Software Is Smart Enough for SAT, but Still Far from Intelligent
nytimes.com
nytimes.com
I went through Stanford in the 1980s, when the "expert systems" crowd had reached the "emperor has no clothes" state and the "AI winter" was starting. Three decades later, AI has become un-stuck, and there's real progress every year. The next decade is going to be interesting.
Fifty years out, the machines will be in charge, but I won't live that long.
Obviously, software is not actually smart enough for SAT ;)
This is kind of an interesting case, in that the article, contradicting the headline, says that the software was only given some "geometry questions" from the SAT, and that it only performed at the level of an average 11th grader, which is shockingly bad. Writing the software to solve SAT geometry problems with perfect accuracy isn't difficult at all, so what's being tested here is the ability to parse visual diagrams and natural English sentences into an accurate representation of the problem. (Again, this is made clear in the article, but seems basically unrelated to the headline.)
But since that's what we're doing, I don't see why we don't give it the verbal questions too. Understanding written English sentences is most of what's left there.