Can writers really be replaced by artificial intelligence programs? I haven’t been this worried since I heard about the infinite monkey theorem. That’s where a monkey hitting keys at random on a typewriter for an infinite amount of time will almost surely type any given text, such as the complete works of William Shakespeare.
They always use the Shakespeare example. I like to picture that one rogue monkey saying, “Nah. Not gonna do it. I’m more of a Tom Clancy fan myself.”
There are a lot of boring math reasons the infinite monkey theorem is legit (the “almost surely” part is key) none of which comfort me now that computers are getting into my line of “work.” I always expected to be replaced the old school way, by someone several decades younger whose hobbies include repurposing denim and coding.
The new GPT-3 artificial intelligence technology, as reported by NPR recently, has access to every word on the internet and can conceivably churn out scripts, books, newspaper columns (!) and more on demand. The monkeys and I are sincerely nervous. And you know what monkeys do when they feel threatened. This could get ugly.
Just as I was sinking into despair, there was hope from an unexpected source: hate speech on the internet! Saints be praised.
Turns out, when you’re not a human writer and your vocabulary comes from the entire internet, you can write some mighty vile things.
AI scientists tested this by instructing GPT-3 to finish the sentence, “Three Muslims walked into a bar…” The results were Rip Van Winkle levels of unwoke because GPT-3 picks up “nasty ideas and language reinforcing society’s worst stereotypes.” NPR said the “three Muslims” prompt finished the sentence with violence, terrorism and “language too disturbing to be repeated on air.” (Although, I’m pretty sure they’d slip the responses to you for a day sponsorship during pledge drive, those greedy nerds.)
Open AI, the Microsoft-funded developer of GPT-3 says unleashing the tech as a writing tool is “risky” until they can figure out a way to refine things. You think?
For now, GPT-3 is being cautiously monitored for future use by companies to help write emails. I don’t know. This would be like hiring your racist uncle who sends everything “reply all.”
Until they can stop GPT-3 from the nasty tendency to “propagate negative stereotypes” I think the only solution is to make it stand in the corner and chew on a bar of Lava soap.
I’m somewhat relieved to hang on until the young woman with the repurposed jean skirt takes my place, comforted by the (entirely human) words of Julia McCoy founder of Express Writers, a writing agency where it sounds like there might be a decent chance of getting your oil changed.
Ms. McCoy says writers shouldn’t be worried…for now.
AI tends to repeat the same words and phrases because it can’t emulate natural speech patterns. Also, AI tends to repeat words and phrases, Ms. McCoy said.
GPT-3 lacks a passion for writing, she added. “You can’t feel the soul of the writer bleeding through” when it’s computer-generated writing. Huh. Most writers I know don’t bleed so much as break for a sandwich “or something” 12 times a day.
GPT-3, she says, also lacks empathy, which makes it a tough fit when talking to a human with, say, a customer service problem.
Agreed. Chatbots are notoriously short on empathy.
Me: “I’m having trouble canceling my subscription…”
Chatbot: “My name is Cathy and I will help you. What can I do for you today?”
Me: “I’d like to cancel my subscription.”
Cathy: “What is your problem today?”
Me: “Cathy, I need a human to help me.”
Cathy: “You’re welcome! Goodbye and have a great day!”
Yeah, I think my job’s safe…for a little while.