AI's Learning Writing Skills, But That Might Not Be a Good Thing

It could not have written this story

  • AI tools are now ready to write blog posts, reports, and tweets. 
  • The results can be realistically bland or surprisingly human. 
  • We may be more sensitive to written plagiarism than we are to ripping off visual artists.
Top down view of someone's hands on a laptop keyboard with a smartphone sitting next to the laptop.

Cytonn Photography / Unsplash

Artificial intelligence (AI) creation tools have exploded this year, from Dall-E to Stable Diffusion for images to Facebook's text-to-video tool. Now, AI is coming for writers.

Notion, the fancy online note-taking database app, just announced Notion AI, designed to automatically write blog posts and reports or brainstorm ideas. You just type in a short prompt, and the AI bot does the rest. It sounds amazing, especially for folks who hate to write but have to generate reports as a part of their job or as a class requirement. But are these tools up to the standard? Can they really replace human writers?

"I think this will be hardest on the writers who have allowed themselves to specialize in SEO content creation. You're already writing with the machine as your customer. Now the machine is going in-house," professional journalist John Brownlee told Lifewire via Mastodon. "For the rest of us, I don't think this is much of a threat and is possibly even beneficial as a drudgery-reducing tool. But just like everyone has already learned to spot the AI-generated art, people will quickly learn to feel the contours of the uncanny valley in writing AIs." 

Fancy Notion

2022 is definitely the year of AI tools. They seem to be coming out almost weekly, and they are actually functional and often impressive. For example, Dall-E, Stable Diffusion, and Midjourney all let you create images in many styles with just a text prompt. These machine-learning models are trained on zillions of existing images, and when you type that prompt, they rehash those human-created pictures, often to startling effect. 

Text AI uses similar methods, building language models from existing works. And in addition to straight-up creation, it can also do translations and fix spelling and grammar. 

"It seems like Notion is planning to mainly help out with inspiring ideas, first drafts, and summarizing—all things that, for most copywriters, are just tedious and take away fun and creativity from the job. For now, tools like these seem to be a great way for companies to take care of the most repetitive tasks that don't actually need any human involvement," self-described content writer Martina Lupi told Lifewire via email. 

You can take a look at a video showing off Notion AI in action, and the results are—less than impressive. They are, in fact, just as bland and uninspired as you'd expect from any report or list grudgingly produced by a human. The list of brainstorming ideas for Notion AI's launch is particularly dull. 

Plagiarism and Jobs

Judging from these examples, text AI isn't going to be taking over creative writing jobs any time soon, but it could be a useful tool just to get started. Just like an artist generating multiple Dall-E images to give them inspiration for their work, an AI report could give structure to a first draft. 

"We utilize AI writing tools every day to help streamline processes," website creator Stephen Heffernan told Lifewire via email. "AI can help you to come up with new ideas by providing inspiration and suggestions you may not have thought of. It helps give the 'spark' some writers need when they get stuck.

But there are other tools out there that are startlingly realistic. Researcher and writer Dr. Bron Eager set the Jasper AI loose on a blog post and published the result. Thankfully, Eager makes it clear which part is the AI and which part is her own introduction. Otherwise, you might not be able to tell the difference. 

Two robots typing on latops.

iLexx / Getty Images

So should writers be worried? It depends. 

"No matter the context, the writing that people actually want to read is authentic, and authenticity comes from idiosyncrasy," says Brownlee. "I have no doubt that an AI can successfully generate 99% of the SEO-optimized filler on the web, but no one actually reads that [stuff]."

And then we get to plagiarism. The image AI bots have been criticized for using copyrighted work without permission because they are. But whereas few people seem to worry about stealing images, plagiarism of the written word is still frowned upon to the extent that politicians have resigned over it

What happens when somebody uses AI? Who is guilty of plagiarism if the results are too similar to existing work? That probably depends on the kind of work you are creating, but the specifics will get murky very fast when this goes mainstream. Which, judging by the velocity of AI right now, could be very soon.

Was this page helpful?