Why bad writers can't use AI to write well

You have to be a pretty good writer to get good writing out of AI.

Sep 15, 2023

The biggest moral panic attending this year’s rise in AI is: Bad writers are using AI to cheat, plagiarize or write good articles (thereby replacing the labor of professional writers)!

I think this is misguided.

AI-generated writing is often horrible, tone-deaf and wrong. Generative AI usually demonstrates in ways large and small that it doesn’t understand the words.

A recent article published by a sports site called Race Track and re-published by Microsoft on MSN.com reported on the death of basketball player Brandon Hunter. Instead of saying he was “dead,” the headline said he was “useless.” The headline was: “Brandon Hunter useless at 42.” (The article remained online for a day and a half before removal.)

The whole story was off. Way off.

Instead of saying that Hunter had “passed away,” the piece reported that he was “handed away.”

The article said hunter is a former “NBA participant” who “performed” for other NBA teams and was “a extremely regarded highschool basketball participant.” His death was “introduced” by the coach. All these weird language flaws appeared in just the first two sentences.

This reveals an important fact about AI. It simulates “knowledge” about what words mean, but doesn’t understand the words or their context.

In other circumstances, “dead” and “useless” can be similar. To wit: “The battery in my flashlight is dead” and “The battery in my flashlight is useless.” Obviously in an obit “useless” is horribly wrong.

Likewise, “passed” and “handed” can mean the same thing: “She passed the bottle of ketchup to me” and “she handed the bottle of ketchup to me” are the same. “Passed away” and “handed away” are not.

Humans understand these differences. AI does not.

The Hunter obit is an extreme example. Most AI-generated content isn’t this bad. If the reader is skimming or indifferent to language, it’s often fine. Data-driven content like financial information or weather reports (where data, not language, is the sole purpose) can be AI generated without incident.

But for general news, social media posts, college essays, business emails and other written communication, AI-generated content can rarely pass for human work. AI is tone-deaf, oblivious to nuance and context and tends to come across as an English-as-a-Second-Language student who’s trying too hard.

But here’s the real reason nobody wants to read AI-generated writing.

Read the rest