In 2014, I got into a heated conversation with a colleague. He argued that, within just a few years, robots would replace writers in content marketing. I rolled my eyes so hard my brain ached. The idea of machines crafting sentences as elegant and nuanced as a human writer was both heretical and impossible, I thought.
Turns out he was right, but the development isn’t the sci-fi creative dystopia I imagined … in most cases. Today, companies indeed use machine learning to generate natural language; and marketers are using AI-powered tools like Narrative Science and Automated Insights to produce practical, well-constructed, high-performing content.
All sounds well, yes? Not so fast. Some applications for robo-writing are deeply problematic — not because the quality of the language generated is poor, but because it’s so believably human. I’ll get to that, but before I raise the red flag of caution, let’s talk a bit about the full range of applications for robo-writing currently available.Some robo-writing apps are deeply problematic because the language is so believably human. @clare_mcd Click To Tweet
In content marketing today, there are primarily three forms of machine-made content:
- Modular content: Highly templated, formulaic content; it’s often data-driven, and the scale and scope of those content projects make them a fantastic candidate for machine-made content.
- Creative support: Tools that give writers superpowers. Rather than writing from scratch, these AI-powered tools enhance the creative and analytical powers of human writers.
- Messaging chatbots: Interaction bots, powered by AI and using natural language processing, sustain one-on-one conversations with prospects and customers — through social media, web applications, or even email.
AI-driven robo-content goes mainstream
The most exciting opportunities for using modular, machine-made content are the types of projects that are predictable, repeating, following a template, and in most cases highly reliant on data.The best opportunities for machine-made content follow a template and are reliant on data. @clare_mcd Click To Tweet
Years ago, when I left traditional employment to become a freelancer, one of my first gigs was transcribing quarterly earnings calls and preparing summaries for the press. Writers like me felt immense pressure to complete tasks quickly because each assignment was priced at a modest, flat rate. If you didn’t move at manic speed, your hourly rate could dip close to minimum wage. (I lasted about two weeks.)
Today, these tasks are completed by machine learning technologies. The Associated Press, for example, uses robo-writers from Automated Insights for corporate earnings calls. In doing so, AP freed up 20% of reporters’ time, reduced errors, and improved turnaround times.
When the program went live, Philana Patterson, an assistant business editor at the AP who was in charge of implementing and overseeing the system, explained it this way: “One of the things we really wanted reporters to be able to do was when earnings came out to not have to focus on the initial numbers. That’s the goal, to write smarter pieces and more interesting stories.” (While there may be some hand-wringing over machines replacing human writers, trust me when I say this type of job is not one that leads to a satisfying freelance writing career.)
AI tools also help companies convert data into stories. Dominion Dealer Solutions helps auto dealers create customer-ready profiles of cars for sale on the lot. In the past, a dealer might cut and paste information from disparate sources to create a sell sheet — a process that is wildly impractical for dealers with thousands of new and used cars in inventory. Using Narrative Science, dealers can access a sell sheet in real time, unlocking plain-English descriptions from a massive trove of vehicle data.
This application isn’t simply about the sales sheet, of course; it’s also about SEO. Using robo-writing helped Dominion Dealer Solutions increase page views by 20% (including a 50% lift in page views across used vehicles), as well as increase inventory revenue.
Creative support applications
These cases show AI replacing humans, yet there are plenty of examples of AI supplementing the work of individual human writers. To understand the future of robot-assisted content, it’s useful to look at the manufacturing industry, where more and more companies are using what are called “cobots” — robots designed to interact with and assist humans in a shared space.
Cobots often perform repetitive tasks in partnership with a human (e.g., packaging, “pick-and-place” tasks, quality checks) and eliminate or minimize the more tedious and dangerous tasks on assembly lines.
AI-powered writer-support tools look a lot like the relationship between human and cobot. The most common content cobots are Google’s Smart Reply and Smart Compose features in Gmail. As you type, Google predicts the words that likely follow. To date, the problem with predictive content cobots like Google’s is they don’t deliver on the promise. Google’s Smart Compose isn’t particularly timesaving: The keystrokes required to complete the sentence are awkward enough that in most cases it’s simply faster to type the same words.
