Math is arguably one of the most logical and rational of sciences, which in theory should be a subject a system should excel at, and AI computer gets a 49% score on high school geometry, essentially proving that we’re taking steps in that direction.
- GeoS was created by researchers at the Allen Institute for Artificial Intelligence
- It managed to get a 49% score, roughly 500/800
- In the next three years, researchers hope to advance it into tackling science problems
Researchers at the Allen Institute for Artificial Intelligence, which was funded by Microsoft, have made advancements in the field by designing a system able to solve rudimentary geometry problems that are often found in the math section of the SAT. It is the standardized tests for all high school students across the United States, and now almost achievable for computers.
According to co-founder Paul Allen, their new artificial intelligence system, called GeoS, was able to solve 49% of the geometry questions it was given. If “extrapolated to the entire Math SAT test”, it means it would’ve achieved a roughly 500 out of 800 score, which was the average for high school students across the United States.
It may not be the world dominating computer that minds such as Stephen Hawking or Elon Musk fear might bring an end to the human race, but it’s a step down that path. We have come far in technology and robotics, so much so that we’re slowly passing over the hurdle of mere recognition.
With personal voice assistants like Siri or Cortana able to recognize what we say, to anticipation engines that can predict our next moves, scientists are moving into a field where AIs will be able to properly interpret problems and solve them. So far, the biggest issue comes in the matter of using common sense or identifying potential trick questions.
GeoS makes sense of the data it’s fed, taking the images and text, interpreting them as questions and then sending it to the solver after a full analysis. It detects the patterns and attempts to understand the diagrams. However, some math questions have little hidden clues between the lines that might require the use of common sense and accurate language interpretation that might not go as deep for a computer.
After it assesses the data, GeoS compares the results with the multiple choices given on the test.
According to Ali Farhadi, senior research manager, the biggest challenge was finding a way to convert questions to a language the computer will understand. In 3 years, they believe that they will be able to progress with building their system until it will solve the entire Math section of the SAT, before they advance further to science questions.
Perhaps, one day, when the AIs will be able to use human-like resources like common sense and knowledge, computers might be able to take on more problems, find more uses in real-world tasks, and hopefully be nice to the human race.
Image source: nancepage.weebly.com
ozonator says
Deniers would call 49% an elitist, dependent on “gummit” funding, and insulting to the coal-fired slide rules.