Jeremy Wolfe

Every day, people miss what is right before their eyes. Simple typos escape our notice. We might look — but fail to see — a cyclist at an intersection. This phenomenon may sound familiar to many who have taken an introductory psychology class. But how and why our visual system misses the obvious — and what we can do about it — remain mysterious. Jeremy Wolfe, PhD, director of the Visual Attention Lab at the Brigham, and his team are studying how people process visual stimuli and the real-world applications of visual search behavior.

In a perspective piece published July 21 in Trends in Cognitive Sciences, Wolfe and his co-authors coin the term “normal blindness.”  Normal blindness, says Wolfe, isn’t a sign there is something wrong with a person’s vision — rather, it’s a problem all of us share.

“There’s a fundamental limit in the way our nervous system is put together,” said Wolfe. “Our visual system is in the business of telling us what’s most likely in front of us. Most of the time, it works fine. But occasionally, it fails us. And we are all subject to these kinds of errors.”

Failing to See the Gorilla

In their perspective piece, Wolfe and co-authors remind readers about the “classic gorilla experiments” conducted by DJ Simons and CF Chabris and published in 1999. In this experiment, participants watch a video of a ball game and are instructed to count how many times one team touches the ball. In the middle of the game, a woman in a gorilla suit walks in.

“When you ask participants if they saw anything unusual, about half of people say no. Even when you ask, ‘Did you see a gorilla?’ many will still say no,” said Wolfe. “People think of missing the gorilla as a one-off — an outlier. But we’d argue that we make similar mistakes all the time, such as when we’re proofreading a paper. The oversights may be different in detail, but they are fundamentally similar.”

A section of a lung CT image with an inserted gorilla.

Wolfe is especially interested in what normal blindness may mean in the context and practice of medicine. In a twist on the classic gorilla experiments, his lab conducted a study in which radiologists were the participants. The 24 radiologists were shown lung CT scans and asked to identify small, white, round lung nodules on the image. What the radiologists were not told is that the researchers had inserted an image of a gorilla — 48 times larger than the average nodule — into one of the scans. More than 80 percent of the participants failed to see the gorilla. Wolfe makes it clear that there was nothing “wrong” with these radiologists.

“They are simply experiencing the same normal blindness that is part of the way we are built,” said Wolfe. “Sadly, being an expert radiologist doesn’t immunize you against these phenomena.”

Everyone experiences normal blindness, Wolfe and his co-authors write, but people are mostly blind to its costs.

“Radiologists are asked to keep an eye out for incidental findings when looking at CT scans,” said Wolfe. “Now, of course, when you’re looking in the lungs, incidental findings don’t include gorillas. But we’ve spent some years in the lab conducting other studies that show that people can reliably miss what they are specifically looking for. And in the real world, this can have serious consequences.”

A Path to Vision Research

Wolfe became interested in visual research during high school when he spent a summer working in the lab of a color vision researcher who was interested in how many pixels were required to recognize a face.

“I got this summer job and I was hooked. I wanted to be a vision researcher,” Wolfe recalls.

Wolfe received an AB in Psychology from Princeton University and a PhD in Psychology from MIT. In the beginning of his career, his research focused on how our two eyes perceive just one world, but his interests changed with time.

“When I was a young faculty member, a colleague came down the hall waving a paper from one of the leading visual attention researchers, saying he was certain she was wrong, and asking why didn’t I do something about it,” said Wolfe. “That piqued my interest, and I began to change my focus.”

In addition to studying medical image perception and cancer screening, Wolfe’s lab has also studied visual attention among transportation security officers who screen baggage. The team is currently working on an approach to circumvent some of the limitations of gorilla experiments —participants can only be asked once (after that, they are more likely to look for the gorilla) and can only be asked after the image is no longer in front of them.

“I’d like to ask, what’s going on while the gorilla is on the screen? We know participants are looking in the right place — we can track where the eye goes and we can see that people look right at it — and yet they still don’t see it. We’d like to study that in more detail.”

Solutions in Sight?

When it comes to catching oversights, Wolfe says one of the best solutions is the one that many writers benefit from: an editor. Having a second pair of highly trained eyes can help catch a typo before publication; likewise, having a second radiologist read a mammogram — which is common practice in Europe — can help catch a finding that might otherwise go overlooked.

“Sometimes, things are missed because they are hard to see. But if it’s just an error like a typo, it’s less likely that two people will make that error than one person will,” said Wolfe.

However, with limited resources and limited numbers of radiologists, this solution is not always feasible. In some situations, artificial intelligence (AI) is used to help radiologists. Wolfe notes that although AI may be able to help, it certainly cannot do the work alone.

“We need the computer and human to work together as buddies,” said Wolfe. “Unfortunately, that doesn’t work that well either — we’re not that good at using the advice AI gives us.”

Wolfe and others are also studying the relationship between AI, humans and normal blindness in the context of self-driving vehicles. Wolfe’s son and daughter-in-law, who are co-authors on the recent Trends in Cognitive Sciences paper, are interested in how AI could alleviate or exacerbate normal blindness when it comes to self-driving cars.

Checklists can also help radiologists — and drivers — to pay attention to the most important things by reviewing and catching their oversights . But no solution is perfect.

“Our visual system evolved to get us through the real world reasonably successfully, but not at 80 miles per hour or at the resolution of radiological images,” said Wolfe. “From gorillas in the lung to typos on the page, we’re likely to continue to fail to see what is in front of us. But normal blindness is just that — normal. If we can better understand our limitations, we may be able to find better solutions that take them into account.”