It sounds like the start of a bad joke: How can you tell the difference between what a human wrote and what a robot wrote?

The punchline isn't particularly funny: You might not be able to.

Here's a snippet of a story that The Washington Post published in November 2016, on a close game between two local high school football teams:

Keeney seemed to be at a loss until the game's final minutes. Coming off four consecutive 300-plus-yard performances, Keeney completed just 10 of 20 pass attempts for 125 yards, two interceptions and one touchdown Friday. [The Washington Post]

And here's a sample from another high school football game write-up that the Post published less than a year later, in September 2017:

The game began with a scoreless first quarter.

In the second quarter, The Patriots' Paul Dalzell was the first to put points on the board with a two-yard touchdown reception off a pass from quarterback William Porter. [The Washington Post]

The first section is written by Nick Eilerson, a high school sports reporter for The Washington Post. The second one is the work of Heliograf, the Post's artificial intelligence system. You know, a robot.

Robo-writing systems like this one are rapidly edging their way into journalism. As of September 2017, The Washington Post had relied on its artificial intelligence to generate an astounding 850 pieces of content over the previous year. The robo-writer made its debut by auto-publishing reports on the Rio Olympics. Many of Heliograf's stories were on D.C.-area high school football games, while others were tweets or Election Day reporting on congressional and gubernatorial races.

The Washington Post certainly isn't the only news outlet taking advantage of artificial intelligence. The Associated Press has relied on robo-writers to generate earnings coverage and niche sports stories, while USA Today has turned to video software to make short videos complete with narration by a synthesized voice, Digiday reports. Last summer, Google gave a British news agency an $805,000 grant to develop software capable of creating more than 30,000 local news stories each month.

Journalism isn't the only industry turning to robo-writers. Wall Street firms have turned to Quill, the natural language generation platform of the company Narrative Science, for financial reports. In 2014, USAA, a company that provides financial products to former and current U.S. military members, signed a software licensing agreement with Narrative Science to create financial advice articles for its millions of members. Other Narrative Science customers include PricewaterhouseCooper, Groupon, Deloitte, Credit Suisse, and MasterCard Worldwide.

So, how does a robot write stories? First they need humans to show them the ropes and point them in the right direction. After all, robo-writers aren't picking up their pens to opine on President Trump's latest Twitter outburst or to make sense of his most recent slew of firings. Rather, robo-writers tend to work on content that relies heavily on data and that's somewhat formulaic in structure, such as a quick report on the results of an election or the outcome of a high school football game. Though the specifics vary from game to game, the general gist of the story — a winner and a loser, a touchdown and a fumble — remain consistent.

To get robo-writers to produce these stories, humans must first feed them the necessary data points or direct them to a data source that will. For instance, for its election coverage, The Washington Post connected Heliograf to VoteSmart.org, a nonprofit research organization that collects information on candidates.

To get personalized narratives, humans must also provide examples of story structure to the artificial intelligence, which can be done by feeding the artificial intelligence system pieces of similar content previously written by humans or templates of the types of articles it will write. Editors can introduce the artificial intelligence system to key phrases that can be used in different scenarios, so the robo-writing system knows whether to say a candidate won in a landslide or the race is too close to call. It's also possible to teach robo-writers how to identify what's worth emphasizing in the story, so the story doesn't dedicate equal amounts of space to a "scoreless first quarter" and a game-winning touchdown in the final quarter.

Once the algorithm is developed, it can churn out articles at a far faster speed than humans can, Wired reports:

In November 2012, it took four employees 25 hours to compile and post just a fraction of the election results manually. In November 2016, Heliograf created more than 500 articles, with little human intervention, that drew more than 500,000 clicks. [Wired]

Other clear benefits are the possibilities of expanding audience size and improving economics, simply by producing more articles more quickly. According to the Content Marketing Institute, producing one story type on Quill costs roughly $70,000 a year, while three story types would cost $175,000. Wordsmith, which generates more than 1.5 billion pieces of content a year for clients like Microsoft and Allstate, starts at $2,000 a month, with additional set-up costs for each new story's data points. That's a heck of a lot cheaper than a team of writers.

While it might seem like robots are coming for reporters' jobs, in many cases it's been a symbiotic relationship. The Associated Press has said robo-writing systems have "freed up 20 percent of reporters' time spent covering corporate earnings" and decreased the number of errors in stories even as production has rapidly ramped up, DigiDay reports.

But as human writers continue to teach robo-writers the tricks of the trade, it's hard not to wonder if the student will someday become the master.

Editor's note: A previous version of this article misstated elements of Wordsmith's business model. It has been revised. We regret the error.