> That combination of results, Williamson and the other researchers wrote, suggested many students from mainland China were using significant chunks of pre-memorized shell text.
Well, that's one heck of a way to game an AI!
It reminds me of the confucian exam system [0] where you had to just straight up memorize a huge book of Master Kong's scripts and hope the person that re-copied your work did so too.
> ... they found that the machine boosted students from China by an average of 1.3 points on the grading scale and under-scored African Americans by .81 points.
Note that this is on a 0-5 point scale. So that says that the mean for the Chinese students was ~26% higher than the human grade and for African American students it was ~16.2% lower. The Chinese students could be considered (very roughly) to be boosted from a C- to an A+, and the African American Students from an A+ to a B-, for comparison.
Well, that's one heck of a way to game an AI!
It reminds me of the confucian exam system [0] where you had to just straight up memorize a huge book of Master Kong's scripts and hope the person that re-copied your work did so too.
> ... they found that the machine boosted students from China by an average of 1.3 points on the grading scale and under-scored African Americans by .81 points.
Note that this is on a 0-5 point scale. So that says that the mean for the Chinese students was ~26% higher than the human grade and for African American students it was ~16.2% lower. The Chinese students could be considered (very roughly) to be boosted from a C- to an A+, and the African American Students from an A+ to a B-, for comparison.
What a mess.
[0] https://en.wikipedia.org/wiki/Imperial_examination#Details_o...