Jason Kottke: How does a computer program compare to the Web's best link blogger?

Will these machines steal your job?
Sept. 27 2011 7:20 PM

Will Robots Steal Your Job?

We wrote a computer program that replicates Jason Kottke. How does Robottke compare to the Web's best link blogger?

(Continued from Page 1)

Kottke is right. Robottke can't write, it can't do headlines, and it can't find the best excerpt to quote. Well, it can't, yet. We'd love to get people to help us improve Robottke; if you're a coder who's got an interest in artificial intelligence and the media, contact Chris Wilson for Robottke's source code. (FYI, the source is in Python.)

We need your help, because the flesh-and-blood Kottke is slowing down. "I started building Stellar because I was basically burned out on blogging," he says. In 2006 and 2007, he would spend nearly all of his waking hours in search of links, and he always found himself frustrated. "Then, over a period of 6 to 8 months, I suddenly started getting a lot more efficient," Kottke says. He attributes the change to experience—"maybe I hit 10,000 hours"—and to new tools, like Twitter. "I found that I could do the site in much less time—now I'm spending only a couple of hours a day on Kottke.org," he says. "Now there are fewer items a day, and I'm writing a little bit less about each thing, and I don't often take the time to really dig into something that's a little more interesting the way I used to."

When I asked if he plans to wind down Kottke.org, he answered cryptically. "I would probably stop working on it as a job, but I don't think I would ever stop blogging," he said. "I feel a certain responsibility to people who are reading, and it is something that I like doing, but I don't like doing it full time, all day anymore."


That's the advantage of Robottke: It will never get burned out.


In this series, I've been looking at machines that can replace high-skilled human professionals. I've already covered my dad's profession, pharmacy, and my wife's medical specialty, pathology. Now, it's time to look in the mirror. As a writer, I like to think of myself as having uniquely human skills. I write columns about stuff that human readers care about, and in doing so I try to elicit human emotions—joy, fascination, fury. Machines can't yet mimic this sort of creativity. But as I surveyed efforts to automate journalism—and as we learned from our work on Robottke—I found that my job may not be beyond the capacities of a robot.

I go through the same routine every weekday morning. First, I launch a half-dozen technology news sites and open up dozens of stories. As I scan each headline and skim a bit of the story, an algorithm executes in my head: Is this interesting enough to write about? If not, skip. When I do encounter a story that appeals to me, my brain launches another script: Can I come up with a unique take? Will people care about this? My process isn't conscious, and it's not always so systematic. Often I'll get distracted by email or the phone, and if I've been up all night tending to the baby, I'm likely to spend more time goofing off on Facebook than looking for something to write about. But these are human failings; if I were a robot, I'd follow the algorithm perfectly—and likely come up with better story ideas as a result.

This is the theory behind current efforts to automate journalism: Whether we admit it or not, many writers loosely follow a script when they work. How do you write a wire-service-style news story? You start by noting the most important thing that happened—what the jury found, who got killed, which team won. Then you get into the particulars. Structural rules even dominate more high-minded genres. The former New Yorker writer Dan Baum once passed along this bit of advice from his editor at the vaunted magazine: "You can use any narrative structure you like. … Just know that when I get it, I'm going to take it apart and make it all chronological." So, here's the algorithm to write a New Yorker story: 1) Say what happened first. 2) Say what happened next. 3) If there's more stuff to say, repeat 2); otherwise, end.

At the moment, no computer on earth can write anything like a New Yorker story. But machines can do a serviceable job with more rote pieces. Look at this excerpt from a site called FriscoFan.com, which posts a game recap every time the San Francisco Giants play:

            Giants Batter Rockies 3-1