If you read news, sports, product descriptions, weather reports, or just about anything else online, there is a very good chance that you have read something written by a computer with little or no human input and never realised it.
A company called Wordsmith has been writing automated content for The Associated Press (AP) since 2014. Using quarterly earnings reports, the software converts the financial data into news copy.
A study from Stanford University found that by drawing attention to the earning reports of smaller companies which would have otherwise gone unreported, AP’s AI produced reports increased the trading volumes and liquidities of thousands of smaller companies.
AP and Wordsmith are not alone. Many leading news sources including Reuters, Yahoo and The Washington Post are using automation to write stories.
As stories become more mass produced, the important crisis communications question becomes, not “is the incident reported,” but “is anyone reading the report.” Here AI plays a massive role.
Who (what) decides which stories we read? The AI systems that curate our social media feeds and our internet search results are arguably more powerful than the ones that write the stories. In the internet age, getting a story online is easy, getting people to actually see that story is much more important.
A shipping incident which impacts a beech in the Canary Islands, for example, would at one time have reached an audience on the Island and maybe in other parts of Spain, but probably would not have reached people in India or Canada. Now, as the systems that create our news feeds get more “intelligent” users will be shown stories that relate to where they grew up / went on vacation / honeymooned / etc., which fundamentally changes the vital process of audience identification in crisis communications.
Computers have for a long time been good at doing complex mathematics, saving large amounts of data and for the last two decades playing chess (things that most humans find difficult). AI is trying to teach machines to do things that even small children find simple, such as creating natural sentences, understanding human emotions and distinguishing between a cat and a dog, a bench and a chair and a ceramic model of an apple and a real apple. While AI systems are getting remarkably good at these things, they are still not nearly as good as humans.
AI’s biggest impact for journalists and crisis communications teams, will be in a support role where it can focus on doing the things that computers are good at rather than on replacing humans completely.
Reuters is one of the leaders in this area and earlier this year announced the launch of “the cybernetic newsroom” where machines will feed journalists, story ideas, details and tips to help them with their jobs. At the moment the cybernetic assistant is being rolled out in the financial news division where data is plentiful, but it will soon be rolled out in other areas.
I recently had the opportunity to trial the beta version of an AI driven media monitoring system. At the moment it struggled with shipping stories, and was much better with political and financial stories, but it is likely to improve rapidly.
In the initial stages of an incident it can be difficult to identify the early social media reaction because people often post photos and videos without titles or tags – with advances in image recognition we will potentially be able to monitor for all images of a vessel regardless of how they are (or aren’t) tagged.
Very soon we might be able to integrate the MarineTraffic data feeds, with publicly available webcams and social media images, to identify potential media crises long before the company’s head office is notified of any situation on board.
As with the development of any new technology, AI systems targeted at breaking news will create a sort of arms race for all players. Such contests are rarely won by either side, but instead just serve to increase the pressures on both sides – news will only get faster.