Mark Henricks 15 Jan 2020 // 8:56PM GMT
In December, angel investor Cyan Banister tweeted that she was using AI to take pitches, in part, to weed out sectors that she’s not interested in investing in. The tweet came on the back of Google announcing it is adding an artificial intelligence-powered writing tool, Smart Compose, to its G Suite offering. The feature, according to Google, is designed to help “you compose high-quality content in Google Docs faster and more easily.”
The implications for public relations seem obvious: how long before AI is deployed for journalist outreach and other forms of communications? Kyle Arteaga, co-founder of tech public relations agency The Bulleit Group, says it’s a better question of when will AI write for PR. The only issues to be resolved are the timing and precise details of the transformation.
That’s not necessarily because there’s anything special about PR, Arteaga says. “Literally every client we have has some form of AI tech story,” he explains. “It’s forced us to find a way to be more up-to-date not only with the AI that’s within our companies’ products, but also how it’s used in everyday life and, more specifically, how it’s used in journalism.”
Pointing to Banister’s tweet, Arteaga noted there are similarities between financiers screening investment opportunities and reporters screening requests for coverage. “How far are we from journalists using AI to parse through received pitches or using AI to better target pitches?” Arteaga said.
At this point, there’s no indication either is happening – yet. But AI could come shockingly close to having the chops to master the highly specialized craft of writing for PR. For instance, in a New Yorker piece from last fall, journalist John Seabrook trained GPT-2 to write an article in the style of The New Yorker – and came tantalizingly near to succeeding. (Research lab OpenAI released GPT-2 last month, although experts warn it could generate dangerously convincing disinformation and spam.)
I spoke to Seabrook about this experiment — and he suggested that PR should take a closer look at AI-powered for writing. “I’m not saying AI can write better than most human PR folks, but it can come close enough so that you can begin to think about deploying the humans for something else,” he says.
The Limits of AI Writing
Seabrook explains the potential for GPT-2 in his New Yorker piece: “GPT-2 was designed so that, with a relatively brief input prompt from a human writer—a couple of sentences to establish a theme and a tone for the article—the AI could use its language skills to take over the writing and produce whole paragraphs of text, roughly on topic.” The malicious implications are fairly obvious, but he goes even further to ask — could it write a New Yorker article? The answer was muddled.
AI is good at looking at data and writing short articles describing it
In Seabrook’s experiment, the text alternated between refined sentences displaying an eerily convincing imitation of the magazine’s distinctive style and crashingly out-of-place verbal blunders that wouldn’t get past a just-fledged copy editor. It is clear that GPT-2 will not soon take a place on The New Yorker’s masthead.
Seabrook writes in his feature: “GPT-2 was like a three-year-old prodigiously gifted with the illusion, at least, of college-level writing ability. But even a child prodigy would have a goal in writing; the machine’s only goal is to predict the next word. It can’t sustain a thought, because it can’t think causally. Deep learning works brilliantly at capturing all the edgy patterns in our syntactic gymnastics, but because it lacks a pre-coded base of procedural knowledge it can’t use its language skills to reason or to conceptualize. An intelligent machine needs both kinds of thinking.”
AI evolves far faster than a human child, however. Writing in the MIT Technology Review in December 2019, journalist Karen Hao reported that Uber AI developed a way to use common-sense prompts to keep AI-powered writing focused. “While the model still doesn’t understand meaning, the technique brings more control,” Hao elaborated. “It takes us one step closer to bringing the leaps in AI-generated language to more domain-specific applications, like health care or financial services chatbots. It could also be used to guide models away from producing offensive results.”
Having said that, it’s also well-publicized that media outlets, including Forbes, Bloomberg and The Washington Post, have been using AI to gin out short pieces using company financial reports and sports box scores as raw material. In fact, AI’s best journalism use is probably for automatically analyzing structured data, says Clive Thompson, who reports on technology and science for The New York Times Magazine, Wired, Smithsonian, and others and is author of Coders: The Making of a New Tribe and the Remaking of the World.
