AI for marketing is getting better at content creation. But caution is needed.
Digital marketers still have a job of course. But it is not going to be quite the same job, as artificial intelligence begins its “second act”.
Yes, AI is still good at compiling, sorting and categorizing massive amounts of data. Only now it’s increasingly able to assist in creating content in ways it could not before.
All you need to do is give an AI app a specific input. Render for me an image of a 1920s gangster taking a selfie with a smartphone. That technology did not exist 100 years ago, but the image does look convincing.
AI has evolved, but how does one now use it? Do you replace people and automate their work? Put the AI in charge? Or have the AI assist the human?
Necessity is the mother of improvisation
There’s long been the promise of using AI to create visual and written content. “One year ago, it couldn’t do that. It seemed like it was always ‘getting there’,” said Adam Binder, owner, founder and CEO of digital marketing and SEO firm Creative Click Media.
Things are different now, a reality Binder encountered when necessity called. “I was trying to make a deadline when my human writer was not there.” he said. Turning to AI, Binder made all the necessary inputs to get the app to do a bit of copywriting.
The app he used was GPT-3 by OpenAI, a nonprofit AI research and deployment company. It uses existing copy and top search picks from Google as raw material to generate its output. “The writing tool created a tone of voice,” Binder said. “It’s scary how similar it was [to the writer’s copy].”
Still, “AI is a long way from replacing the writer,” Binder said. “It can’t come up with a thesis.” It can’t compare and contrast, and it isn’t generating copy at the same level as an op-ed piece in the New York Times, he added.
Google put down its marker, declaring that AI-generated content violates its webmaster policy . However, Google may not have the means of detecting such copy. “AI writing is perfect,” said Chris Carr, president and CEO of digital agency Farotech. “Humans make mistakes.”
Making it sing
GPT-3 sources its input from the top 20 pages of a Google search. While it can turn out usable copy, it is not “final”, or ready to print. “You have to hone it,” Carr said. “The human has to humanize it.”
Carr demonstrated GPT-3 during the interview. After entering the question “How to use a hammer?”, Carr got a series of responses broken down across a range of options. By cherry-picking bits of feedback, Carr was able to cobble together some 600 words of copy in about five minutes. It would take an average writer about an hour or two to produce the same output.
While the AI could produce copy, it could not “make it sing”. The AI-generated writing was flat, factual, and addressed the chosen points. The AI could not “connect the dots” or “show personality”, Carr remarked. A professional writer could write an argument or add a turn of phrase to enliven the copy, things the AI could not do.
If you included written content by a known author, then GPT-3 could mimic their writing style. But that still leaves the copy in need of editing and polish. “Great content comes from synergy. It’s about moving people’s minds with a great argument,” Carr said.
What AI should not do
Sadly, AI is vulnerable to “GIGO” — garbage in, garbage out. Asking an AI app like GPT-3 carries a risk that it will draw from content better left untouched and unread. The internet, from which it draws the raw content, still teems with misogynistic and racist content, cautioned Pieter Buteneers, director of engineering for machine language and AI at communications PaaS company Sinch.
“There are certain places you do not draw from.” Like Twitter, Buteneers noted. Instead, one should pull from news sites or Wikipedia. “Yes, there is some junk there, but it’s human-curated content.” AI is keyed to pick up patterns but has no way to judge what it is picking up. “It doesn’t have common sense,” Buteneers said.
Another potential for abuse is false content — spam or phishing — noted Carr. “The power to do bad is so easy, it’s ridiculous.” Google cannot separate the wheat from the chaff, he added. A good parallel example is the use of junk phone calls, now so commonplace that people will let voicemail screen the call instead of answering it.
In Sinch’s case, the company does offer a “fraud filter,” where its AI can make a judgment call on what looks “phishy”. “It can be wrong, but it is better to stop one false positive than let go the other 999.” Buteneers said. While suspicious material can be flagged, a human must make the judgment call whether to let it go or stop it cold.
AI is typically tasked with seeking bytes of insight from terabytes of data, but even so, one must worry about GIGO. AI must consolidate data and resolve identities to eventually achieve a customer-centric approach, where more of the data can be accessed and put to good use, explained Joyce Gordon, AI/ML product manager at enterprise CDP Amperity.
“This must be achieved before brands can move into using AI to make judgment calls — otherwise, they’re making judgment calls on partial or inaccurate data.” Gordon said. “The algorithm is only as good as the data that is being used to power it.”
“All AI models include human bias. Therefore, it’s impossible to exorcise all the ghosts in the machine,” said Paul Hebert, director of enterprise strategy at customer engagement platform Cheetah Digital. “Going forward, I’d say the trend would be using ML for tasks, AI for suggested judgment calls, and humans for making the final decision.”
Here Hebert cited the HBR article linked earlier. “As [Garry] Kasparov expressed it, ‘Weak human + machine +better process was superior to a strong computer alone and, more remarkably, superior to a strong human + machine + inferior process.’”
What digital marketers can do
AI’s capabilities will improve over time, despite its limitations. Digital marketers will have to learn on the fly and try to identify best use cases.
Carr offered this advice to marketers: “Get out in front [of AI] or get run over by it. This is coming, like it or not.” Also, understand how Google works so you can get the best outcome and the most thorough source content. Creative people are the glue that will make the content stick.
And finally, Carr revealed his biggest worry: if the internet is ruined by false content and spam, “search results become a wasteland of bots.”
“Try small first,” Buteneers said, but learn from it rapidly. Learn quickly from mistakes. “Jump in — with one toe at a time.”
“So, crawl, walk, run,” Hebert said. “[M]arketers need to move slower than the tech and as fast as the ethics will allow. Ethics are the governor of the AI engine.”