Even Computers Can Be Fooled By Optical Illusions
We here at Zenni have written about optical illusions extensively (see here, here, and here). However, it’s not just humans whose eyes can play tricks on itself. Computers can be tricked too, a new study finds.
Each computer model was shown a pair of lines, one longer than the other, and each line had both an arrowhead and an arrow tail or an X at both ends. The computer model then had to guess which line was longer. Over time, the researchers were able to train the system, named HMAX, to correctly identify the longer line 90 percent of the time.
Testing like this can result in something that sounds a bit like what a mad scientist would do; as researcher Astrid Zeman told LiveScience, “If we think of this visual system as something we implant in a robot, this means that we can grow whole bunch of robots up in different environments. Then, once our robots have matured and have learnt to see things, we can then smash their brains open to see what they are thinking. This is something that we can’t quite do with humans.”
The second part of the study showed a pair of lines to the computer system, but this time the top line always had two arrow tails and the bottom always had two arrowheads. For humans, if both lines are the same length, we are duped to believe the top line looks longer. And the study showed that the computer system was also duped around 1.6 percent of the time.
With a finding like this, the researchers are able to eliminate previously believed explanation for this illusion in humans—was it our brains misinterpreting the arrowheads and arrow tails as depth cues? Or do we focus more on overall information about shapes than their elemental parts? These findings show, as LiveScience wrote, that it may result simply by how our visual system processes information that requires further elucidation.
“If we build robots with artificial brains that are modeled off our brains, the implication is that these robots would also see illusions much like we do,” Zeman added. “By imitating the amazing accuracy, flexibility, and robustness that we have in recognizing objects, we could also be copying potential errors in computation that manifest in visual illusions. … These illusions bring to light new questions about how we perceive the world and the assumptions we make about the world.”