Give it time. As Google gets access to even more of your data [shudder], its deep-learning AI will more accurately predict larger tracts of what you’d like to say. Ray Kurzweil leads the effort, and his team is working on a way to expand on Smart Reply. He explains, “You could have similar technology to help you compose documents or emails by giving you suggestions of how to complete your sentence.”
Aside from these types of predictive, NLP cobots, there are dozens of creative-support cobots on the market (see my picks in the sidebar). Among my favorites is Frase, an AI-powered word processor with an integrated research assistant.
As I write a story, Frase searches for third-party media and research, and summarizes relevant sources. The tool even creates footnotes/endnotes based on the information you cite. Frase isn’t intended to replace human storytellers but rather make them more efficient and effective.
Human fakers may spoil the party
All this seems relatively benign, no? For the most part it is, but a small and growing number of AI-powered interaction tools are blurring the line between machine-made content and something a bit more disturbing.
AI-powered interaction tools blur the line between machine-made content and something more disturbing. @clare_mcd Click To TweetMarketers are well aware of chatbots — AI-powered interaction tools that can answer simple customer questions and replace the role of humans for more mundane conversations. Most chatbots divulge their machine-identity in these interactions, even though an astute observer can usually tell when they are interacting with a machine. But all this will change quickly.
Lisa-Maria Neudert is a doctoral candidate at the Oxford Internet Institute and a researcher with the Computational Propaganda Project. Among other things, she studies how businesses and political groups use chatbots to interact with their audiences. Neudert predicts that as machines become more adept at imitating the quirks of human language, as well as carry on wider-ranging conversations, people will be a lot less able to identify bots. And this change will make it easier for bad actors to exploit chatbots.
She explains, “[Propaganda bots] will present themselves as human users participating in online conversation in comment sections, group chats, and message boards.” What is particularly frightening about the development is the ability for bots to engage in highly customized, one-on-one, private conversations … and to do this at scale. In the MIT Review, Neudert writes it won’t be long before “conversational bots might seek out susceptible users and approach them over private chat channels. They’ll eloquently navigate conversations and analyze a user’s data to deliver customized propaganda.”
To be clear, there are powerful and positive applications for bots in content marketing. Conversation-bot maker Spectrm, for example, created a public health chatbot to offer advice about the morning-after contraceptive pill. Spectrm co-founder Max Koziolek explains, “It is one of those times when someone might prefer to speak to a bot than a human because they are a bit embarrassed. They talk to the bot about what they should do about having had unprotected sex and it understands naturally 75% of queries, even if they are writing in a language which is not very clear.”
Yet, good examples aside, it won’t be long before marketers can develop messaging chatbots that can so closely mimic human language they will be tempted to use them more covertly.
While marketers don’t have to divulge when content is machine-made, they absolutely should when conversations are machine-made. California has already implemented a law — dubbed Botaggedon — that says companies must disclose when users are interacting with a bot (e.g., “Hi, I’m a bot”). The law is directed mostly at social media chatbots, leaving companies to figure out how to apply it to other use cases.
It’s still about message, not method
Despite the legislation in California, marketers are currently in a grey space with the technology evolving much more quickly than the regulation of that technology. Just look at the way governments are playing catch-up to the privacy misdeeds of companies like Facebook.
People have known for years that Facebook, Google, and thousands of other smaller players are monetizing people’s personal data. But governments and consumers were slow to realize the implications of this data exchange.
The same is true for bot-powered conversations. Governments are still grappling with how to limit bot-armies influencing elections and political activism. They haven’t yet begun to wrestle with the implications of business-sponsored bot conversations. (Let’s be real: it’s not even on their radars.)
Given that regulation won’t likely arrive in time to curb bad behavior with machine-generated conversations, perhaps consumer choices will. A study from the Pew Research Center in October 2018 found only half of U.S. consumers believe businesses using bots to promote its products is acceptable (not exactly a ringing endorsement). And among those who have a higher level of understanding of what bots are, the acceptance rate is significantly lower.
As marketers race to find new and different ways to reach their audience (hello obnoxious lock-screen advertising, uninvited text messages, and web browser notifications), consumers are finding new ways to limit their access. Ultimately, it’s still about the message, not the method. Publish great content, provide useful information, and your audience will open the door to let you in — whether that content is created by machine or human.
Unless stated, all tools mentioned in CCO are suggested by authors, not the CMI editorial team.