Phone showing Chat GPT answering questions

Does Chat GPT Plagiarize? How the AI Creates Content

How ChatGPT works, and what that means for plagiarism.

Silent PC Review is reader-supported. When you buy through links on our site, we may earn an affiliate commission. Learn More

Table of Contents

ChatGPT is the household name when it comes to language AIs: used by many as a tool where it’s fed a prompt, and it responds in seconds with lengthy text. But since its very introduction, the question has been posed: does ChatGPT plagiarize? The language model must get its information from somewhere, and that somewhere is a particularly shaky source.

As the debate rages on, we’re here to highlight the various critical perspectives on whether ChatGPT plagiarizes content to create its responses – and where it may plagiarize from. The AI debate is a pertinent one, as schools that detect ChatGPT use argue that the language model is both unoriginal and uninsightful. So understanding how the language model works is key to understanding whether it uses the work of others.

✓ TL;DR

Quick Answer

No, ChatGPT doesn’t technically plagiarize, but the grounds are shaky – and arguably the result is worse. Being fed a slurry of data from across the internet means that ChatGPT’s responses are sourced from any number of places, without validation or a clear means of understanding how it came to that response.

How ChatGPT Works

ChatGPT is what’s known as a Large Language Model (LLM), which means that it is an AI able to simulate concise written English and string words together in responses. It ingests huge quantities of textual data, which is then assimilated into its word banks and can be used to respond to indirectly related prompts. It is fearsomely impressive technology, regardless of your stance on the AI debate.

Where that data is sourced from, however, is slightly more ambiguous. OpenAI uses a web-scraping process to harvest data from across the internet, effectively creating a slurry of data. Though this data is periodically updated, these updates are few and far between. The last update to the free version of ChatGPT was January 2022, which means that a majority of its data will likely now be outmoded.

Additionally, ChatGPT is prone to simply proclaiming misinformation, and makes mistakes. If you ask it to solve the same math problem twice, you’ll likely get two different answers.

Does ChatGPT Plagiarize?

OpenAI states that ChatGPT doesn’t plagiarize – at least, not intentionally. Because the results of ChatGPT are an indirect result of the prompts fed into it, that means that results are based on the works of others. This is why many academic bodies are hesitant about AI use. Whilst ChatGPT make for great accessibility tools to make menial tasks easier – such as checking grammar or even transcribing notes with the new image input technology – the waters are muddied when it comes to more intensive work. Rather than drawing on one source for a response, ChatGPT draws on dozens, if not hundreds, to formulate its answer. But with no means of verifying where the information was originally sourced, there’s no real means of checking whether or not the data was plagiarized – or from whom.

There are some interesting safeguards in place, however. If asked on certain specific topics, ChatGPT will cite the source where possible. Particularly for open-source formats such as Wikipedia or YouTube, the language AI will fairly reliably provide a source video or text. However, the responses are unverified and based on the same training data as the language model, meaning that the responses from ChatGPT are still somewhat rote.

In Summary

The waters around whether ChatGPT plagiarizes is murky, and it’s incredibly difficult to verify in either direction. The ethics surrounding generative AI use, particularly in academia, are fairly clear-cut: plagiarism is strictly prohibited. However, ChatGPT’s use of the internet to come up with responses creates an interesting Ship of Theseus debate: how many aspects of the writing have to change before the writing is no longer considered plagiarized? Whilst ChatGPT is not explicitly ripping off the works of others, it would be nigh-impossible to tell if it did.

For more on the limits of ChatGPT, including how accurate the LLM is, stay tuned with us here at Silent PC Review.

Leave a Comment

Your email address will not be published. Required fields are marked *

About the Author