Artificial neural networks are already being used to teach self-driving cars to “see” and train Facebook’s new personal assistant, but they are also being deployed for a less ambitious purpose: automatically generating the seductive and carefully calibrated news headlines that drive traffic to some of the world’s most popular websites.
Lars Eidnes, a Norwegian developer, created software that uses Recurrent Neural Networks (RNN)—a form of “deep learning” that has been able to demonstrate surprisingly human-like abilities—to write new clickbait headlines, after training it with several million articles from BuzzFeed, Gawker, Jezebel, the Huffington Post, and Upworthy. Eidnes explained on his blog:
We can show an RNN a bunch of sentences, and get it to predict the next word, given the previous words. So, given a string of words like “Which Disney Character Are __”, we want the network to produce a reasonable guess like “You”, rather than, say, “Spreadsheet”. If this model can learn to predict the next word with some accuracy, we get a language model that tells us something about the texts we trained it on.
The resulting model can then take a seed (“Taylor Swift”) and use it to generate new headlines. ”It surprised me how good these headlines turned out,” Eidnes wrote. “Most of them are grammatically correct, and a lot of them even make sense.”
Here are a few samples:
- John McCain Warns Supreme Court To Stand Up For Birth Control Reform
- Biden Responds To Hillary Clinton’s Speech
- Here’s What A Boy Is Really Doing To Women In Prison Is Amazing
- Why Are The Kids On The Golf Team Changing The World?
- Taylor Swift Becomes New Face Of Victim Of Peace Talks
- WATCH : Mitt Romney’s New Book
- Kim Kardashian Is Married With A Baby In New Mexico
Eidnes didn’t stop there: He went on to create an entire auto-generated news site called Click-o-Tron, which pairs the headlines with photos and short articles, also assembled by the neural network, that bears more than a slight resemblance to the sites that inspired it.
Google is using headlines from newspapers like the Daily Mail to teach its neural networks to parse language. But Eidnes suggested that the technology also “gives us an infinite source of useless journalism, available at no cost. If I remember correctly from economics class, this should drive the market value of useless journalism down to zero, forcing other producers of useless journalism to produce something else.”
As a recent story by Bloomberg suggested that many of the so-called “clicks” that drive the online news business are actually fraudulent software “bots,” perhaps the future will consist entirely of computer-generated “news” consumed by non-human “readers,” with actual humans completely superfluous to requirements. Bet you didn’t see that coming.