The old and new of literary plagiarism

T.S. Eliot is misquoted as saying “good writers borrow, great writers steal”. What he actually wrote was: “immature poets imitate; mature poets steal; bad poets deface what they take, and good poets make it into something better, or at least something different.”

The thin line between plagiarism and ‘drawing inspiration’ has always been difficult to parse. People are inspired to write by different things every day – people they know, books they’ve read, films they may not remember watching. A loose set of criteria are typically used to determine whether something is an inspired original or a direct ripoff:

1. The extent of the similarity (in substance or essence);
2. The extent of the difference (has the author done something transformative with the work themselves or just paraphrased it); and
3. Whether the source material is acknowledged.

The use of AI heightens the existing risk of plagiarism and introduces new ones. The training data for AI features books, blogs, and just about everything else, stolen at scale. People incorrectly assume that LLM outputs are unique only to find that the “edits” a program provided to their text have inserted copied passages of someone else’s work. The risk is heightened given the inability of LLMs to provide meaningful citations for outputs.

With human writing there is an expectation that it is imbued with the specific style and perspective of an individual that has been impacted by, and learnt from, other people and other works. These can be enumerated in the footnotes or the acknowledgements section for anyone interested in learning more. Written works created by LLMs are ‘inspired’ by everything and can therefore properly credit none of it. LLM responses may appear to ‘cite’ sources, but given responses are predictive there is often no connection between the text and the source cited. The citation isn’t there to credit a specific source with the information preceding it, it’s there because the algorithm determined that it should be. If this is too abstract, consider the times you thought you found a source supporting your argument in a panicked Google Search the day before an assignment was due, only to click through and discover it refuted your thesis statement. LLMs are a bit like that, with an added gamble that the text they appear next to may not appear in them, or the source itself may not exist at all.  

Jonah Lehrer is a kind of one-man, proto LLM. Lehrer built a career off invented quotes and recycling sentences without attribution. The latter was called out first and deemed ‘self-plagiarism’ in early 2012. Later that year, a reporter claimed several quotes attributed to Bob Dylan in Lehrer’s book, ironically titled Imagine: How Creativity Works, seemed too trite (even for Dylan). Only when the reporter publicised the claims in an article did Lehrer admit:

1. “The quotes in question either did not exist, were unintentional misquotations, or represented improper combinations of previously existing quotes.”
2. When originally questioned by a reporter about the quotes, he had lied again about their source – claiming they were from archival footage supplied directly to him and that is why they didn’t appear elsewhere. “This was a lie spoken in a moment of panic. When Mr Moynihan followed up, I continued to lie, and say things I should not have said.”

(sound familiar, Claude user? This could be is the source material for those bonkers LLM apologies). Following this admission, Lehrer resigned from the New Yorker, was terminated from Wired.com, and several of his books were pulled from publication.

There have been multiple scandals where memoirs, purportedly non-fiction, have been found to be more fiction than fact. A Million Little Pieces by James Frey was a bestseller promoted by Oprah before reporters discovered it contained Many Large Lies. This shouldn’t have been surprising, given it was originally pitched as a novel. Only after it was repeatedly rejected by publishers did Frey start claiming it was a memoir and ultimately secure a book deal. In a move that feels inconceivable in the current publishing landscape, following the settlement of a dispute between Frey and his publisher, readers were offered a refund if they felt defrauded into purchasing the book by the claim it was a true story.