Quite frankly, they’re doing a pretty decent job at it.
Business reporters get stuck with writing about quarterly earnings, and that demands a daunting combination of speed and accuracy. But as The Verge notes, it’s painfully necessary.
So the Associated Press partnered with Automated Insights (AI) and solved this problem. Using the company’s Wordsmith platform, up to 2,000 stories per second can be generated.
Welcome to the future
Wordsmith generates millions of articles every week. Other companies that have partnered with AI include Allstate, Comcast, and Yahoo, which uses the platform for its fantasy football reports, James Kotecki, AI’s public relations manager told The Verge.
So how does a robo-writer really work?
For the robots at Narrative Science, the first order of the day is to swallow a batch of data. This idea sprung up in domains that are stuffed with data — like weather domains, where there were plenty of natural language generation systems to mine; now this is also being applied to sports and finance. Data is pretty standardized in these domains. Although this medium is in its’ infancy, it may well provide a wealth of data for producing automated stories. That data needs to be clean and comprehensive, Diakopolous writes.
Once the algorithm reads the data, the next step is to glean newsworthy or interesting features from the data. Now the algorithm tries to figure out critical aspects of an event, and it has newsworthiness criteria built into its statistics. It seeks surprising statistical deviations — any sort of big swings, changes in a value, minimums, maximums, or anything outlying. “Any feature the value of which deviates significantly from prior expectation, whether the source of that expectation is due to a local computation or from an external source, is interesting by virtue of that deviation from expectation,” according to the patent for Narrative Science. In a baseball game, for instance, the algorithm computes “win probability” after every play, Diakopoulos writes.
After a few interesting features have been identified, it’s time to look for angles for the story, and these are selected from a pre-authored library. Angles are explanatory, adding coherence to the overall story. What they really are is patterns of events, entities, circumstances, and their features. For a sports story, for instance the angle might be “heroic individual performance,” “strong team effort,” or “came out of a slump.” Some of the angles are triggered according to certain features derived from the previous step, and each angle is given an importance value ranging from 1 to 10, and that is used to rank that particular angle against all of the other proposed angles, Diakopoulos writes.
Once the angles have been determined and placed in order, they are linked to specific story points that connect to individual pieces of data — the names of players or perhaps something with numeric values, like a score, for instance. Now story points can be chosen and personal interests, such as home team players, can be pointed out. Then the narrative can look to internet databases to furnish a story with factual content, such as a player’s hometown, a quote or a photo of the player.
The last step on this automated journey is when the robo-journalist traverses natural language generation. For Narrative Science, this is done by checking the angle and story point representations, using phrasal generation routines that generate and splice together the text.
That’s how a robot-generated story is born.
Before implementing the program, the AP estimates it was writing up quarterly earnings coverage for around 300 companies, The Verge reports. Now it automates some 3,000 reports each quarter. Out of that number, the human touch will be felt in 120 of these stories, either through updating the original story or by doing a follow-up piece.
Will journalists become extinct?
As an assistant business editor at the AP, Philana Patterson was given the job of implementing the system, and at least some staff members were skeptical, she told The Verge.
“I wouldn’t expect a good journalist to not be skeptical,” she said.
When the program first started in July, all automated stories were imbued with a human touch. Errors were logged and sent to AI to fix the mistakes. In October, the program was fully automated and stories were published without human intervention, and no jobs were lost due to this new service, the AP and AI told The Verge. On a potentially ominous note, they also said the automated system is logging in fewer errors than those produced by people in years past.
Noted journalist and social activist Barbara Ehrenreich is cautious about this new technology. Ehrenreich, the author of “Nickel and Dimed: On Not Getting By In America,” in her review of “Rise of The Robots” by Martin Ford, writes in The New York Times, that “tasks that would seem to require a distinctively human capacity for nuance are increasingly assigned to algorithms, like the ones currently being introduced to grade essays on college exams.” She’s worried about computer programs that can write publishable articles and notes that an expert quoted by Wired magazine predicts that within the next decade 90 percent of news articles will be computer-generated.
Journalists are safe…for now
There’s not going to be any “slaughter of all the journalists,” Ford told TheMediaOnline. Instead, changes are likely to be subtle.
He doesn’t believe huge numbers of journos are going to get canned; what’s likelier to happen is that fewer journalists may be hired in the future, especially at the entry level.
