Microsoft and Alibaba AI now better at reading than us humans

16 Jan 2018434 Views

Share on FacebookTweet about this on TwitterShare on LinkedInShare on Google+Pin on PinterestShare on RedditEmail this to someone

Image: Bas Nastassia/Shutterstock

Share on FacebookTweet about this on TwitterShare on LinkedInShare on Google+Pin on PinterestShare on RedditEmail this to someone

After mastering games such as chess and Go, AI is now better than humans at one of our most-loved hobbies: reading.

If artificial intelligence (AI) is to perform a multitude of tasks – from driving us to a destination, to learning about our likes and dislikes – it will need to master the art of reading.

However, as we have seen a number of times before with facial-recognition technology, the results can often be as disastrous as they are impressive.

But now, Chinese tech giant Alibaba has announced that its own research into AI has produced a new first: a machine capable of reading faster and better than a human.

According to CNN, the company revealed that its system had beaten the performance of humans who took the highly regarded Stanford University reading comprehension test on 11 January.

The test was created at the university by computer scientists to quantify the progress of the computer’s ability to read, and Alibaba’s AI is the first to beat it.

The test is designed to generate questions based on a series of Wikipedia articles. For example, it could give an AI the article on World War I and ask it: What was the name of Archduke Franz Ferdinand’s assassin?

‘This milestone is just a start’

When the results came in, Alibaba’s AI showed a competency score of 82.44, thereby narrowly beating the human opponent who scored 82.304.

In response, the company’s chief scientist of natural language processing, Luo Si, described it as a “great honour” and said its applications will be numerous in the future.

“[The technology] can be gradually applied to numerous applications such as customer service, museum tutorials and online responses to medical inquiries from patients, decreasing the need for human input in an unprecedented way,” Luo said.

The next step, according to Stanford researcher Pranav Rajpurkar, is to improve the overall accuracy of the AI, as humans still lead by 2.5 points, according to the F1 metric.

Microsoft also put its AI to the test, achieving a score of 82.650, meaning both it and Alibaba are tied for first place on the Stanford leaderboard.

Ming Zhou, assistant managing director at Microsoft Research Asia, said of its own result: “Natural language processing is still an area with lots of challenges that we all need to keep investing in and pushing forward. This milestone is just a start.”

Colm Gorey is a journalist with Siliconrepublic.com

editorial@siliconrepublic.com