The pun-loving computer programs that write adverts
Machines are now writing advertising copy as well as basic news reports, but are their efforts any good and can they be taught to be more inventive?
"Have a suite stay" read an ad for a hotel offering all-suite rooms. A neat - if obvious - pun you might think.
But what made this ad noteworthy was that it was created by an automated copywriting programme developed by Dentsu Aegis Network, the marketing giant.
The firm launched its natural language generation algorithm last year to increase output after changes were made to Google's advertising system, explains Audrey Kuah, the firm's managing director.
The programme creates 20 to 25 full ads a second in English and is "trained" by feeding it thousands of the kind of ads it is meant to produce, she says.
But what caught their eye about this ad was that it was quite witty.
Google's "cost-per-click-basis" advertising system, whereby the cost of an ad falls the more it is clicked on, encourages clients to play it safe, says Ms Kuah, making the ads rather pedestrian.
The algorithm couldn't learn to be more creative based on such a back catalogue of humdrum copy.
"We got quite fascinated by how we inject this concept of creativity," she says.
So they began to "feed" the algorithm with editorial headlines from travel articles and idioms to see if it could learn "more flowery" language.
- Could a computer ever create better art than a human?
- Russian TV launches robot news presenter
- US lawmakers to probe algorithm bias
- China unveils AI news presenter
"Our ambition is to train this AI [artificial intelligence] copywriter to learn how to inject a little bit of that human creativity, which today is taken out of the search advertising system because it may not be so readily rewarded," she says.
The idea of making AI more human-like and inventive is already happening to a certain extent in China.
Retail giant Alibaba, for example, enables merchants on its e-commerce ecosystem to dictate the tone of the language when using Alibaba's AI-generated copywriting service, a company spokesperson tells the BBC.
On the Taobao shopping site, for example, merchants can choose between descriptive "short-title" copy, more promotional "selling point" copy, and more emotional "heart-warming" copy.
One heart-felt ad for a hoodie read: "A windbreaker is enough to withstand the autumn wind in England".
The AI copywriter learns from millions of existing samples and can generate 20,000 lines of copy a second in Chinese, the spokesperson says.
While the copy created is not necessarily perfect, the service makes life easier for the many, smaller merchants on Alibaba's e-commerce sites, which do not have the resources to do a marketing push themselves.
"A single product might require up to 10 versions of copy for different advertising formats, like posters, web banners, product pages, and event pages," says Li Mu, director of Alimama Marketing Research and Experience Center. Alimama is Alibaba's digital marketing arm.
"Many merchants, and especially smaller ones, lack the marketing expertise or resources. We aim to solve this problem with easily-accessible and user-friendly technology."
Companies generally welcome the greater use of AI, says Parry Malm, chief executive of Phrasee, an AI-powered copywriting firm, headquartered in the UK.
An AI-generated email subject line for a Virgin Holidays campaign continuously outperformed a human-written one over a testing period, he says.
"Shop the sale - don't hang around, book today!" proved more popular than the human-written "There's still time to book that dream holiday for less".
"Ongoing testing resulted in a revenue increase of several million pounds for their email campaigns - which Virgin Holidays has confirmed was a direct result of using Phrasee's AI technology for email marketing," says Mr Malm.
But the scope of AI can be limited and making it more "creative" is not without its challenges.
Ms Kuah says the Dentsu Aegis algorithm sometimes gets confused when you give it new information.
"It will start to go haywire," she says. "You will suddenly have things that don't make sense appear. So it's a little bit like teaching a wayward dog that doesn't want to sit."
Phrasee meanwhile seeks to avoid what it calls "simplistic emotional tagging" altogether, says Mr Malm.
Instead, it tries to differentiate itself by allowing clients to create AI-generated copy that adapts to the "voice" of a particular brand through a "bespoke language model".
"The traditional approach for many marketers and providers has been to tag language with certain emotions - 'happy', 'sad', 'scary' - but such an approach is problematic," argues Mr Malm.
"It not only limits the language available to use but is highly susceptible to human error and prejudice. In other words, it's not scientific."
While the use of AI in advertising copywriting is still very niche, such automation is widely employed in the selling and distribution of digital ads, and increasingly, in journalism.
"The adoption of AI is growing significantly among advertisers with the increasing use of mass marketing," says Venkata Krishnan Seshadri, industry manager for information and communications technology at market research company Technavio.
- The doctor who invented 18 medical devices
- Conjuring designs from thin air in a virtual world
- Could aviation ever be less polluting?
- Why is the white hot Chinese tech sector cooling down?
- Could high-flying drones power your home one day?
"In 2018, over 40% of advertisers used AI for analyzing their target audience. This is expected to increase in the coming years, as numerous organizations are utilising their AI capabilities to streamline their marketing and sales process."
The bottom line is that if AI produces better responses to adverts than humans manage to achieve - and at lower cost - marketers will jump at the chance to use it.