What A.I. Writing Can (and Can’t) Do

What A.I. Writing Software can do:

• Produce strings of language that look & sound like they could have been written by a human being (albeit a very utilitarian & mechanistic-sounding human, mostly).
• Write in the voice/tone/conventions of relatively stable & formulaic genres with easily identifiable conventions (five-paragraph essays, simple rhyming poems, lab reports, songs by well-known bands, obituaries, wedding speeches, etc.).
• Summarize and provide basic analyses of texts that have been extensively written about by human beings.

 

What A.I. Writing Software can’t do:

• Invent truly new, truly “original” language (in the sense that we typically think about “originality”). Every word an A.I. software “writes” is pulled from a variety of other texts written by human beings & remixed to create something that feels original (because, again, its training data / source material was originally written by human beings).
• Reliably cite sources. More often than not, they will fabricate sources, which could put students in the awkward position of having to “claim” writing with made-up information as their own, or confess that an A.I. made up the source (both of which are academic integrity violations).
• Analyze or produce new figurative language. Unless particular metaphors (and explanations of them) have been programmed into its system, an A.I. cannot tell you that, for example, that “life is a highway” (nor why you may want to drive it all night long), or why you might call a human being an “early bird” who rushes to work in order to “get the worm.” Its output operates primarily at the level literal interpretations of words and phrases and cannot reliably invent novel figurative language.
• Summarize or analyze texts that have not been processed / written about in their corpora of training data. There is a better chance that the A.I. can write a relatively decent essay about the “themes” (generically defined) in Shakespeare’s A Midsummer Night’s Dream than a rhetorical analysis of a recently-published op-ed in the New York Times.
• Substantively reflect on “choices” it has made about why it wrote the way it did (at least, not in natural language / concepts we use to refer to human cognition & decision-making).