MIT Developing Fake News Detection AI

Researchers at MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL) has used one of the world’s largest fact-checking data sets in order to create more developed automated systems that could detect fake news.

This was built on a study that MIT CSAIL has conducted last year, wherein they have produced an AI system that could identify whether a source is accurate or not. 

The researchers’ first paper features a framework based on OpenAI’s GPT-2, an AI model that aims to corrupt the meaning behind the human written text before feeding it to a fake news detector. 

However, the fake news detector couldn’t differentiate fake from real text if they’re both machine-generated.

In their second paper, the team sourced Fact Extraction and VERification (FEVER), a collection of false statements that were cross-checked with evidence from Wikipedia articles in order to develop a best-in-class fact-checking algorithm. 

Despite their efforts, the researchers have noted that FEVER contains bias that could cause errors in machine learning models if it were not addressed immediately.

The team fixed the issue by engineering an entirely new algorithm. 

They trained it using a de-biased dataset, making it outperform previous fact-checking AI across all metrics.

CSAIL hopes that combining fact-checking into defenses that already existed will make models more robust against attacks.

In the future, they aim to develop new algorithms and construct data sets that cover more types of misinformation to further improve these existing models.


© Fourth Estate® — All Rights Reserved.
This material may not be published, broadcast, rewritten or redistributed.