Clickbait is the scourge of our generation of journalism. Skim the digital pages of most media sites and these kind of headlines are most likely to grab your attention:
- X Number of Things You Won’t Believe Happens in A Certain Place!
- X Number of “Facts” That Will Blow Your Mind!
- This Part Sounds Normal, But The Next Part Will Shock You!
- You Won’t Believe What This Person Did/Said!
- The Most Terrifying Thing About _________
- Which ________ Are You? (Quiz)
- What Does Your ___________ Say About You?
We can’t help it. Even after years of visiting sites whose sole source of incoming views come from clickbait titles like one of the ones above, we find ourselves hooked and needing to see what lies beyond. What if this thing will shock me? What did that celebrity do or say that will blow my mind?
With this in mind, Lars Eidnes–a developer from the IT consulting company Itema in Trondheim, Norway–has dedicated some time to see if he could teach a computer to write clickbait headlines.
Using what are called Recurrent Neural Networks–or RNNs–Eidnes could teach a system of computer algorithms to slowly learn how to create headlines without even teaching it what the words mean. By feeding the RNN headlines from over two million articles, gathered from such websites as Buzzfeed, Gawker, Jezebel the Huffington Post and Upworthy–that RNN developed a better idea of how to write newsworthy or clickbait-level headlines. With a little tweaking over time to minimize prediction errors and maximize it’s guessing ability, the RNN went from making some humorous and horrendous headlines to generating pretty solid titles.
These are some examples the RNN produced after only 40,000 titles:
- 2 0 Million 9 0 1 3 Say Hours To Stars The Kids For From Internet
- Adobe ‘ s Saving New Japan
- Real Walk Join Their Back For Plane To French Sarah York
- State 7
- Dr 5 Gameplay : Oscars Strong As The Dead
- Economic Lessons To Actress To Ex – Takes A App
- You ‘ s Schools ‘ : A Improve Story
After a few tweaks and multiple passes, these are some better examples of what it could do:
- John McCain Warns Supreme Court To Stand Up For Birth Control Reform
- The Most Creative Part Of U . S . History
- The Children Of Free Speech
- Romney Camp : ‘ I Think You Are A Bad President ‘
- Why Health Care System Is Still A Winner
- Why Are The Kids On The Golf Team Changing The World ?
- 2 1 Of The Most Life – Changing Food Magazine Moments Of 2 0 1 3
- World ‘ s Most Dangerous Plane
Not all of these are based on accuracy, mind you, and many of them don’t make sense, like the following headlines:
- Chef Ryan Johnson On ” A . K . A . M . C . D . ” : ” ” They Were Just Run From The Late Inspired ”
- A Tour Of The Future Of Hot Dogs In The United States
- This Guy Thinks His Cat Was Drunk For His Five Years , He Gets A Sex Assault At A Home
And that’s not including the headlines from the “good” batch of headlines that were also erroneous but linked together through algorithms and likely possibilities. Considering that this was done by A.I. that had no knowledge of the meaning of English words but learned how to link them together through trial and error, it’s amazing to see how far the system went in terms of coherence and quality.
But what does this mean for the future of journalism? Will there be A.I. writing our clickbait articles for us? RNNs telling us 17 things that will help us lose weight in a week? Computers luring us to taking a quiz that reveals what spirit animal belongs to us?
Is there one writing this article right now?
Source: Lars Eidnes Blog via Popular Mechanics
Feature Image: Penn State