by Andrew Cochran
January 1, 2019
updated October 2019
AI systems in newsrooms are doing tasks not jobs. Examples are outputting templated stories from structured data or clustering social media posts to flag breaking stories. Three themes have emerged:
1 – SPEED-TO-MARKET
For example: time-sensitive information in minutes or less
Among the earliest implementations was QuakeBot, an algorithm at the Los Angeles Times that monitors and analyzes seismic activity. For any earthquake registering more than 3.0 on the Richter scale it prepares a story, associates a map, and pings a Times editor for a vet. A QuakeBot story can be posted in minutes or less. Or at Bloomberg, where milliseconds in market reporting may have financial consequences, an algorithm can be prepped to parse forthcoming financial results and analyze if the results meet, exceed, or fall short, note any salient trends, and have a story ready for an editor’s approval, seconds after the information is released.
2 – SCALE-TO-MARKET
For example: large numbers of convincingly-original stories in fact-laden subject areas
An automated publishing system enables the Associated Press to increase its quarterly earnings reporting from 300 to 3,700 stories, without degradation in quality. This kind of system needs (a) data that is well labelled, aka structured data, and (b) a fixed format for using the data, such as a template. Corporate earnings, sports scores, and weather are examples of structured data and they are the kinds of stories presently done most often by AI systems. At RADAR, a start-up paired with the UK Press Association, six journalists prime an AI system to parse public data reports (like housing or crime figures) and prepare news stories that are customized to hundreds of local districts.
3 – DEPTH-TO-MARKET
For example: associating pieces of information to provide context quickly
Research-oriented systems present writers with a range of relevant materials automatically assembled by an AI-enabled search. Additional facts, references, or pictures, generally providing more context with less effort, are delivered to the writers’ desktop. Some can also suggest new stories, new angles, or additional text. Two such systems in use are the Lynx Insights tool at Reuters or the Bertie system at Forbes.
A Force Multiplier
AI systems act as a force-multiplier, enabling people to do many more things. The Associated Press says automated processes deliver a 20% time efficiency, which they have redirected to tasks better suited to humans.
- First stage AI systems automate
productionof simple, fact-intensive narratives, such as earnings reports or fantasy-league sports results
- Second stage systems augment the work of human journalists, for example, presenting options to enhance their writing. The choice to accept, reject, or modify machine-made suggestions remains a human decision.
Examples of other ways AI tools are being used in news work:
- Moderating, discerning between acceptable and unacceptable reader’s comments
- Transcribing, converting spoken words into text very rapidly
- Monitoring, ‘listening’ for key words in briefing calls
- Translating, allowing news gathering to be done in native languages other than the language of publication, or making possible presentation in other languages
- Simulating responsiveness, having ‘news bots’ interpret user questions or replies, and then responding in a way that mimics conversation. Taken further by ‘intelligent assistants’ like Alexa, Google Home, and Siri, in some instances tailoring news presentation based on user input
- Predicting reader relevance, weighing prior choices to predict likely interest, and then selecting, displaying, or ranking stories accordingly
- Flagging news-in-the-making, for example in social media: searching for keywords (like ‘shooting’ or ‘accident’), clustering a number of similar posts, and then alerting a human journalist that follow-up may be warranted. Or calculating trends or anomalies, such as fluctuations in GDP or currencies, again to ‘suggest’ a human journalist look further to determine its significance.
Is AI Good or bad for journalism?
Possible favourable outcomes:
- AI systems can free time for journalists to do more fulfilling work. Machines will take care of routine needs so humans can apply advanced critical thinking skills to imaginative or investigative projects.
- AI tools can open new areas of journalistic inquiry using AI systems to analyze data otherwise unreachable or unfathomable by human reporters.
- AI systems can improve relevance for audiences, helping match stories to interests.
Possible negative outcomes:
- AI journalism can create legal entanglements, such as challenges in copyright, privacy, and defamation.
- AI generated material introduces ethical dilemmas, for example an inability to verify results derived by algorithms.
- AI systems can reduce opportunities in entry-level positions, and eventually, might lead to layoffs.
So far, news practices using AI systems are receiving positive notice. They are removing drudgery while enabling greater audience service. Current and expected AI systems are augmenting, not replacing, human journalists. It remains to be seen if this continues as AI systems become more capable in more areas.
Currently, AI operations in many of the largest newsrooms are achieved with in-house systems. They may be supplemented with outsourced suppliers for certain tasks. The number and range of third-party suppliers is growing. Newsrooms still commencing AI capabilities have the choice of building their own systems or buying capacity from third-parties. Costs can be significant in either case, creating a growing potential for a divide between ‘have’ and ‘have-not’ newsrooms.
Today’s force multiplier effect could, for others, become a force creator. New entrants and competitors could gain legacy-scale capabilities with new systems, similar to how desktop publishing upended the industry. More alternatives could alter consumer behaviour with increased choices, diluting the addressable market for existing publishers.
We could, over time, see a transition from machines augmenting human journalists, to human journalists making final judgments for AI-generated journalism. It would all still be augmented journalism. The question would be who, or what, is augmenting who?
The difficulty of seeing ahead
In computing most AI scientists say the field is still at an early stage. Advances in AI are hard to predict, ironic as it is for a technology that exists by prediction. Deep learning pioneer Geoffrey Hinton worked on principles of deep learning for decades, only to see things leap ahead in one year, 2012. That was when computing speed and big data caught up with his ideas. He says ‘it’s impossible to predict beyond five years.’
In journalism many news organizations prefer to wait-and-see, long into the innovation curve. A 2017 survey within the World Publishers Association said the single greatest business challenge they face is ‘a failure to innovate.’ This creates parallel questions, not only (1) when will new capabilities be possible? but also, (2) when they will be adopted?
In society the big steps forward in AI are being directed to big-ticket items, such as climate change, health care, and public safety. Journalism is a small field in comparison. This may mean advances for journalistic applications will trickle-down over time.
Together, impacts for journalism most likely will result from:
- Spin-off capabilities, as advances in other areas are adapted for use in journalism
- Incremental improvements, as existing tools or platforms evolve
- Economic advantages, as they become impossible to ignore
- Explainer | Is deep learning the same as AI & machine learning?
- Explainer | How do algorithms work?
- AI in the newsroom | 7 challenges for journalism | WEF
- Reports/White papers | Artificial Intelligence: Practice and Implications for Journalism | TOW
- Video panels/talks | The augmented newsroom: How AI will impact the journalism we know | GEN Summit 2018
- AI in the newsroom | AI Applications at New York Times, Reuters, and Other Media Giants | EMERJ