Six AI teams excel in reading comprehension tests
Language could be another segment where artificial intelligence (AI) defeats humans in one of our own inventions after chess, go, strategic games and solving the Rubik’s Cube to name but a few.
LG CNS, a system-integration affiliate of Korea’s LG Group, said on March 20 that up to six AI systems developed here have beaten a majority of adults in reading comprehension tests.
A total of 15 AI teams were engaged with the Korean data-set, which the Seoul-based company unveiled last December to check AI’s ability to answer reading comprehension based on provided paragraphs.
Six got higher marks than Korean adults who achieved an average of 91.2.
Topping the list was an anonymous participant, which received 94.08 followed by systems made by Naver with 92.42 and Kakao, 92.1. The former is the country’s dominant search engine while the latter is the leading provider of mobile chatting services.
The six AIs were not only more accurate but also far speedier than humans. They were asked to deal with 3,898 questions in just 10 minutes, a Herculean task that humans would not be able to contemplate.
For comparison, LG CNS asked six university graduates to answer the 3,989 questions, and it took them 33 hours in total, or roughly half a minute for each subject. They scored an average of 91.2.
“We do not know who the top scorer was because the teams did not disclose their identities. But one thing is certain that the team came up with a very competent AI,” an LG CNS official said.
Will film ‘Her’ become reality?
Similar testing, developed by Stanford University, took place in the United States. It consisted of more than 100,000 questions posed by crowd-sourcing on various Wikipedia articles.
Like its Korean counterparts, Google’s AI programs have outperformed humans producing the hype that human-machine communications could be possible soon.
In other words, the 2013 U.S. scientific-fiction film “Her” could be a reality in the not-so-distant future.