For the first time, artificial intelligence models have outperformed humans in a respected Stanford University reading test used to gauge neural networks’ cognitive prowess.

Chinese web giant Alibaba Group Holding Ltd. announced on Monday that it was an AI developed by its research arm that reached the milestone by earning a score of 82.44. That compares with the 82.304 record of the human participants. A day after Alibaba’s results were finalized, Microsoft Corp. repeated the feat with a deep learning model that produced a slightly higher 82.65.

The test, which is known as the Stanford Question Answering Dataset, consists of more than 100,000 question-and-answer pairs derived from 500 Wikipedia articles. All the text used for the exam was written in natural language as opposed to a structured format that would be easier for a computer to understand. As a result, the fact that AI models can now produce better scores than humans represents a major technical achievement. 

There are many potential applications for a deep learning model capable of interpreting written materials with humanlike accuracy. Luo Si, the chief scientist for natural language processing at Alibaba’s research arm, named customer service and healthcare as two areas where the technology could prove particularly useful.

The tech industry is working to advance AI on other fronts as well. In August, Microsoft researchers revealed that they’ve developed a speech recognition system capable of transcribing text with an error rate of just 5.1 percent, which is on par with the human average.

Alibaba, meanwhile, is actively working to gain an edge over its fellow tech giants in the AI arms race. Three months ago, the company announced plans to invest a hefty $15 billion into engineering over the next three years. The effort will place a particular emphasis on technologies such as deep learning that are important to its long-term growth plans.

Image: Pixabay