Writing and publishing are not beyond the capabilities of AI, but the robots aren’t taking over just yet, writes Shane Richmond
Artificial intelligence (AI) was once seen simply as a threat to manual jobs such as manufacturing, but it is becoming increasingly clear that it will reshape knowledge roles, too.
While a new World Economic Forum report forecasts that machines and automated software will be handling half of all tasks within seven years, it’s easy to assume that the creativity needed in journalism, content marketing and online publishing means they will survive the rise of AI.
But this argument was made by sceptics about the board game Go – until an AI beat one of the world’s top players. And, actually, AI is already demonstrating its writing capabilities.
Working with clear rules
We’re some way from AI being able to write features or a news story about complicated political manoeuvring, but it is being used to produce some fact-based sports and finance articles. Match reports are perfect, for example, because they follow clearly designed rules. There are two teams, a final outcome, a list of players who scored and so on. And this information is usually delivered in the same way every time. Organisations such as the Associated Press have been using ‘robot journalists’ to write sports reports since 2016, while Bloomberg automatically generates reports on corporate earnings releases.
Reuters is using a system called Lynx Insight that can analyse large data sets and identify patterns that could make news stories. These can be flagged to journalists, who can decide whether they are worth pursuing. If they are, the system will provide them with background information – say, on a company – so they don’t need to look it up. It can even write some of the sentences.
Reg Chua, executive editor of editorial operations, data and innovation at Reuters, says: “The real value is using machines to do what they’re good at and then presenting that to humans – that’s the best of both worlds.”
Chua gives the example of share-price data, where Lynx can deliver “insights that build up in complexity and depth” far more quickly than humans. It can go from “Shares of Johnson & Johnson (JNJ) are up 0.68 per cent to $132.07 at 21:01 GMT” to “Johnson & Johnson, based in New Brunswick, New Jersey, is expected to show a 9.2 per cent increase in revenue to $19.394 billion when it reports results, according to the mean estimate of 14 analysts, based on Thomson Reuters data.”
Both sentences were written by AI. Chua says that, in time, the system could be expanded to sports data.
The grammar checker on your word processor is a kind of artificial intelligence, as is your calculator – and both have been part of the office toolkit for a while now. In future, the grammar check might suggest sentences before you start.
To grasp how AI will change writing and content marketing, it’s helpful to understand that we’re talking about a very limited kind of intelligence. AI is an algorithm designed for a specific task. When you want to accomplish a new task, it’s easier to use a new algorithm. For example, we have AIs, like Amazon’s Alexa, that handle voice recognition and AIs, such as those produced by Google’s Waymo, that can drive cars. Both are ‘smart’, but you cannot teach one to carry out the tasks of the other.
Taking over the tedious
AI for specific tasks, however, is amazingly advanced and typically better than humans once it ‘learns’ what is needed – so anything that involves crunching large amounts of data and spotting patterns is a good target. It doesn’t forget things and it doesn’t get bored – both of which can be helpful for certain jobs in content marketing.
For starters, robots are already taking on some of the labour-intensive and time-consuming tasks around search engine optimisation (SEO) and social-media management.
Market Brew, for instance, simulates every possible change you could make to the SEO of your website, compares their effectiveness and then recommends the best actions based on ROI. It’s able to do this because its algorithms are trained to understand what matters to search engines based on vast numbers of simulated searches. Meanwhile, services like Cortex use AI to optimise social-media content by finding the best times to publish – again by modelling many more scenarios against content that has already been published than a human could test manually.
Improving the news experience
Media companies have been experimenting with AI over the past couple of years, both to help news consumers and journalists. Chatbots are one of the most common deployments of AI across the world of customer services and have been used by organisations such as The Financial Times, Guardian and Quartz to deliver news in a more interactive and conversational way.
This is now beginning to move on to voice assistants such as Alexa. Guardian readers who own an Alexa can say “Give me the headlines” or “Tell me the latest film reviews” and the AI will oblige.
It’s easy to imagine how this technology could be used in content marketing. A company website could ask visitors to describe their problem and then suggest related content-marketing case studies.
Sitting between reader-facing and writer-facing products are ideas like the BBC’s Juicer, a BBC News Labs experiment that ingested around 850 news sources from all over the world, identifying and tagging the concepts included. The articles were then “searchable and therefore useful for trend analysis”. For example, one developer built a responsive map showing worldwide headlines related to corruption in FIFA, the football governing body.
In 2015, the New York Times began a similar experiment with an application called Editor that scans an article even as a journalist is writing it and suggests ways that it could be tagged or annotated to connect it to relevant content in the paper’s archive. The journalist can then simply approve or reject the suggestions at the end.
The ability to spot trends that humans would miss, bring together data from thousands of original pieces and organise information by concept could be transformative for producing new content that people want to read, and then helping them actually discover it.
The limits of AI
But what about writing articles that aren’t largely based on data sets?
“A must of the world’s largest computer scientists have shown that the cost of transporting the sound waves into the back of the sun is the best way to create a set of pictures of the sort that can be solved.”
So began an article in The Economist in December 2017, written by an artificial intelligence program trained on articles in the newspaper’s Science and Technology section. The Economist noted that “although the sentences are grammatically correct, they lack meaning”.
Does that mean that an AI cannot write a news article or blog? Not quite. The Economist’s algorithm produced an article. It probably wasn’t asked to produce something that made sense, or that would interest an Economist reader. In the most literal sense, it succeeded in its task. With any kind of AI, how you define the task matters.
To get an AI to write a news article, then, one might feed it some data and set a task such as ‘assemble this information into an article that matches the style of our publication’. Machine learning would allow the AI to understand the style of the publication from previous stories, but it would still need some kind of guidance to understand which data goes where.
How is the AI to know that “Jenny Smith won the election last night” is a more interesting fact than “10 million votes were cast last night”?
IBM’s recent success in creating a computer than can compete in a debating competition suggests opinion pieces might one day be computer-generated. However, knocking on doors, making calls and building the rapport than comes with a face-to-face interview are going to continue to be human roles – for now, at least.
The content marketers of the future might well be algorithms – identifying topics, interviewing decision makers (other bots, naturally), writing, publishing and optimising articles – but that’s some way off. The immediate future for AI in content marketing is a co-writer who never forgets anything, never needs to sleep and is happy to do all the boring jobs while you get on with the fun stuff.