TECH NEWS – While Western artificial intelligences are not very hardwired to uphold core values, this is not the case in China.

Chinese tech companies have also created chatbots. But they also have government oversight: The Financial Times reported that the companies have been subjected to rigorous vetting to ensure that their AI conforms to core socialist values. China’s Cyberspace Administration Center (CAC), for example, is reviewing the AI models of ByteDance (TikTok) and AliBaba (AliExpress…) to see if they comply with the country’s censorship rules.

According to the Financial Times, more cybersecurity officials are showing up at AI companies to interrogate employees about large language models (LLMs), and many politically sensitive topics are also being raised to make sure they actually have the technology they want. The Tiananmen Square massacre, parody memes about Chinese leader Xi Jinping, or anything that “undermines national unity” and “subverts state power.”

AI bots are not easy to defeat, and the CAC does not always give a clear explanation of why a chatbot has failed. The authorities don’t want AI to talk much about politics, but only five percent of queries on the subject can be rejected, making it even more complicated. So some companies have added a layer to their LLM that replaces sensitive responses in real time, and others have simply applied a ban to queries related to Xi.

Fudan University created the security layer. ByteDance leads in “compliance rate” (66.4%). Baidu (the local Google replacement) and AliBaba lag far behind (31.9% and 23.9%, respectively), although this is still much better than OpenAI, which scored only 7.1%. This must be because it is a US technology, not a domestic one.

Still, it’s scary that the country is dealing with this when it has a million other problems.

Source: PCGamer, Financial Times