"Virtual Robots" Befuddled by Optical Illusions

Joab Jackson
for National Geographic News

October 11, 2007

Look this way, R2D2! Hah—fooled you!

As it turns out, robots—like humans—can be duped by optical illusions, a new study says.

That's because context and experience is everything when it comes to perceiving the world, researchers at University College London found.

"The patterns of light that fall to your eyes are meaningless," said lead study author R. Beau Lotto.

It's your prior experiences that let you know whether to fear or approach what you see, according to Lotto.

For the experiment, Lotto's lab did not build actual robots. (See a photo of a lifelike robot from 2005.)

Rather, the researchers gave commercial software used in some robots the ability to process visual cues into an artificial nerve network.

Although far simpler than our brains, the nerve networks of these "virtual robots" can, with training, mimic how nerve cells of the brain respond to light.

But, Lotto said, "we did not aim to model human perception or the brain explicitly."

Instead, the researchers modeled the patterns of light humans might typically see. The team then let the software "learn" on its own how to recognize these patterns.

Making the Illusion

The researchers exposed the software's nerve network to 10,000 images of varying gray scales placed in overlapping circles, like fallen leaves in a forest.

After exposure, study co-author David Corney presented the program with a set of images that are optical illusions to the human eye.

Optical illusions are images that trick the viewer into seeing something different from what exists in reality.

One simple illusion presented to the software is "simultaneous brightness contrast," involving pictures of two identically shaded tiles.

When the tiles are placed against contrasting shades of gray, a person will see the tile on the dark background as lighter than the tile on the lighter background.

The lab's software program misjudged reflective ability just like humans would: Tiles that would look brighter than they were to us were depicted brighter than they really were by the artificial nerve network, Corney said.

Based on its earlier experience with other images, "the network assumed that the object on the dark background is more likely to be in shadow than the one on the light background," Lotto said.

The study was recently published in the Public Library of Science's journal Computational Biology.

Artificial Fools

For the researchers, this finding suggests that optical illusions aren't unique to humans.

That's a "radical departure from the approach to vision taken in the last 50 years," said Dale Purves, director of Duke University's Center for Cognitive Neuroscience. He was not involved in this research, but Purves was a co-author with Lotto on a book about perception.

"Most vision scientists have been glued to the idea that to understand perception you must understand the properties of the cells individually," Purves said.

In this theory, our photo receptors evolved over time to fine-tune our vision.

By contrast, Lotto's research suggests all visual distinctions aren't hardwired but rather represent the statistical accumulation of an individual's earlier experience.

In that sense, over time, a living creature's brain understands images in terms of their usefulness or danger.

"When the brain looks out to the world, it doesn't see the absolute of things. It sees relationships and uses relationships to build context," Lotto said.

Take-Home Lessons

The research can benefit the robotic realm, Lotto said. For instance, designers need to consider that their industrial workplace robots might be confused by what they see.

For humans, there are lessons as well.

"Time is short, and for many of us, we have the luxury to decide how we want to spend what few conscious hours we have," Lotto said.

For example, watching television will cause us to perceive the world in one way, while climbing Mount Kilimanjaro or swimming with dolphins may allow us to understand the world more deeply, Lotto added.

"How we spend those hours will shape our perceptual and conceptual truths in the future," he said, "not just as individuals but also as a society."

Free Email News Updates
Sign up for our Inside National Geographic newsletter. Every two weeks we'll send you our top stories and pictures (see sample).