9 September 2017

A robot did better than 80% of students on the University of Tokyo entrance exam

CHRIS WELLER

At the 2017 TED Conference this past April, AI expert Noriko Arai gave a talk presenting her Todai Robot, a machine that has been programmed to take the entrance exam to Japan's most prestigious university, Tokyo University.

While Arai discovered Todai didn't pass muster to gain acceptance, the robot still beat 80% of the students taking the exam, which consisted of seven sections, including math, English, science, and a 600-word essay writing portion.

But it wasn't necessarily cause for celebration, Arai said. "Instead, I was alarmed."

When Arai thinks about all the evidence claiming machines will replace huge swaths of the global workforce - first in manufacturing and low-skill jobs, and then perhaps in white-collar professions - she sees it as an indication that education is flawed.

Instead of absorbing meaning from their studies, Arai has observed children behaving more like her Todai robot. They ingest facts and spit them back out, without comprehension. The problem is, Todai and other forms of AI will inevitably surpass human memory and cognition at some point, research has suggested. The human brain can never compete with the rote fact-checking power of a computer.

Humans excel at pattern recognition, creative projects, and problem solving. We can read and understand. As Arai notes repeatedly in her talk, computers cannot - at least, so far.

Sometimes Todai makes mistakes. Even with 15 billion sentences in its repertoire, Arai's robot failed to grasp a multiple choice question whose answer would have been obvious even to young children. But as Arai found in her follow-up tests, in which one-third of middle schoolers couldn't answer a simple reading comprehension question, easy questions can also beguile human minds.

"So we have to think about a new type of education," Arai said, one in which kids are taught not just to deposit facts, but also to analyze and think critically about them. "How we humans will coexist with AI is something we have to think about carefully, based on solid evidence. At the same time, we have to think in a hurry because time is running out."

No comments: