AI Attribution
AI attribution is the process of identifying and meaningfully labeling content that has been generated in whole or in part by an artificial intelligence (AI) system. This can include things like news articles, social media posts, and research papers, as well as many other formats of both online and offline content..
The goal of AI attribution is to make it clear to readers or viewers whether the content (in its entirety, or elements of it) was created by an automated tool and not a human, as well as to give other meaningful data about the specific provenance of the article. End users can then make their own informed decisions about the content they consume. (For example, some users might choose to disallow all AI-generated content altogether, or only allow content from approved AI information providers.)
There are at least three levels on which AI attribution might occur in online publishing systems such as blogs or social media.
- The profile level: the social media account or blog identifies itself as as being a publisher of AI-generated, or AI-assisted content (i.e., hybrid human-AI content). This self-identification would also carry through in ideal circumstances to any byline on articles published by the account.
- The post or article level: the content, whether a blog or social media post, or other type of online published article informs the viewer at a high level that AI-generated or AI-assisted elements are present. This might occur in different ways depending on the product or media context, including in the byline, as a subtitle, some kind of tag or other prominently displayed visual element (clearly-defined badge or icon), etc.
- The inline granular level: the article’s contents themselves are marked up to indicate which parts were input by a human, and which were generated by an AI tool. We explored the experimental method called “AIMark” to apply custom markup or markdown to hybrid AI-assisted texts in more detail here.
This is a big topic, which we will continue to explore in subsequent posts.
Tim B.
one point to keep in mind is that enabling content producers to self-identify AI-assisted content in meaningful ways could potentially reduce the burden on third parties to scan content, and make possibly incorrect determinations about the presence of AI-generated content.
Tim B.
ChatGPT’s eli5 summary:
“AI attribution is like putting a label on something to say that it was made by a computer and not a person. It helps people understand that the thing they’re looking at was made by a machine, and not a person. This is important because it helps people know if the thing is real or not, and also helps protect the rights of the people who made it. It’s like when you see a label on a toy that says “Made in China”, it helps you know where the toy came from and who made it.”
A bit imperfect, but getting there…