AI failed to enter Tokyo University.
Torobo-kun failed again. But the developers of the affectionately named artificial-intelligence software system need not feel bad, since the robot is in the same boat as nearly all high schoolers when it comes to the entrance examination for Japan’s top university.
The National Institute of Informatics launched in fiscal 2011 a project designed to push the limits of AI software, setting a goal of developing by fiscal 2021 a system capable of passing the exam to get into the University of Tokyo.
But in the fourth mock test, the institute said Monday, Torobo-kun did no better than the previous attempt last year. It scored 57.1 on the standard-deviation curve. The software performed much better in physics but worse in mathematics.
According to project director Noriko Arai, the software has limited reading comprehension, which is essential to understanding the deep meaning of the questions.
The project will thus shift focus. Rather than trying to develop an AI capable of passing the test, it will work to boost the software’s reading comprehension and develop industrial applications.
The system did score higher than the national average and has an 80% or higher chance of success at 23 public universities and 512 private universities.