The plagiarist sins twice, by stealing and lying.
To plagiarize is to steal the creation of another, without credit or compensation, and then lie that the creation stems from one’s own talent. Stealing and lying are, of course, each condemnable in their own right. But an accusation of plagiarism asserts that these two have been committed in tandem.
The archetypical plagiarist is a student who submits an essay copied verbatim from an authoritative source, perhaps from a book on the topic. This student has stolen1 the words of the original author, and lied to the instructor about their writing ability. But plagiarism is not limited to written content. It is also plagiarism when a musician incorporates a guitar riff taken from another song, or when an artist adds an image created by someone else to his portfolio.
To contrast, we can consider examples of non-plagiarism.
An online clothing retailer, who purchases designs from artists and sells t-shirts printed with the designs, is not plagiarizing. This company is not stealing, not lying, and on the whole completely unproblematic.
If that retailer were to download an image from an online portfolio, and print the design onto a t-shirt without compensating the artist, that would be stealing. However, it would not be lying: the company is clear that designs come from various artists, and that is the case here as well. This would therefore not be plagiarism.
We can flip this. Consider the case where an individual artist, who claims to design his own t-shirts, pays another artist for an image, and then signs the bottom and prints it on a t-shirt. This was not stealing: the second artist was fully aware of the first artist’s intentions. The first artist is lying, because he has passed off the work of another as his own. However, he did not plagiarize.
Plagiarism is not always cut-and-dried. For example, there is only a thin line between making a reference to a popular scene, and plagiarizing the content of said scene. Discriminating between the two comes down to navigating assumptions: is the audience likely to be familiar enough with the source material to recognize the homage, without explicit credit being given? These judgement calls can be difficult to make. It is important that rules around plagiarism be strict and inexploitable, yet it seems wrong to tar someone as a plagiarist for overestimating their audience.
Perhaps plagiarism comes in degrees of varying severity. First-degree plagiarism, like first-degree murder, is intentional and premeditated. Second-degree plagiarism might be an homage gone awry, or a partial-but-inadequate attribution. This distinction adds the necessary nuance.
A natural question, then: what is the plagiaristic equivalent of manslaughter? Until recently, it would have seemed odd to consider that one might plagiarize negligently, since the act requires, at minimum, actively identifying a source to steal from. But AI systems like ChatGPT and Stable Diffusion open up an interesting new possibility. Although the vast majority of the content created by these systems is wholly original2, every so often they will generate an exact replica of a training datum. It is possible to avoid this by double-checking, but since this occurrence is so rare, some may not bother. This might lead to completely unintentional plagiarism, plagiarism by pure negligence:
Pageslaughter.
In the classroom, lying is a major concern: specifically, cheating. The goal of school is to give students skills and knowledge, and exams & assignments validate that these skills were, indeed, learned. Cheating on an assignment is lying about one’s abilities.
On a fifth-grade math homework assignment testing long division, it is cheating to use a calculator. But of course, it is not plagiarism: neither the calculator itself, nor its engineers or manufacturers, can claim to have ownership of the answer to a particular long-division problem. Similarly, it is cheating, but not plagiarism, to have another student complete the assignment, or to copy answers from an answer key.
Similarly, in fifth-grade English class, there are many ways to cheat without plagiarism. It is cheating to hire another student to complete an essay for you, for example; or to re-submit an essay you used for a different class. It is cheating to have your essay ghost-written by a parent, older sibling, or an AI writing tool such as ChatGPT. None of these are permitted. But none of these constitute plagiarism, because they do not involve stealing.
Although calculators enable certain types of cheating, they can also facilitate learning. For example, on a tenth-grade physics homework assignment, use of a calculator is typically permitted, even encouraged. The goal of the assignment is to evaluate the student’s ability to model ballistic motion, not do long-division. Offloading this sub-skill to the calculator focuses learning on the subject at hand.
From this perspective, the advent of AI-driven writing agents such as ChatGPT represents an exciting opportunity for educators. Perhaps in the future, assignments can offload eloquence to ChatGPT in the same way that they offload long-division to calculators. This would allow evaluations in these subjects to focus even more tightly on the advanced skills being learned. An example, it is easy to see how physics classes might benefit from this: lab assignments, wherein students run experiments and gain invaluable hands-on experience with the scientific method, are often compiled into long, tedious “lab reports” for submission. Much of the content of these reports is generic boilerplate, which can easily be automated away using ChatGPT without impacting the real objective: learning to conduct science.
I expect that at first, this mindset will engender some resistance. Teachers used to evaluating both content and eloquence may feel that it is their job to develop both, and see it as a dereliction of duty to offload eloquence to a tool. More cynically, teachers who have been simply evaluating eloquence as a proxy for good thinking (without giving thought to actual content at all), will find themselves left with no grading criteria at all. Like calculators in the 1970s, the incorporation of ChatGPT (and similar systems) into existing educational frameworks will be both gradual and controversial. And I am optimistic that, like calculators in the 1970s, the end result will be a net positive for learning.
Thanks for reading! Please subscribe or follow me on Twitter @jacobmbuckman to be notified of future posts.
If the book has been purchased by the student, this is still stealing, as the payment only gives the purchaser the right to read the words in the book, not re-use them without attribution.
At least, as original as content created by a human whose learning has drawn upon many of the same sources. This is admittedly still the subject of some debate.