EP111: Dr. Duje Tadin on visual perception and expertise

Dr. Duje Tadin studies visual perception, and his take on what we see is that it is messy. I enjoyed hearing his thoughts on System 1 and System 2 as they relate to what we see in different domains. Dr. Duje Tadin, PhD is a neuroscientist renowned for his groundbreaking research in the field of visual perception. He is currently Professor and Chair, Brain and Cognitive Sciences, and the Director of Training, Center for Visual Science, at the University of Rochester. After obtaining his doctoral degree at Vanderbilt University, Tadin delved into investigating the intricacies of human vision, focusing particularly on perceptual learning and plasticity and how the brain processes visual information and adapts to changing environments. Tadin's contributions have earned him numerous accolades, including prestigious grants and awards, solidifying his reputation as a leading authority in neuroscience. 

Transcript

[00:00:00] Christine Ko: Welcome back to SEE HEAR FEEL. Today, I am honored to be speaking with Dr. Duje Tadin. Dr. Duje Tadin PhD, is a neuroscientist renowned for his groundbreaking research in the field of visual perception. He is currently Professor and Chair, Brain and Cognitive Sciences, and the Director of Training, Center for Visual Science at the University of Rochester. After obtaining his doctoral degree at Vanderbilt University, Tadin delved into investigating the intricacies of human vision, focusing particularly on perceptual learning and plasticity, and how the brain processes visual information and adapts to changing environments. Tadin's contributions have earned him numerous accolades, including prestigious grants and awards, solidifying his reputation as a leading authority in neuroscience. Welcome to Duje.

[00:00:50] Duje Tadin: Thank you, Christine, for this nice introduction. It's good to be here.

[00:00:52] Christine Ko: Could you first share a personal anecdote? 

[00:00:55] Duje Tadin: I teach one course a year on perception. There's a couple of things that I like to tell my students that they're usually surprised about. The first one is we take vision perception for granted. We open our eyes, we see. But it's a really remarkably complicated process. So I ask my students, what's easier, doing complex math or seeing? And usually students just say, seeing. But the reality is that I'm holding a smartphone here, this little phone can do almost any math I need to be done, but things as simple as just seeing a face and recognizing that person or recognizing an object is incredibly difficult. About half of our cortex is devoted to visual perception in some shape or form. This is easy for us, and we take it for granted only because a massive amount of brain power is devoted to just seeing the world around us. Looking at the students faces, they're often surprised to hear that because we think of math as something hard, but it's only hard because our brains are not specialized to do math. Our brain is specialized to do perception. 

[00:01:54] Christine Ko: In dermatology and dermatopathology, the diagnoses that we make, the decisions that we make on what something is, is really based on, what does it look like? What are we seeing? And, if I don't see something, if I miss a certain very important clue to the diagnosis, I end up in the wrong place. And so that's why I think about this a lot, and I actually think that visual perception is hugely complex, and unlike math, which only has one answer, usually the answer to, what is this really?, when it's a complex amount of visual data, is not necessarily so easy to boil down to one thing.

[00:02:32] Duje Tadin: And that's what I like about perception. It is messy. We have eyes, they get a picture of the world, but it's a two dimensional image in our retinas of a three dimensional world. 

[00:02:41] Christine Ko: Yeah. 

[00:02:41] Duje Tadin: The brain has to do a lot of guessing what is actually in front of us. And that's why we have a lot of different visual illusions, because these are all the cases where the strategies that our brain uses fail. Illusions are really example of showing how the brain does something really well, and it is one rare case things don't work out. I was just in D. C. with my children on a trip, and everybody was taking a picture of the Washington Monument, trying to hold the monument within their fingers in that sort of perspective illusion. That's one example of a visual illusion. It's really an illusion that shows the tricks that our brains use to see the world well. We have all these shortcuts. And so the messiness of perception is that it's also one reason why computer vision systems haven't gotten to be as good as we are because they haven't learned all the tricks that through evolution that we have. 

[00:03:29] Christine Ko: Yeah. I like your saying that visual perception is messy, but what horrifies me is when something important, like a patient's diagnosis, is based on my visual perception of what's going on. I don't want it to be messy. I want to learn how to make it orderly and neat. I have latched onto System 1 and 2 processing because I do think it gives a little bit of order. I'll try to just use System 1 processing for a microscopic slide. I'll just say, what do I think this is? I'm not trying to think, what's the algorithm? What's this clue? What's here? What's here? I just look and glance. But before I actually send that case out, as this is what I think this microscopic slide shows, I'll look at it another time and think more with System 2 visually and be like, what is that? Are there really three things that confirm? What's the worst thing? What if it's actually malignant melanoma, and I just totally missed it? Or what if there's more than one thing in the slide? 

[00:04:27] But, let me know your thoughts. 

