Some top Silicon Valley executives and a members of the public at-large fear that super-intelligent computers will someday use their smarts to take over humanity. Current technology is far too rudimentary for such a sci-fi reality, but it is making slow progress.
On Monday, the Allen Institute for Artificial Intelligence—a research organization funded by Microsoft (MSFT) co-founder Paul Allen—said that it has built an artificial intelligence system capable of answering geometry questions nearly as well as the average high school student taking an SAT test. The system, called GeoS, correctly answered 49% of the questions it was given, according to a research paper published Monday.
“If these results were extrapolated to the entire Math SAT test, the computer roughly achieved an SAT score of 500 (out of 800), the average test score for 2015,” the Allen Institute said in a statement.
That might not sound like the type of computer capable of world domination, but—sensationalism aside—it’s actually a respectable start. That’s because while so many of the latest artificial intelligence advances are essentially advances in detecting patterns (Siri’s voice recognition, Google’s facial recognition for photos and Cornell’s anticipation engine for cars that can predict what a driver does next, for example), GeoS is actually trying to make sense of the data it’s fed. It analyzes the diagrams and text it sees, but then applies what it knows about geometry to figure what the question is actually asking.
Put simply: While a typical computer vision algorithm could tell you whether an image was of a geometric diagram or, say, a guitar, GeoS is actually trying to understand the diagram.
An example of how GeoS analyzes an exam question.
GeoS is not the first effort to build a computer that can take high-school level tests. The Allen Institute is working on another, called Project Aristo, that targets (for now) fourth-grade science exams. A few years ago, Fujitsu, along with IBM (IBM) and other partners, kicked off a project to build a computer that could pass the University of Tokyo entrance exam. Some subjects are more difficult for computers to master than others, especially if the exam questions involve using common sense or parsing subtle clues rather than mathematical formulas or recitation of facts.
Truly intelligent computers like we read about in sci-fi novels or see in movies like Her will need to be able to do all of these things and more, and to do them across a number of disciplines. A practical example is a security robot, which would need to be able to accurately distinguish a gun from a phone and understand what people say. If its creators wanted the robot to be really useful, it should also be able to decide how to act based on the environmental and legal factors at play in any given situation.
Building that same robot to also help with your math homework, make coffee and predict stock market performance would be something else altogether.
Viewed in this light, projects like GeoS are neither confirmations that computers will take over the world nor evidence that computers will never be as smart as humans. Rather, they’re signs that replicating human intelligence in computers is complicated, and that we’re taking baby steps in that direction.
To learn more about artificial intelligence, watch this Fortune video:
Sign up for Data Sheet, Fortune’s daily newsletter about the business of technology.
