Alibaba’s ‘Ai’ out to prove it can recognise aesthetic beauty by predicting winner of reality TV singing contest

Artificial intelligence can master the world’s most complicated board game, beating South Korean Go grandmaster Lee Se-dol in the process. But can it predict the winner of a reality TV singing contest?

That is the challenge facing “Ai”, an artificial intelligence programme developed by Alibaba Cloud, the cloud computing arm of China’s e-commerce juggernaut Alibaba.

This Friday, Ai will attempt to prove that it can perceive something as subjective as the aesthetic beauty of musical notes by predicting the winner of Hunan TV’s “I’m a Singer”. The broadcaster is named after the southern Chinese province in which it is based.

Ai will make its judgement based on an analysis of the seven finalists’ singing ability, the popularity of the songs sung and the online reaction of viewers, among other factors, according to Alibaba Cloud.

The prediction will be released at the same time that the judges announce the winner, determined by a judges’ vote and the votes of 500 audience members.

Chinese start-up on track to deliver artificial intelligence-on-a-chip

“[The result] is very random and almost impossible to predict using human intelligence,” said Min Wanli, chief scientist for artificial intelligence at Alibaba Cloud.

“We aim to achieve a real-time prediction by Ai.”

The move comes just one month after Google’s AlphaGo stunned the world by beating Lee in four out of five games of Go, a game where players use black and white tiles to dominate territory.

But some experts say Ai’s job is even more daunting due to the emotional and subjective aspects of assessing how a person sings.

“In order to appreciate a singer’s performance, there are a lot of human factors to be taken into consideration,” said Neil Wang, greater China president for research firm Frost & Sullivan.

AlphaGo on a smartphone? Chinese brothers plan to put AI in your pocket with start-up based on Cambrian deep-learning processor

“You have to figure out whether it is a popular song, the genre, whether it is appealing to the judges, as well as whether the singer sang well. All these have to be taken into consideration during every single performance,” he added.

Wang said that while both systems are based on neural networking, AlphaGo differs from Ai in that it has been fed a lot of historical game data to allow it to better learn how a Go player may play. But this kind of data is lacking when it comes to singing.

Neural networking is a technique that simulates how a human brain operates via a host of data structures.

“Ai has been taught how to appreciate a person’s singing by assessing the pitch, the range and the tone,” Wang said, adding that analysing these elements is difficult due to the high level of subjectivity involved.

They think it’s all over. It is now. Google computer wins final game against S. Korean Go grandmaster

Chen Xinhe, deputy secretary general of the Zhongguancun Big Data Industry Alliance in Beijing, said artificial intelligence is usually more effective when there are predetermined rules for the programme to follow.

“For singing, there are a lot of emotional ties involved, such as the audience’s preferences to a song. This is hard to quantitatively capture and evaluate based on the program’s logic,” said Chen.

Alibaba Cloud hopes to apply Ai to areas such as personal assistance, weather analysis, smart cities and social trend predictions, it said.

Ai has correctly predicted two of three winners on the singing show this month already, the company added.

Alibaba Group is the owner of the South China Morning Post.