Support Provided ByLearn More
Tech + EngineeringTech & Engineering

Robots Are Now Smart Enough to Take the SAT

ByAllison EckNOVA NextNOVA Next

Receive emails about upcoming NOVA programs and related content, as well as featured reporting about current events through a science lens.

A new software program might just be smarter than an 11 th grader.

For the first time, an artificial intelligence program capable of seeing and reading has demonstrated the ability to answer

Support Provided ByLearn More
geometry questions from the SAT that it had not previously encountered. Though machine vision is still in its infancy—even the most sophisticated AI programs struggle to comprehend the symbolic meaning of an arrow in the context of a diagram, for example—this most recent achievement is a major step for computer scientists.

robot-writing_1024x576
This robot could take the SAT for you... if the SAT were written in Hebrew.

The team, comprised of experts from the Allen Institute for Artificial Intelligence and the University of Washington reported the milestone yesterday in

a paper presented at a conference in Lisbon, Portugal. They say that what distinguishes this program from its contemporaries is its capacity for abstract and common-sense reasoning. In this way, the software combines advanced perception (the focus of most AI research as of late) with the reasoning skills inherent in reading and understanding complete sentences.

Here’s John Markoff, reporting for The New York Times:

The Allen Institute’s program, which is known as GeoSolver , or GeoS, was described at the Conference on Empirical Methods of Natural Language Processing in Lisbon this weekend. It operates by separately generating a series of logical equations, which serve as components of possible answers, from the text and the diagram in the question. It then weighs the accuracy of the equations and tries to discern whether its interpretation of the diagram and text is strong enough to select one of the multiple-choice answers.

The Allen Institute approach has more in common with an earlier generation of artificial intelligence research that relied on logic and reasoning.

Moreover, the Allen Institute researchers said, machine-learning techniques have continued to fall short in areas where humans excel, such as problem solving.

Of course, no one test is adequate in measuring a machine’s—or a child’s—intelligence. As neural networks become more adept at complex pattern recognition, they will offer another pathway toward intelligence assessment. Eventually, experts may have a constellation of options that, combined in different ways, could give us super-smart robots.

Photo credit: Wikimedia Commons