Communications of the ACM
Machine Learning Sniffs Out Its Own Machine-Written Propaganda
Researchers at the Allen Institute and the University of Washington's Paul Allen School of Computer Science and Engineering created a natural language processing algorithm that generates, as well as detects, fake articles.
Credit: ZDnet
Researchers at the Allen Institute and the University of Washington's Paul Allen School of Computer Science and Engineering have modified a neural network to create a natural language processing algorithm that generates, as well as detects, convincing fake articles.
The researchers tweaked OpenAI's popular GPT-2 network to produce the "Grover" program, which serves as both a fake-news "generator," and a "discriminator" to identify that false content.
Grover produces disinformation after being fed a massive volume of curated human-written online news texts, supporting a language model that the network utilizes to create its own texts.
The discriminator can identify Grover's fake text because it knows the generator's word-assembling "decoder" component chooses the most likely word combinations in a specific pattern.
The researchers said innovations like Grover offer an "exciting opportunity for defense against neural fake news," as "[t]he best models for generating neural disinformation are also the best models at detecting it."
From ZDNet
View Full Article
Abstracts Copyright © 2019 SmithBucklin, Washington, DC, USA
No entries found