WORD REACHES US of a breakthrough in the ability of artificial intelligence (AI) to filter information online with a degree of accuracy previously unachieved.
According to a paper in the Journal of Chinese Computer Systems, entitled Research on Sensitive Information Recognition of Internet News, researchers at Shenyang Ligong University and the Chinese Academy of Sciences claim their new AI is 91% accurate in blocking ‘harmful information’.
In addition, unlike previous AI, which has achieved 80% accuracy rates, it does not need to be trained by humans but can be self-taught through deep machine learning. Internet companies employ armies of humans to check and censor content manually, augmentations to older software systems that rely on keywording and are only 70% accurate.
The researchers also say their AI can keep up with the expanding volume of content on the internet and the rapid evolution of online language used to get around content filters. The use of homonyms to avoid sensitive words is a particular challenge for censors of Chinese text.
The AI is partially based on software Google developed more than five years ago to let it understand the context in which users of its search engine were using search terms.
While there is a universal need to censor online content that promotes criminal activity, terrorism, pornography and human trafficking, China also has a particular demand for blocking political content deemed sensitive.
The AI advance should cut the cost of censorship and opens up a potentially lucrative export market for automated censorship systems, decreasing the economic price of political repression for authoritarian governments.
While no censorship system is ever likely to be 100% effective, better-than-90% effectiveness would be good enough to keep mass political movements from taking hold.