Skip to Content

A new tool uses AI to spot text written by AI

July 26, 2019
Text from Infinite Jest analyzed by the Harvard-MIT-IBM tool.
Text from Infinite Jest analyzed by the Harvard-MIT-IBM tool.
Text from Infinite Jest analyzed by the Harvard-MIT-IBM tool.Harvard University

AI algorithms can generate text convincing enough to fool the average human—potentially providing a way to mass-produce fake news, bogus reviews, and phony social accounts. Thankfully, AI can now be used to identify fake text, too.

The news: Researchers from Harvard University and the MIT-IBM Watson AI Lab have developed a new tool for spotting text that has been generated using AI. Called the Giant Language Model Test Room (GLTR), it exploits the fact that AI text generators rely on statistical patterns in text, as opposed to the actual meaning of words and sentences. In other words, the tool can tell if the words you’re reading seem too predictable to have been written by a human hand.

The context: Misinformation is increasingly being automated, and the technology required to generate fake text and imagery is advancing fast. AI-powered tools such as this may become valuable weapons in the fight to catch fake news, deepfakes, and twitter bots.

Faking it: Researchers at OpenAI recently demonstrated an algorithm capable of dreaming up surprisingly realistic passages. They fed huge amounts of text into a large machine-learning model, which learned to pick up statistical patterns in those words. The Harvard team developed their tool using a version of the OpenAI code that was released publicly. 

How predictable: GLTR highlights words that are statistically likely to appear after the preceding word in the text. As shown in the passage above (from Infinite Jest), the most predictable words are green; less predictable are yellow and red; and least predictable are purple. When tested on snippets of text written by OpenAI’s algorithm, it finds a lot of predictability. Genuine news articles and scientific abstracts contain more surprises.

Mind and machine: The researchers behind GLTR carried out another experiment as well. They asked Harvard students to identify AI-generated text—first without the tool, and then with the help of its highlighting. The students were able to spot only half of all fakes on their own, but 72% when given the tool. “Our goal is to create human and AI collaboration systems,” says Sebastian Gehrmann, a PhD student involved in the work.

If you're interested, you can try it out for yourself.

Deep Dive

Artificial intelligence

chasm concept
chasm concept

Artificial intelligence is creating a new colonial world order

An MIT Technology Review series investigates how AI is enriching a powerful few by dispossessing communities that have been dispossessed before.

open sourcing language models concept
open sourcing language models concept

Meta has built a massive new language AI—and it’s giving it away for free

Facebook’s parent company is inviting researchers to pore over and pick apart the flaws in its version of GPT-3

spaceman on a horse generated by DALL-E
spaceman on a horse generated by DALL-E

This horse-riding astronaut is a milestone in AI’s journey to make sense of the world

OpenAI’s latest picture-making AI is amazing—but raises questions about what we mean by intelligence.

labor exploitation concept
labor exploitation concept

How the AI industry profits from catastrophe

As the demand for data labeling exploded, an economic catastrophe turned Venezuela into ground zero for a new model of labor exploitation.

Stay connected

Illustration by Rose WongIllustration by Rose Wong

Get the latest updates from
MIT Technology Review

Discover special offers, top stories, upcoming events, and more.

Thank you for submitting your email!

Explore more newsletters

It looks like something went wrong.

We’re having trouble saving your preferences. Try refreshing this page and updating them one more time. If you continue to get this message, reach out to us at with a list of newsletters you’d like to receive.