching incomplete information is the way Wikipedia has always operated in practice. ChatGPT will continue to exist and develop rapidly, why not use it while emphasizing the necessity of manual intervention? It's a bit biased to think of ChatGPT as a raging beast.

But before we think about it, let's see if it's possible. Is ChatGPT sufficient to directly write Wikipedia?

On March 30th, Wikipedia founder Jimmy Wales discussed this issue in an interview with EveningStandard. Wales believes that,There is still some distance left for ChatGPT to independently write a complete Wiki entry, but it is difficult to say how far it is.The industry calls this situation hallucinating - I call it fabricating

ChatGPT has a tendency to fabricate things out of thin air, which is really not good for Wikipedia, "Wales said in an interview. In practice, the same applies. If you search for a word on Wikipedia, it may provide feedback that the item does not exist. However, if you ask ChatGPT, it may generate an unfounded fake message for you.

ChatGPT's ability to 'lie' is no longer new. But ChatGPT has only been around for half a year, and its self iteration ability is already amazing. It seems like it's just a matter of time before ChatGPT "speaks the truth". So what is Wikipedia worried about now?

03

Limited manpower,

Infinite computing power

The Wikipedia team is not so concerned about whether the content originates from humans or AI, but rather about whether the content quality is satisfactory.

In the conference call summary report of the Wikimedia Foundation, "Challenges" were placed at the beginning, and the most significant part of it, which is also the biggest concern of the Wikipedia team, is:Wikipedia flooded with a large amount of AI generated content, drowning out truly high-quality and accurate information.

Wiki projects have a large amount of high-quality, reliable, structured, and categorized content. This is the value we bring to the world.What scares me the most is not that people use big language models like GPT to acquire knowledge, but that the AI generated content that needs to be patrolled will explode.

For high-quality content, the time cost of creation is much higher than that of consumption. Just like a relatively complete Wikipedia article, it requires many people to participate in writing, takes a lot of time, goes through many processes, and is completed in a few minutes for readers.

edit AI ChatGPT But it does not directly provide the source of the argument for the conclusion, and manual labor still needs to find the argument again. In the end, revisions may take longer than writing.

There will be many extended reading links under Wikipedia's entries | Wikipedia

Currently, Wikipedia volunteers have identified many issues with ChatGPT's automatic generation of content. For example, ChatGPT can easily be defined too broadly, resulting in unclear meaning. Also, ChatGPT's wording and sentence making are too positive and do not match the objective and neutral writing style that Wikipedia wants to present.

The most important thing is that information sources are difficult to search for. The credibility and expanded readability of Wikipedia are largely based on the rich information reference sources under the entries, but ChatGPT will not actively provide references, and may even fabricate them out of thin air.

20edit Andrew Lih ChatGPT ChatGPT But after carefully reviewing, Lih found that these source articles do not exist, and even the URLs provided by ChatGPT are automatically generated fake links that cannot be found on the page.

In summary, the Wikipedia team directly stated that the speed and efficiency of AI generated content may exceed the project's operational capabilities.

In addition, there are many concerns that the Wikipedia team may have. For example, among Wikipedia contributors today, white men who use English are still the main body, and Wikipedia content is already biased towards language and content. ChatGPT's AI machine, which feeds on internet information, generates content that further amplifies bias.

Wikipedia co founder Jimmy Wales talked about AI's involvement in writing Wikipedia in an interview with Standard | Standard.co

The Wikipedia team is also unable to grasp the tendency of volunteers to use AI tools. Lih believes that Wikipedia people do not lack motivation, what they lack is time. The poor draft generated by ChatGPT can stimulate Wikipedia volunteers' desire for revision. This also conforms to the "Canning Stability Law" proposed by Ward Cunningham, the father of Wikipedia:The best way to get excellent answers on the internet is not to ask questions, but to post a wrong answer.

The Wikipedia team is also concerned that when Wikipedia is filled with AI generated content, users will reduce their trust in other information and instead trust media content with a more "human author" logo, such as appearing videos or media newspapers that label authors.

04

Can't we coexist

Wikipedia and ChatGPT have many similarities, such as being primarily textual and attempting to 'answer everything'. But the most significant difference between the two lies in the different ways of answering.

Wikipedia is an encyclopedia style information with a framework, system, and detailed index. When you click on an item, you can learn about its historical changes from the simplest summary introduction. Through the rich extension links in the item, you can delve deeper vertically or jump between different items horizontally, expanding your understanding of the entire field.

ChatGPT currently presents a questioning style interaction, requiring users to have a clear understanding of the questions they want to know. Expanding outward also requires building on the answers provided by ChatGPT and further questioning.

Different users have different tendencies to obtain information and choose different tools,Wikipedia cannot answer very specific questions like ChatGPT, and ChatGPT does not have as precise and well organized knowledge information as Wikipedia.The way these two are used is like whether we choose to read textbooks or directly ask the professor questions.

Just like Wikipedia was not wiped out by the rise of the search engine Google, it will instead appear in the first and sidebars of Google's search results.

The Wikipedia team does not have the same sense of crisis as Google. At the annual meeting, although the challenge brought by ChatGPT was highlighted at the beginning, more time was left for "opportunities" throughout the meeting.

What happens when there is a large amount of AI generated content on the internet?In an internet composed of trillions of low-quality and low credibility pages, both the people and users who create large models need to find reliable information, and they may use Wikipedia more.This is the "opportunity" in the eyes of WikiLeaks.

Using a large language model to investigate bugs, translate, summarize content, and enrich media formats, such as video generation reflected in GPT-4, AI generated images can also be placed in many abstract concept entries, increasing readability, and enabling mutual conversion between text and speech.

But the above premise is that the big language model cannot be used as an auxiliary tool, and it cannot dominate. Wikipedia is about humans gathering together to try to define the truth. These tools are unreliable and can distract us from practical tasks. We should be careful how quickly we catch up with this trend, rather than giving it up. We should focus on the people who create knowledge

The attitude of the Wikipedia team also reflects our current cautious attitude towards AI.Not being replaced, wanting to fully utilize AI while not trusting it enough, wanting to maximize its value, but not daring to easily deliver truly valuable content, making it difficult to let go of the core concept of human centeredness, approach cautiously, and catch up carefully.

How Wikipedia coexists with big language models may answer the question of how humans coexist with AI.

This article is an original article by Geek Park. For reprinting, please contact Geek Jun's WeChat geekparkGO

Geek's question

How does Wikipedia coexist with big language models?

Hot Video

Live display! Let's see how Google uses AI "magic" to optimize photos! Even parts that are not in the original image can be redrawn?

Like FollowGeek Park Video ID