Geometry may be a bane for many students but an artificial intelligence system called GeoS appears competent in the subject. The AI is capable of solving SAT geometry questions on par with the level of an average 11th grade student in the U.S.
The system, which was developed by researchers from the University of Washington and Allen Institute for Artificial Intelligence (AI2), uses computer vision for interpreting diagrams, natural language processing, and a geometric solver that allowed the AI to achieve a 49 percent score on official SAT test questions.
GeoS answered unaltered SAT question that it has not yet encountered before and which required an understanding of ambiguous references, implicit relationships, and relationships between natural language texts and diagrams.
The AI solves the problems by interpreting a question using the text and diagram in concert to come up with the best possible expressions of the problem that it sends to the geometric solver for solution. The machine then compares this answer to the choices in the multiple choice for that question.
"In our experiments, GEOS achieves a 49% score on official SAT questions, and a score of 61% on practice questions," the researchers reported. "We show that by integrating textual and visual information, GEOS boosts the accuracy of dependency and semantic parsing of the question text."
Should the results be extrapolated to the entire Math SAT test, GeoS will receive an SAT score of 500, the average test score for this year.
AI2 CEO Oren Etzioni said that standardized tests such as the SAT offer a means for researchers to measure the reasoning ability of a machine and compare this with that of a human.
"Much of what we understand from text and graphics is not explicitly stated, and requires far more knowledge than we appreciate," Etzioni said. "Creating a system to be able to successfully take these tests is challenging, and we are proud to achieve these unprecedented results."
Many of the new AI systems are advanced in detecting patterns what but makes GeoS different is that it tries to make sense of the data that it is fed. It analyzes the diagrams and texts and then applies geometry information that it knows about to determine what the question is asking for.
While a typical computer vision algorithm can determine if an image is a geometric diagram, GeoS is in essence tries to understand this particular diagram.
Photo: Philipp | Flickr