AI to Fight Fake News
Here are five important AI stories from the week.
With the major breakthroughs in the field of natural language processing over the past 18 months, it is much easier than ever before to generate fake—yet seemingly real—news. To study and detect fake news, the Allen Institute developed a model that both generates fake news and detects whether an article was written by its AI—called Grover—or a human. I highly recommend you check out this demo.
Facebook AI Research (FAIR) discovered a method to convert text to speech and produce near-realistic human-like audio. The Facebook model trained on Bill Gates’s voice so the conversion emulates his voice. This is a remarkable breakthrough but raises concerns about the potential explosion of fake audio content in the coming years. Fake news is already a major problem today. To check out the Bill Gates samples, please see the article.
Given the advances in computer vision over the past six years, it is now possible not only to identify faces in video but also automatically notify authorities based on suspicious human actions in videos such as entering a forbidden area, loitering, urinating in public, etc. Powered by AI, surveillance will be ubiquitous in society, raising concerns of a total surveillance state similar to what is happening in China today.
Amazon’s success in online shopping is partly due to its incredibly good recommendations engine. Now, Amazon is releasing its recommendation system—called Amazon Personalize—to the masses, allowing customers of Amazon Web Services to use the recommendation system in their own applications. This will lead to more personalized product and content recommendations, search results, and marketing campaigns for more businesses.
In addition to fake news and fake audio, fake images and fake video (e.g., a fake video featuring Mark Zuckerberg giving a sinister speech) are becoming commonplace. To combat fake content, Adobe—the creator of Photoshop—developed an AI to detect edited media. In its internal tests, Adobe’s AI spotted 99 percent of edited faces. In this age of disinformation, firms are developing tools to fight back.
More Stories Worth Reading and Watching…