“If you want to have a career in journalism and you’re just graduating from school, the place you might start, your first assignment may be one of those routine formulaic things like sports stories and corporate earnings reports and maybe even obituaries or something like that, you know, those kinds of things where there’s not a lot of creativity to it … that’s been the way that journalists have learned the ropes from the beginning,” he notes.
For journalists in the UK, the risk of losing a job to automated programs appears pretty low, according to statistics compiled by researchers from Oxford University, IFLScience reports.
To highlight the results of the statistics, the BBC cobbled together an infographic that detailed the risks robots pose to different professions, including journalists. The infographic listed 366 jobs.
The infographic showed that telemarketers, typists, and legal secretaries faced the most risk of losing their jobs to artificial intelligence in the near future. Hotel owners and shop managers were among the safest. Out of the jobs listed, journalists came in at number 285, with only an eight percent risk of losing their jobs to automation.
How was the data gathered?
The researchers assessed nine crucial skills in each job. These skills included social perceptiveness, negotiation, and persuasion. Jobs requiring more “human-like” skills were determined less likely to be automated. In total, however, about 35 percent of jobs were found to be at risk of computerization in the next 20 years – and it isn’t known what other jobs may spring up as a result.
There’s also plenty of reasons to be positive about automation.
In his article for The Daily Intelligencer, Kevin Roose writes that he thinks automated reporting is good news for journalists.
“Not only am I not scared of losing my job to a piece of software, I think the introduction of automated reporting is the best thing to happen to journalists in a long time,” he writes.
And, for right now, we humans still have the edge over the bots, talent-wise. AI’s software can only produce certain types of news stories — it’s well-suited for generating short articles that rely on structured data for input, and the output of these stories follows a regular pattern, Roose notes. The software’s algorithms sift data feeds for facts and important trends. Then it combines this with historical data and other types of contextual information to write narrative sentences.
Victoria Nicks, who holds a Master’s degree in Information Technology, is a tech consultant and the CEO of Decoded Everything. She says it makes sense to use programs like AI’s.
“Google’s using similar technology to pull the most relevant answer to your question when you Google a question,” she told Digital Journal. “And there are artificial intelligence algorithms that can identify skin cancer lesions more accurately than experienced doctors.”
“If a piece of software can use a formula to compile a brief article, then it works for me.”
She believes there will always be room for journalists.
“Real-life experts will always be in demand for more detailed insights, so I don’t see that this takes away from anyone any more than ATMs take away from bank tellers.”
She also says she thinks this can be reasonable and ethical, as long as there’s an editor checking the articles before they go live.
The ethics of robo-journalism
There are still questions regarding the ethics of the software, however. It’s something that should be carefully considered, Associated Press standards editor Tom Kent told the University of Wisconsin Center for Journalism Ethics. “We take our ethics very seriously.”
Among the concerns: There can be copyright issues, and this mainly depends on how data is collected, Kent said. The data itself might be copyrighted.
“Just because the information you scrape off the Internet maybe accurate, it doesn’t necessarily mean that you have the right to integrate it into the automated stories that you’re creating – at least without credit and permission.”
Aside from collecting data, the way in which algorithms structure the information also requires ethical consideration. For an article to sound natural, the algorithm has to know the language, BBC reporter Stephen Becket wrote in an article in September.
“Each type of story, from finance to sport, has its own vocabulary and style,” he noted.
Language is something that must be programmed into the algorithm at human discretion, the Center notes.
This is why Kent believes that stories written by robots aren’t necessarily more objective than those that are produced by humans, and therefore, they are subject to the same objectivity considerations.
“I think the most pressing ethical concern is teaching algorithms how to assess data and how to organize it for the human eye and the human mind,” he said. “If you’re creating a series of financial reports, you might program the algorithm to lead with earnings per share. You might program it to lead with total sales or lead with net income. But all of those decisions are subject – as any journalistic decision is – to criticism.”
News judgement and organization are ethical questions, and as such, this carries over from traditional reporting to robot journalism, and Kent suggests we should combat them in the same way.
“Everyone has a different idea about what fair reporting is,” he noted. “The important thing is that you devote to your news decisions on automated news the same amount of effort you devote to your ethics and objectivity decisions at any other kind of news.”
The advent of robo-journalism opens up a new realm of possibilities, and it will keep journalists on their toes, but it has the very real possibility of making our jobs easier.
Welcome to the future.