Advertisement
Artificial intelligence
TechInnovation

Fighting fake news the Chinese way: A peek inside China’s biggest news aggregator

Facebook, Google and Twitter have all come under scrutiny over the degree of responsibility they should bear for misinformation spread by others through their social networks

Reading Time:2 minutes
Why you can trust SCMP
Beijing-based ByteDance Technology, which operates the popular mobile app called Jinri Toutiao, or Today's Headlines, is using artificial intelligence to fight fake news. Photo: Bloomberg
Meng Jing

While Facebook is attempting to stamp out fake news by contracting third-party fact-checkers to call out false information, researchers at ByteDance Technology – operator of the popular mobile app Jinri Toutiao, China’s biggest news aggregator – have drawn inspiration from the age-old Chinese medicinal practise of “fighting poison with poison”.

Beijing-based Toutiao, with 120 million daily active users, is working on artificial intelligence (AI) algorithms to detect fake news and create a bot that is an expert in generating misinformation, said Ma Weiying, head of AI Lab at ByteDance.

Without the fake news writer, researchers may not be able to garner the steady diet of misinformation to feed and “train” the fact-checking algorithms, according to Ma.

Advertisement

He said the idea of using a bot to fight another bot helps improve the AI algorithms used by Toutiao. The approach was similar to how Alphabet’s DeepMind AlphaGo program mastered Go, the ancient Chinese strategy board game, and beat a professional Go player.

Advertisement

London-based DeepMind, a subsidiary of Google owner Alphabet, revealed earlier this year that AlphaGo Zero, the latest evolution of its Go-playing computer program, learned to play in a very short time by playing games against itself.

The previous AlphaGo program had trained with thousands of human amateur and professional Go players.

Advertisement
Select Voice
Choose your listening speed
Get through articles 2x faster
1.25x
250 WPM
Slow
Average
Fast
1.25x