Spotting Machine-Written Text

Creating 3D Models from Flat 2D Images
March 10, 2020
Ambient Computing Expands
March 10, 2020

Spotting Machine-Written Text

It turns out that A.I. can also be used to detect when text was machine generated, even if we humans can’t spot the fake.

In the past year, researchers showed how A.I. could be used to compose text so good that humans couldn’t tell it was machine written. The team at OpenAI demonstrated the many reasons why this was problematic, from mass-generating salacious social media posts and fake reviews to forging documents by world leaders.

It turns out that A.I. can also be used to detect when text was machine generated, even if we humans can’t spot the fake. That’s because an essay written by A.I. tends to rely on statistical patterns in text and doesn’t have much linguistic variation.

Researchers at the MIT-IBM Watson AI Lab and at Harvard University developed the Giant Language Model Test Room (GLTR), which looks for words that are likely to appear in a particular order.

This is good news for those concerned about the spread of automated misinformation campaigns.