As the parent said, plagiarism has very little to do with copyright or legalities. If a human essay farm assigns the copyright of what they've written to you--which is very common in a work for hire situation--it's still cheating in the eyes of pretty much any school.

ADDED: There is a degree of I know cheating when I see it. But, absent specific instructions to the contrary, spell checkers and grammar checkers/recommenders seem to fall pretty clearly into the assistive tools category whereas GPT-3 does not.

The author still had to write the prompt, edit it, etc. these are still skills that are (imho) more valuable than the writing itself. It is no more “assistive” than turning “waht” into “what”. In both cases the computer must infer what I mean and replace it with the appropriate string.

Legalities do matter here. The definition is “taking someone else’s work.” AI (currently) has no rights and no ownership. It is an “it” and not a “someone.” You are using a tool, an advanced tool that should be dealt with on a case by case basis. IOW, “don’t bring a graphing calculator.”

No, this does not enter into it at all. The universities set their assessment policies how they see reasonable, and will generally not allow auto-generated text. Just like it is not illegal to have a smartphone in an exam, but you can still be failed for it if it’s against the rules of the university.

Not to be tooo pedantic, but you agree spellcheck is against the rules? It’s “auto-generated text.”

But it appears we are on the same page. If it’s directly against the rules, that’s different (like bringing an advanced calculator to an elementary math exam). But it def isn’t plagiarism.

If you didn't create it, it isn't your creations. It's algorithmic output.

You can argue that it is no different than spell check and grammar check. It takes your inaccurately written text and turns it into accurate text. It doesn’t generate anything from thin air, so no. It isn’t “algorithmic output” unless we want to consider other tools as “algorithmic output” as well.

Spellcheck works (and often fails) at the level of individual words. Grammar checkers involve more extensive sets of relationships, typically at the level of a sentence, but are still limited to parts-of-speech agreement, relationships, tenses, and the like.

GPT-3 is constructing entire narratives from prompts. It operates at the level of multiple sentences and paragraphs, to the scope of an entire essay itself. Based on a provided prompt, it generates an essay.

By your same logic, an instructor assigning a class essay would be considered the author of those essays as that instructor had generated the prompt.

Moreover, as a matter of pedagogy, where a course either assigns group work (such that multiple students are contributing to an essay), or requires drafts and reviews with an instructor, TA, or other students as part of the overall essay-writing assignment, those become acknowledge parts of the process. In the case of the draft / review / revise cycle, it would seem that that is one way of putting a check on AI-generated essays, at least for now, as though generators can follow prompts, they're rather more limited in taking criticism well, and adapting to it.

> By your same logic, an instructor assigning a class essay would be considered the author of those essays as that instructor had generated the prompt.

Hol up, didn't you just describe academic research?! Sounds like we're on the right path here.