The model is said to improve fact-checking systems worldwide, mitigating social biases that otherwise come as part of AI-powered technologies
(Subscribe to our Today’s Cache newsletter for a quick snapshot of top 5 tech stories. Click here to subscribe for free.)
A number of artificial intelligence-powered tools today help spot grammatical and factual errors in online books, webpages and news articles. In June 2020, Facebook said it would display warning labels if users chose to share COVID-19 articles older than 3 months on the platform.
Now, researchers at the Massachusetts Institute of Technology (MIT) have devised a machine-learning (ML) model that will monitor updates to news articles and suggest edits to irrelevant and unverified information. It uses deep learning to verify edits and updates related texts, the team noted in a study titled ‘Get Your Vitamin C! Robust Fact Verification for Contrastive Evidence’.
They examined edits to popular Wikipedia pages. The website has an open design and welcomes over 6,000 edits per hour, making manual fact-checking cumbersome, the team noted.
Also Read | Google extends fact-checking to image search
The team then collected 200 million revisions to popular pages, including COVID-19 and famous personalities, and assigned human annotators to filter and edit outdated information. The ML model imitated the task of human annotators and detected almost 85% revisions with factual change.
The team also developed a model to automatically revise texts and potentially suggest edits to related articles. Changes made by the model were rated as accurate as those made by humans, they noted.
The model is said to improve fact-checking systems worldwide, mitigating social biases that otherwise come as part of AI-powered technologies. It could also save time and effort as compared to manual fact-checking.