Could a robot write your copy? Yes.
It’s not an exaggeration to say that Artificial Intelligence (AI) has infiltrated pretty much every aspect of our lives.
So it’s no surprise to find AI (and especially GPT-3) already embedded in the way we create copy and content.
For better or worse, writing and content creation is the next frontier for writing and human creativity – here’s how to prepare for it.
Think of the predictive text that Google suggests when you’re composing an email.
Or the system-generated responses you get when you contact a Customer Support Representative through Live Chat.
That’s all AI.
But it doesn’t end there — in fact, those earlier forays into AI-driven content were just the beginning.
GPT-3 and the Rise of AI-generated Content
When it was released last year (amid much media fanfare), GPT-3 was hailed as an artificial intelligence breakthrough, and had publications like the MIT Technology Review raving about its capabilities.
So what’s all the buzz about?
First, a little background. GPT-3 stands for Generative Pre-Trained Transformer 3 (it’s the third iteration of the tool developed by OpenAI, an AI research lab based out of San Francisco).
Essentially, GPT-3 is a language generator — it’s designed specifically to produce streams of human-sounding text based on a simple opening prompt.
So, how does it do that?
GPT-3 is a neural network. In AI terms, that just means that its computing system mimics the way the human brain processes information.
And like the human brain, it strives to identify patterns in vast amounts of data — in this case, digital data — so it can quickly recognize, sort, and process new pieces of information, based on what it has already encountered.
Once it’s “trained” (that is, once it’s fed an adequate amount of data to learn linguistic patterns) it can predict what words are likely to come next when someone types in a few words as a prompt.
Why GPT-3 Works: Massive Datasets and Sophisticated Algorithms
What sets GPT-3 apart from other language models?
The sheer amount of data it’s been “trained” on as part of this pre-training process is unparalleled.
OpenAI researchers trained it with 570GB of data, including the largest dataset scraped from web pages, Wikipedia, and digitized books.
GPT-3 is also what researchers call a “few shot” learner.
With all the information it already has about words and linguistic usage, it only needs a few examples of a task in order to learn how to perform that task.
So the list of language-related tasks it can perform is potentially limitless.
But it doesn’t end there.
Not only can GPT-3 predict what words to use together when generating text, it can also mirror the form and rhythm of a written task.
You can give it instructions like “translate the following sentence” or “write me a poem about cars”, and it’ll spit out the response in the right format.
In fact, The Guardian had some fun with this when it got GPT-3 to generate a full article.
Obviously, the possibilities go far beyond generating text for chatbots or suggesting a response to a common customer support request.
GPT-3 can perform a lot of language-related tasks, but it’s the text generation that’s most relevant for anyone who produces content.
So the question remains: Is AI going to replace human writers?
Why GPT-3 Doesn’t (Always) Work: Garbage In, Garbage Out
Despite all of its promise (and, let’s be honest, hype), GPT-3 has its drawbacks.
Like any machine learning model, it’s prone to bias.
Experiments have already shown that GPT-3 generates text with biases against certain populations.
Researchers are taking steps to mitigate this effect, but given the vast amount of data already ingested, it’s not clear that these are more than Band-aid solutions for a much bigger problem.
Since GPT-3 was trained, in large part, using data from the internet, it might end up reproducing the disinformation that has become widespread in the online world.
But apart from its tendency to amplify human bias, GPT-3’s ability to learn and predict from existing content sources may be a double-edged sword, at least for content creators: It’s trained to produce more of the same.
Now, this can be very helpful sometimes — a jumpstart on a new writing project, as a shortcut to a first draft. But if you’re in the business of grabbing someone’s attention with your content, you need to stand out, not blend in.
Think about it in terms of the “long tail” of the distribution curve.
Less common ideas or uses of language may be some of the most interesting and engaging — but GPT-3 isn’t designed to generate that kind of content. It’s designed to mirror what it sees the most of.
And for content creation, that’s a recipe for getting lost in the tsunami of content that’s already out there.
AI (and GPT-3) Is a Writer’s Friend
Even if tools like GPT-3 might not be accessible (GPT-3 itself is a closed API, and is extremely expensive), AI is not the writer’s enemy.
In fact, there are many AI-driven writing tools that are worth your time and attention.
AI-driven writing tools can be a boon to your writing, and most of us are already using them.
You can install some lightweight tools as browser extensions. Grammarly checks your grammar, while Wordtune suggests better phrasing.
Email plugins like Just Not Sorry will flag words that make you sound weak or deferential. This can help you improve your writing in real-time, on the fly.
Other self-styled “online editors,” like Hemingway, scan your text for overall readability and flag things like jargon or passive voice.
Then there are other, more comprehensive tools that can do more of the heavy lifting for you.
Tools like Copysmith and Writesonic can generate full-fledged copy, from blog posts to PPC ads.
Persado, another AI-driven writing service, can provide personalized content, and claims it can predict which ideas and messages will resonate most with an audience at different touchpoints.
Help like this is a marketer’s dream, and it can be a good starting point for a copywriter willing to do the work to polish the words.
The Future of Copy and Content Includes AI (with GPT-3)
There’s no question that AI is changing — even disrupting — the way marketing organizations and copywriters produce content.
So does that mean the machines will soon be running the show?
No, not by a long shot.
Like any disruptive technology, it will create challenges — and opportunities.
