Baidu Believes Its AI Chatbot Is Safe For Important Topics

Baidu CEO, Robin Li, said that the company was awaiting approval from the government before launching Ernie's chatbot. (photo: twitter @baidu_inc)

JAKARTA - Baidu Inc revealed on Tuesday May 16 that their experience in adjusting their search engines to Chinese regulatory requirements led them to believe that their artificial intelligence-based chatbots would not make mistakes in "important and sensitive topics".

In a call with analysts, Baidu CEO Robin Li said that the company was awaiting approval from the government before launching Ernie's chatbot similar to ChatGPT. In a trial conducted by Reuters, it has been proven that this chatbot is reluctant to answer questions about politics, especially those related to Chinese government leaders.

"For important and sensitive topics, we have to make sure that artificial intelligence will not experience hallucinations," said Li. He uses industrial terms for when artificial intelligence models produce different outputs than expected.

"Given that LLM (big language model) is basically a probability model, this task is not easy at all," he added, referring to the models used by many artificial intelligence chatbots, such as ChatGPT and Ernie.

Li said that the industry regulations were not final, and the company would continue to update its strategy over time.

"Baidu has been operating search engines in China for more than 20 years and has extensive experience in Chinese culture and the regulatory environment," he said.

"On the other hand, companies that do not have extensive experience in providing appropriate or non-performing online content with regulators will face significant challenges," he said.

China's cyber regulator last month revealed draft measures to regulate services driven by generative artificial intelligence, such as Ernie's chatbot, by stating that the content generated by this new technology must be in accordance with the values of the country's core socialist.

Li said that these steps would benefit Baidu.

"We believe that regulatory active involvement in generative artificial intelligence in the early stages will increase the threshold, and we are in a good position for that," he said.