GeoS – short for geometry problem solver, an artificial intelligence software devised by the University of Washington, has just successfully completed 49% percent of a high-school math test.
Though the result might not sound all that impressive at first, it was in fact on par with the average results of the students that took the same exam. Despite common preconceptions, computers are not that great at math as some of us are inclined to believe. The artificial intelligence is a significant leap forward in the field, even if it might not pose a great threat for humanity, at least yet.
The problem with computers is that they generally do not solve questions that are anything but in a binary format. Therefore, problems that are in a complex text format are simply not understood by the artificial intelligence. So computers can solve simple formulas by following the calculations that have to be processed step by step and successfully reach the result.
GeoS is different, as the artificial intelligence had received the same questions as the students. The test was your regular geometry test, which means that the software was able to understand the text and diagrams printed on the piece of paper.
GeoS was not only able to identify that the diagram was a diagram, but also it managed to understand how was the diagram related to the text – a breakthrough in computer vision. That is the true achievement here, it is not about solvin geometry problems, it is about identifying geometry problems from high-school geometry tests. What the artificial intelligence software did is understand the problems from a human point of view, by using human language.
By getting a 500 score out of a total 800 on the SAT, the artificial intelligence managed to be on par with the average students’ score for this year. Scientists claim that this is a monumental achievement.
GeoS isn’t the first artificial intelligence project that was conceived to partake at a human exam with different subjects. Other projects such as Aristo, from the Allen Institute for Brain Science is working on a fourth grade exam solving machine, whereas Fujitsu has started working on an artificial intelligence computer that will attempt to pass the entrance exam of the University of Tokyo.
GeoS’s intelligence is definitely far from what the Terminator franchise fanbase is expecting from a computer. We should not start worrying yet, because replicating the human mind is one, if not the most difficult thing science is trying to achieve.
Photo credits: Pixabay