Artificial intelligence recognizes fake news better than humans, a university study reveals. It works well in 76% of cases
Artificial intelligence recognizes the fake news better than people, just one algorithm to overcome the human capacities of immediately recognize false news. Researchers from the University of Michigan and the University of Amsterdam worked on developing an artificial intelligence tool based onmachine learning for detect false news the famous fake news. When they started development, there was not enough data available to train the algorithm, but they made the only rational thing possible: they collected hundreds of articles and fake news and fed them to artificial intelligence.
How the algorithm that recognizes fake news works better than humans works
The algorithm, developed by researchers from the University of Michigan and the University of Amsterdam, uses thenatural language processing (NLP) to look for specific models or linguistic indications that indicate that a particular article is a fake news, false news. This artificial intelligence is different from an already diffused algorithm of fact checking that checks the information contained in an article by comparing it with other sources to see if the article examined contains inconsistent information. The new solution based on artificial intelligence, entirely based on themachine learning, could completely automate the tracking process.
The implementation of a NLP algorithm to analyze the structure of sentences and perfect the recognition of keywords is not a new application for intelligence. However do it for detect fake news better than people, a real innovation.
Examples of Fake news to study
The problem is that there are not enough data. You cannot simply download the data from the internet and tell an algorithm what to do: the machines need rules and examples. The commonly available datasets for this type of artificial intelligence training include a set of Buzzfeed data, which was used to train algorithms to detect fake news on Facebook. Other datasets focus mainly on training an artificial intelligence on satirical content, such as those of The Onion – unfortunately, this method tends to turn an algorithm into a satire detector.
How to teach AI to recognize fake news
Researchers from the two universities had to decide first what the false news is. For this reason, they turned to the requirements for a fake news corpus research developed by a team of researchers from the University of Ontario.
These nine requirements basically indicate that an algorithm for detect fake news must be able to recognize the news that is not false, verifying the truth and taking into account factors such as the development of news and linguistic and cultural interpretations.
The fake news have some features in common in all their variations: news that is intentionally false; hoaxes (created with the intention of becoming viral on social media) or humorous or satire articles.
How to teach to recognize fake news
Researchers from the University of Michigan and Amsterdam created their own data sets by asking Amazon Mechanical Turk workers to reinterpret 500 real news as if they were fake news. Study participants were asked to imitate the journalistic style of the original article, but inflated the facts and information to ensure that the result was clearly false.
The team provided both the fake news and the real news to the algorithm and learned to distinguish the two. Once trained, the team provided news directly from a data set containing it both real and false news directly from the Web and did better than humans in the understand what were the fake news.
THEnot perfect algorithm, misses 24% of the time, but humans have more mistakes: at least 30% of the time, so the point on fake news recognition goes to machines.
One of the project's researchers, Rada Mihalcea, told Michigan News
You can imagine many applications for this algorithm to be put at the beginning or at the end of a news story or on social media sites. It could provide users with an estimate of the reliability of individual news or of an entire site. It could be a first line of defense against a site's false news, reporting suspicious stories for further review. A 76% success rate leaves a fairly large margin of error, but it can still provide valuable information when used together with humans.
Lera of fake news
The world in the midst of what historians will almost certainly call "the era of fake news". likely that this problem cannot be solved simply by relying on the authors forcing them to never lie by even forcing users to news to always check everything they read on multiple sources.
A more elegant solution involves the implementation of protocols and protections before publication and at the platform level, but until this process is completely automated we will continue to see the flow of false newsfrom various sources.
At least, according to the white paper of the research team, we now have a new tool for detecting false news that outperforms human capabilities in terms of performance.
Google News: here is the new version
YouTube against fake news focuses on authoritative videos and invests
WhatsApp marks the messages forwarded to fight fake news