By JACK HOUGH
Journalists may not> have to report on the loss of jobs to technology for much longer. Computers may soon handle that job on their own.
Software programs can already turn data into passable sports reports in seconds. "An outstanding effort by Willie Argo carried the Illini to an 11-5 victory over the Nittany Lions on Saturday at Medlar Field," wrote a program called Stats Monkey last April. Stats Monkey's Northwestern University creators say it can handle any event such as an earnings report or market update that produces heaps of quantitative facts. It can "write" using different tones and points of view, and even add quotes so long as they're part of the data feed.
Sorry, editors: Stats Monkey also handles headlines.
That got me thinking about the Stock Screen columns I write for SmartMoney (whose name is already too similar to Stats Monkey for my comfort). I've long advised readers that computerized screens are a fine way to begin a search for attractive shares, because they tap into a much larger universe of candidates than humans can possibly consider on their own. They also bypass human prejudice to point out unpopular but worthy stocks investors might not otherwise have considered. But computers just guide the research. They can't pick stocks for you, I tell readers.
That's not exactly true. Trading algorithms already account for more than half of the shares bought and sold in an average day. Robo-Buffetts, these programs ain't; typical hold times are measured in milliseconds. Buy-and-hold investing requires human judgment to make sense of things like economic trends and manager competence. Doesn't it?
Maybe not. Investors can already buy into mutual funds based on indexes like the FTSE RAFI US 1000, which "picks" its members based on fundamental measures like sales, free cash flow, asset values and dividends. And then there are computer-driven research firms like ValueEngine, whose 10-page company reports are based on a computer model that back-testing has shown to pick winning stocks. ValueEngine covers more than 6,000 companies with a full-time staff of just 12 people, mostly programmers.
"Our lack of human researchers visiting company managers is a strength," says founder Paul Henneman. "We remove the emotion from investing and make it as scientific as possible."
The Motley Fool recently gave me a scare. "Avery Dennison's Dividends Might Not Last Forever," read a headline on the media company's popular investing site, Fool.com, in mid-December. The same headline has run with more than 50 different company names over the past six months, and the text for each story is remarkably similar. "It's obvious that, at least on the surface, there aren't any problems with [company name] generating enough income to support that nice dividend of [percentage]," the stories say of companies whose dividends make up only a modest portion of earnings. There are other fill-in-the-blank stories with titles like "Is [Company Name]'s Stock Cheap By the Numbers?" and "Is [Company Name] the Perfect Stock?"
So have the machines already taken over? No. Brian Richards, managing editor of Fool.com, says some of his writers have been using templates for about six months to extend their coverage, and that he's seen other sites experiment with what he calls template-style content. Fool.com's template stories draw readers on highly trafficked partner sites like Yahoo Finance and contain pitches for Motley Fool's paid newsletters, as well as sponsor advertisements. "The majority of readers realize these are formula stories," says Richards, although "a minority might think they're brand new." Computers don't do the writing. Humans create the templates and tweak the results, and editors review the stories after they're submitted, Richards says.
For the last word on when R2-D2 will take my job, I spoke with Kristian Hammond, one of Stats Monkey's creators and chief technology officer of Narrative Science, an Evanston, Ill., company working to commercialize robo-journalism.
"You have two different skills," he told me. "One is an ability to interpret the data, which is easy. The other is an ability to articulate the process to your readers, which is more difficult."
I felt better than I probably should have about that. But surely computers will be able to handle the articulation part eventually. "Are we talking about 10 years or 100 years?" I asked.
"Three months," Hammond said.
After a quick mental tally of my retirement savings, I asked why he was being so specific.
"I can't give you all the details yet, but the technology is amazing," he said. "A computer will win a Pulitzer Prize in five years."