I’m not saying AI can write better than most human PR folks, but it can come close enough
For instance, he says, The Los Angeles Times uses AI to monitor the readings coming from earthquake sensors and generate news alerts about tremor activity. “AI is good at looking at data and writing short articles describing it,” Thompson explains.
Looking at data is a good example of where AI can really help, Thompson continues. “The most promising use of AI is not replicating human judgment, but doing things humans can’t do,” he says.
However, tech media consultant Sam Whitmore guesses AI’s first major use in PR will be actually generating content from large masses of unstructured data in the form of previously published reports. “AI will spot opportunities among all that has been published before,” he says. “On the other hand, humans will beat AI at imagining opportunities irrespective of what has come before.”
“For both PR and editorial, AI will make SEO unimaginably effective,” Whitmore adds. “AI will be able to sift through enormous data sets and craft articles beyond the capabilities of even the best analytical reporters.”
'Seems like AI Could be of Enormous Help'
When asked whether PR professionals should use AI for crafting pitches, Seabrook says it’s possible. “Of the many thousands I have received over the years, I could count on one hand the number that have led to stories I’ve written,” Seabrook told me. “This isn’t a very good ratio. Surely the PR industry can do better. Seems like AI could be of enormous help in several areas.”
Seabrook suggests an AI pitch generator could parse a writer’s previous work or style more cost-effectively than a human. “I get it that no one has time to tailor every pitch to each individual writer, or each publication, but an AI could do it, and if it wasn’t obvious that it was an AI, I’d imagine blind pitches would have a much greater chance of succeeding,” he says.
For his part, the New Yorker writer says he’d welcome better-targeted pitches from an AI and he expects other journalists would as well. “All they care about is whether the pitches are topical and whether it can save them some time,” he says. “That doesn’t mean relationships don’t matter anymore, because they do. But they don’t matter as much as they once did.”
Not everyone finds machine-written pitches agreeable. Forbes Associated Editor Alex Konrad, in a Twitter exchange on the topic, tweeted that he doesn’t think it’s a good idea. “This would further commoditize whatever the news was, meaning it probably wouldn’t be worth the trouble to write,” Konrad warned. He guessed AI would have more positive impact helping to flag events such as financial filings or lawsuits that could trigger coverage.
Thompson, however, feels using AI to target pitches makes so much sense that it’s inevitable. “That’s certainly going to happen and would probably be good and better than the spray-and-pray style.”
It almost sounds as though AI researchers were listening in to similar comments from media professionals. As described by MIT Technology Reviews' Hao, the Uber AI prompting technique nearly directly addresses this need to better target pitches generated by AI-powered writers that predict upcoming sentences by referring to a model.
When an initial prompt is fed into the first model, it begins the process of predicting subsequent words. But after every word, it checks its score with the evaluation model and readjusts on the basis of the feedback,” Hao explains. “The final sentence ends up with the desired attribute, while also retaining the giant language model’s fluency."
'AI is Going to be Talking and Writing to You'
Use of AI in professional communications including journalism and PR is increasing, although Thompson suggests it’s not happening as quickly as some observers may think. But Seabrook’s deep dive into machine-generated writing has convinced him that big change is on the way. “An AI is going to be talking to you and writing to you,” he says. “I think it’ll get more and more familiar. It’s not going to get less familiar.”
That doesn’t mean relationships don’t matter anymore, because they do. But they don’t matter as much as they once did.
Arteaga advises that it wouldn’t be wise for any communications professional to ignore the potential impact of the technology. “What’s wise is understanding how the computer works and the algorithms work so you can deploy it,” he says. “Because if you don’t someone else is.”
Meanwhile, Arteaga raises the intriguing possibility that AI-assisted pitching and writing could help boutique firms compete effectively against the goliaths of the PR industry. After all, he notes, similar scenarios have played out as technology transformed countless other industries.
“Those firms that adequately use technology and are constantly pushing the edge outperform the incumbents.” Arteaga says. “That’s what’s going to happen in PR and journalism, too.”
This article is part of a series exploring the topic of innovation in partnership with the Bulleit Group. You can read more from the series here.