Slashdot Mirror

← Back to Stories (view on slashdot.org)

Can a Japanese AI Get Into University?

Posted by ryuzaki0 on from the I'm-afraid-I-don't-like-my-schedule-dave dept.

the_newsbeagle writes "Japanese researchers are trying to develop an artificial intelligence program that can pass the standardized test required of all college-bound high school students. Interestingly, the AI is showing good progress in the history portion of the exam, because it's fairly adept at looking up answers in a vast textual database. But the so-called Todai Robot is having trouble with math, 'because the questions are presented as word problems, which the Todai Robot must translate into equations that it can solve,' as well as with physics, which 'presumes that the robot understands the rules of the universe.' If the AI does succeed in mastering the general university exam, researchers will next tackle the notoriously difficult University of Tokyo entrance exam, which will require the bot to write essays."

3 of 91 comments (clear)

  1. huh by Anonymous Coward · · Score: 2, Insightful

    There is something fundamentally broken if tackling the University is considered easier than passing the Turing test.

    1. Re:huh by Anonymous Coward · · Score: 3, Insightful

      There is something fundamentally broken if tackling the University is considered easier than passing the Turing test.

      Not necessarily. In a Turing test, you can ask every question. Including questions about feelings (which can easily narrow down the other side to either a computer or an autistic person), about its own biography (where a computer obviously cannot tell the truth without revealing that it is a computer; inventing a coherent biography is much harder than telling your actual biography), about things which belong to the experience of every human, but not of intelligent computers (and there's a good change that the programmer has not thought of the very thing you are asking about), ...

      On the other hand, the university tests typically only request fixed knowledge and abilities of the type you can easily classify as right or wrong.

      Now the essay tests are more interesting, but then, in an essay you don't have much interaction, so it's much easier to stay clear of weaknesses in the software than in a Turing test.

  2. Re:Dragon Zakura by Eivind · · Score: 5, Insightful

    That I've experienced to. It's a *really* stupid way to grade someones language-skills, but it's an easy way to do it, just count the mistakes, so it's basically about caring more about ease of grading than whether grades are meaningful or not.

    "My name is Eivind. I am a boy. I come from Norway. Norway is in Europe. Norway is cold." should *never* score higher than:

    "I'm called Eivind and come from Norway, it's a coldish place over in Europe, thoug not as cold as some folks assume."

    Yeah, the latter has more mistakes. But despite this it demonstrates far higher skills in english. Failing slightly at constructing a complicated sentence should be preferable to constructing a entry-level sentence perfectly.