Hannaneh Hajishirzi Leads Breakthrough AI Research
UW EE Assistant Research Professor Hannaneh Hajishirzi was the lead faculty member working on an artificial intelligence system that is able to, for the first time, solve SAT geometry questions as well as an average human test taker.
A major breakthrough in the field of artificial intelligence, the research was undertaken by UW researchers, including Hajishirzi and CSE graduate student Minjoon Seo, in collaboration with the Allen Institute for Artificial Intelligence.
A paper detailing the research, titled “Solving Geometry Problems: Combining Text and Diagram Interpretation” was presented recently at the 2015 Conference on empirical Methods in Natural Language processing in Lisbon, Portugal.
Called GeoS, the software uses natural language processing, computer vision and a geometric solver to answer questions accurately enough to achieve a SAT score of 500, which was the average test score in 2015.
“For students, the language and diagram understanding parts are easy, but the reasoning and problem solving are challenging,” Hajishirzi said. “For computers, both steps are challenging.”
The techniques and algorithms for GeoS were designed by both Hajishirzi and Seo. The researchers designed algorithms to recognize the semantics of questions by understanding texts and diagrams.
Next steps for the research are to use similar techniques in other domains such as science questions and interactive systems, Hajishirzi said. They also plan to develop tutoring systems for educational purposes.