Is AI for content creation over-hyped or will all writers eventually be replaced by bots? Businesses don’t necessarily need more content, they need better content that actually performs.
AI is doing a lot to help streamline content marketing and management for companies across the board. You can get things researched, prepped, edited, and published in minutes (as opposed to days or weeks).
The problem is that while AI can automate time-consuming publishing tasks and help predict what people want to read, it can’t really write that well – yet.
Today, AI still relies heavily on stringing together concepts or facts into some semi-coherent ramble, but it can’t massage the phrasing or other intangibles that get customers to stand up and take notice.
AI’s underlying technologies in the area of content creation currently include Microsoft’s Turing Natural Language Generation (T-NLG) – boasting 17 billion parameters – and OpenAI’s Generative Pre-trained Transformer 3 technology (GPT-3), which has 175 billion machine learning parameters.
In September 2020, Microsoft announced it had licensed access to GPT-3’s technology for its ‘exclusive’ use, which offers a clue as to where this fast-growing industry is heading.
In this interview, Wordable CEO and Founder & CEO of Codeless, Brad Smith outlines AI’s current limitations for content creation, tells us how best to leverage its capabilities, and looks at where we are headed in the not-too-distant future.
MORE FOR YOU
Key problems with content AI todayarbage in. Garbage out.
Smith says the biggest problem with AI right now is its overreliance on patterns and the probability of certain words or phrases showing up next to each other when you reference certain topics.
“This means that currently, it can only take a mediocre pass at factual, information-based content. But even then it struggles to actually understand anything it’s saying. It is merely taking what’s already out there on certain topics and then playing a Robocop version of the word game Mad Libs,” he explains.
AI can’t string together long, persuasive text.
“Most good content builds on itself,” says Smith. “So you might lay the groundwork for an argument in the first section, and then come back later to build on top of that reader’s new understanding in the fourth section or paragraph.”
The trouble with AI is that it can’t reference itself in that way, says Smith. “Its knowledge of a few topics are completely isolated from each other, so it can’t ‘connect the dots’ that your reader would importantly be expecting to see.”
Smith says that these two issues alone would completely rule out AI-based uses for a significant amount of online content.
AI can’t do audio or video
He also goes on to point out that AI can’t do audio or video, nor write scripts for this type of media, “which is how most people will consume digital media in the next few decades, but that’s another topic for another day”, says Smith.
AI can’t do subjective
Most content online isn’t objective, but subjective. Comparing alternatives, or providing a few recommendations, each with its own pros and cons.
“Unless AI is basically robotically plagiarizing other content already on this subject, it can’t compare alternatives like this or provide additional context as to why one argument might or might not be legitimate.”
AI can’t do emotion
And AI can’t do emotion, such as style, jargon, inside jokes, meta-references, anecdotes, and storytelling. “All the things that get someone to stop dead in their tracks, take notice of what they’re reading, and actually want to continue reading the full thing. At the end of the day, people are still emotional human beings, hardwired via their centuries-old lizard brains to use feelings to convince themself of logical decisions, and not vice-versa.”
Where can AI help with content creation?Research and prep
Given that most long-form content (1,000-2,000 words) takes 4-5 hours on average to write, with maybe half of that for research and prep, Smith says that AI can be a huge help.
“AI and its underlying content technologies can help shortcut this dramatically, providing ideas for how an article should look, or what subtopics to mention, within seconds versus hours.
Pattern-matching works well for SEO
While pattern matching can create content “like a Robocop version of Mad Libs,” Smith says that AI-based research that’s heavily based on pattern-matching can help structure something for strong SEO.
“Search engines like Google only exist to help searchers find answers to their queries. To do that, a lot of content they show tends to be fairly formulaic, where the top 10 results might all mention certain subtopics, semantic ideas, questions they’re answering.”
First drafts – in specific cases
Smith explains that in some very specific cases, AI might be able to provide short-form, basic fact-based content that’s passable for a first draft. “Again, you’ll still want writers and editors to actually review it, polish it, edit, or add on. And again this might save you significant time and money, especially if you can work with AI to vet or manually approve an outline before the AI attempts to write it out.”
So while bots might not be replacing writers anytime soon, AI for content creation is developing rapidly. Given that GPT-3 launched just a couple of months later than T-NLG with 10 times the capacity of its rival, it will be interesting to see what happens next. Preliminary tests, on only 80 subjects showed only 48% success in distinguishing short stories written by people from those created by AI.
Smith’s key advice is: when using underlying technologies for content, ensure you keep humans involved at various stages of the process.
“Of course, you still need experts and humans to vet, filter, tweak, and throw out things to make sure it’s legit. Nevertheless, used correctly, AI can potentially be a huge timesaver, and will no doubt feature heavily in the future of content creation.”