Chinese internet giant Baidu is now the world’s leader in a critical AI segment for understanding human language, dethroning Microsoft and Google.

Baidu, China’s biggest search engine, got the “highest ever score” in a test aimed at measuring AI language processing, MIT Technology Review reported, underscoring accelerating competition for the most efficient and human-like AI.

The test is called General Language Understanding Evaluation (Glue). While an average person score around 87 points on the test, Baidu was the first to go beyond 90 with its AI system called Ernie, Review reported.

Ernie was influenced by Google’s Bert and both are inspired by Sesame Street characters. 

In addition to supporting power-efficient technology and autonomous systems, the Holy Grail of AI research permits conversational communication, not just informational “what is the weather like?” tasks and simple assignments.

AI technologies that were first developed in a different language sometimes have advantages, like Russian search engine Yandex that developed AI assistant Alice and provides more nuanced responses and conversational replies to Russian-language users.

Baidu’s Ernie is trained to “give users more applicable search results, remove duplicate stories in its news feed, and improve its AI assistant Xiao Du’s ability to accurately respond to requests.”

Ernie was developed to differentiate “between meaningful and random strings” of Chinese characters, and as a result is “much more accurate at predicting the missing pieces,” Review noted.

“When we first started this work, we were thinking specifically about certain characteristics of the Chinese language,” Hao Tian, chief architect of Baidu Research, told Review. “But we quickly discovered that it was applicable beyond that.”

Even though Baidu is known as China’s Google, the company is aggressively expanding to other sectors like virtual assistants, smart speakers and autonomous cars.

Earlier this month, Baidu partnered with Samsung to develop power-efficient AI chips, which can be used for “large-scale AI workloads, such as search ranking, speech recognition, image processing, natural language processing, autonomous driving, and deep learning platforms.” 

  • Glue’s guidelines were developed by researchers at New York University and University of Washington.