AI Scores as High as the Average High Schooler on the Math Section of the SAT

Wednesday, 23 September 2015 - 2:37PM
Wednesday, 23 September 2015 - 2:37PM
AI may be rapidly approaching human-level intelligence. Chinese researchers recently built an AI that was shown to score slightly higher than the average human on the verbal portion of an IQ test, and now, a new study shows that AI can achieve the same score on the math portion of the SAT as the average high schooler. 

Researchers at the Allen Institute for Artificial Intelligence in Seattle uses a mathematical algorithm in order to understand and solve math problems. The algorithm creates a formal model of the problem that is able to combine text understanding and diagram interpretation, which the researchers claim "boosts the accuracy of dependency and semantic parsing of the question text."

Using this algorithm, the program was able to correctly solve 61% of the geometry practice problems and 49% of the official problems on the SAT. The researchers then extrapolated these results to a score on the entire math section, and found that the AI scored a 500, precisely the same as the average score of 11th graders in 2015. 

Does this bring us one step closer to the singularity? The infamous Turing test, in which a program needs to convince a human that it is human as well, has technically already been passed, but researchers have come to a near consensus that it is primarily based on trickery, and therefore is not an accurate measure of intelligence.

Opening quote
"Unlike the Turing Test, standardized tests such as the SAT provide us today with a way to measure a machine's ability to reason and to compare its abilities with that of a human," said Oren Etzioni, CEO of the Allen Institute for Artificial Intelligence.
Closing quote

To be fair, the SAT, similar to the IQ test, has problems of its own as a measure of intelligence. While researchers generally believe that it is correlated with IQ, there have been many accusations of cultural and socioeconomic bias. Furthermore, students (and probably computer programs as well) can improve with practice, especially since answering the specific types of questions the SAT asks is a skill set unto itself, which means it is not entirely dependent on inherent intelligence. But still, it seems to be at least associated with intelligence, so this is a significant step towards a humanlike artificial intelligence.

Via Business Insider.
Artificial Intelligence

Load Comments