The use of fake news stories for political disinformation has become a major concern for governments around the world in the wake of the 2016 U.S. presidential election. The Federal Bureau of Investigation concluded Russia used false news reports, spread through social media, to try to sway voters. Writing these stories still needed someone to sit behind a keyboard.
Now OpenAI, a non-profit artificial intelligence research group in San Francisco, has unveiled a machine learning algorithm that can generate coherent text, including fake news articles, after being given just a small sample to build on. The algorithm can be tuned to imitate the writing style of the sample text.
Here’s an example of the system at work that OpenAI provided. The system was only given the first two sentences, which were written by a human:
And this is the rest of the news article that the AI wrote, with no human guidance:
Because of its potential to be misused for creating disinformation, OpenAI said it had chosen not to make the most powerful versions of its algorithm available for public use nor was it publishing the source code and training data used to create the software.