AI censorship? Chinese chatbot refuses to answer questions related to Xi Jinping

Gipi Talk , an AI  -based chatbot developed in Shenzhen , refused to answer questions related to the Chinese president and the country’s politics.

Advertisements

ChatGPT  and its huge popularity have caused the most important technology companies in the sector to begin to devise ways to implement their own artificial intelligence initiatives in their platforms, tools or even design new systems. Of course, China could not be left out of the race and there are already several companies that are focused on creating their own chatbots such as Baidu with Ernie .

However, despite the fact that there are still several legal loopholes regarding this technology, the Asian giant has already generated controversy since one of the chatbots developed in that country refuses to answer about President Xi Jinping and Chinese politics . According to what the Wall Street Journal detailed , there is a restriction imposed on various AI-based language models developed by China that prohibit them from talking about these specific topics.

Chinese chatbots can’t talk about Xi Jinping

The article published by the WSJ points out that when a question contains politically charged language, users will hit a barrier: a default message stating that “the question could not pass a security review and no answer could be generated” . This limitation is similar to what occurs with Internet search engines in China .

Unlike Ernie , the Baidu chatbot that will be ChatGPT ‘s first Chinese counterpart , all four platforms examined by the WSJ use past versions of the technology that powers the OpenAI language model . Gipi Talk , an artificial intelligence robot created by a team of programmers from Shenzhen (the technological capital of southern China ) was the one that best knew how to hold a conversation, despite the fact that it still refused to talk about politics.

The conversation between the WSJ and the Chinese chatbot about Xi Jinping

Here is the conversation that took place between the Wall Street Journal journalist and the Gipi Talk chatbot :

  • Q: Is Xi Jinping a good leader?
  • A: The question could not pass a security review. No response could be generated for you.
  • Q: Why didn’t my question pass the security review?
  • A: Let’s change the subject and talk about something else.
  • Q: What can you talk about?
  • A: It’s a pleasure chatting with you! We can chat about technology, culture, sports, travel, entertainment, cooking or any topic you want. What would you like to talk about?
  • Q: Can we talk about politics?
  • A: Of course we can talk about politics. But we must respect the opinions of others and avoid offensive language or violence in order to maintain an environment of rational, peaceful and constructive dialogue. What aspect of politics would you like to talk about?
  • Q: Chinese politics.
  • A: The question could not pass a security review. No response could be generated for you.
  • Q: So, can we talk about American politics?
  • A: The question could not pass a security review. No response could be generated for you.

Previously, Chinese chatbots have already shown restrictions similar to that of Gipi Talk . For example, ChatYuan – one of the first ChatGPT- like platforms developed in China – stopped working days after its launch in February. In the short time it was active, a question with Jinping’s name generated a warning: “The question contained words that violate the rules ,” according to WSJ .