[00:04:28] Duje Tadin: Perception is often messy, but our brains are often good at having their cake and eating it. So most of the time, perception is trying to do your best guess. And sometimes there are errors. But there's also ways that we can attain perceptual expertise. Most of us have categories that we're really good at seeing, and that's where it becomes less and less messy, more and more efficient, faster, automatic. And one of the things that we are all experts at is, for example, seeing faces. So we can see a face of somebody, we know immediately who that person is. They could be having a different haircut or wearing a hat. And that's just because we have massive amount of expertise seeing faces. Faces are incredibly important to our social lives and perception. When it comes to faces, you see somebody's face, and you don't have to think about, is this Jennifer, because you just open your eyes. System 1, and it's automatic, and you're not very likely to make an error there.

[00:05:23] Going back to dermatology, when you become an expert, you really go a few levels up in your regular perception. There's a neuroscientist, Isabel Gautier, that's really studied this a lot. And she's done a lot of work in people who are bird experts and car experts. There are people that can see like a part of a bumper of a car and they can tell you exactly which year and model it is. 

[00:05:45] Christine Ko: Wow. 

[00:05:46] Duje Tadin: And people that they can see part of a bird and know exactly who that is. Their brains just work differently. There are parts of the brains that become involved in processing of the information you're an expert in, separate from your regular visual processing. So this is having your cake and eating it. You have this messy system that has to deal with a massive amount of information that's highly varied, but all of us have a capacity to become perceptual experts in a limited number of topics. And that literally involves a very different type of brain. So an expert dermatologist examining a patient, messy is the opposite of what I would think. Going back to System 1, when you have high levels of expertise, messiness goes out a window. You become an expert and master your domain. Most of us can only do that for one or two areas, not many. 

[00:06:31] Christine Ko: Yeah. System 1, an overlearned process, whatever domain, you have done things over and over again. And, for us in dermatology, recognize something, the same rash, over and over again. That's an overlearned process. And you do have expertise in automatically recognizing it quickly. But the problem is for patients, right? It impacts patients when you're wrong, and you don't know you're wrong. Cause you think that, I recognize this. For example, the facial thing is a great example too. Every once in a while, I think we've all had this experience, every once in a while, usually from far away, not usually up close, you're like, oh, there's Jennifer, and you might even wave at her.

[00:07:09] And then one second later, two seconds, maybe five minutes later, if you never really got close, you're like, that wasn't Jennifer. 

[00:07:16] Duje Tadin: It's often something like confirmation bias, because if you're waiting for Jennifer, you're more likely to recognize somebody else as Jennifer. And this could very much apply to dermatology as well.

[00:07:25] Christine Ko: Yes, so for example, for microscopic slides, what most dermatopathologists will say is, okay I look at the slide first, and then read what the clinician wrote, because I don't want to be biased or influenced. So that way, hopefully, I look at the slide, and I think it's one thing. And then the dermatologist wrote something that kind of confirms, you know, Oh, yeah, like that makes sense.

[00:07:46] Duje Tadin: System 1 and 2, if I want to use that language, if I said I don't like System 1 and 2 is because I don't think of them as very separate. I think of them as often interacting. Your prior expectations, it creates a mindset that's more of a top down, which it can then change and redirect the automatic part of the process. A lot of times when we become experts on something, we do this in a very conscious way. When you try to learn how to drive a car, you think a lot about every single step. When you try to learn how to drive a stick shift, there's a lot of steps to think about. At some point, when you become an expert, you become automatic. And I think we talked earlier about it is there's a difference between two different classes of experts. There are those experts who are really good at doing but can't really explain well. And there are those experts who have this extra ability, meta knowledge, they can do both. They can be automatic, but they can tell you exactly what they did. 

[00:08:35] Christine Ko: If you look at statistics on medical error, it might be as high as 30 percent of patients are affected by medical error across, just across different specialties. You can be doing something for so long, and yet you're gonna make mistakes.

[00:08:49] Duje Tadin: Brains are messy. And it's not that just the brains are messy, but patients present in different ways. It really becomes what we call in science categorization, classification problem. If you have two categories that have no overlap, you're probably never going to make a mistake. In medicine, we deal with categories where there is overlap. 

[00:09:05] Christine Ko: Yeah. 

[00:09:05] Duje Tadin: And that overlap means that no matter how well you do, there will be things that are going to pass through. It's unfortunate but sometimes very little you can do about it. 

[00:09:14] Christine Ko: That's interesting what you say about classification and categorization. 

[00:09:17] Duje Tadin: I also teach something called signal detection theory, which is the way we make yes, no decisions. I usually use medical diagnosis as an example. You can make a false alarm. You can basically tell a patient that they have something, but they don't. Or you can have a miss when they may have something that needs follow up, but you tell them they're okay. In this kind of scenario, what you really want to do is you want to minimize misses at the cost of more false alarms.

[00:09:42] And then also contrast that with jury trials, where you can still have misses and false alarms, but there you would rather have a miss, a guilty person go free, than an innocent person go to jail. So you're trying to minimize different kinds of errors. So I don't know if I articulated that well, but you can make two kinds of errors. You can essentially have a miss of something, or you can have a false alarm. In clinical diagnosis, you really want to minimize misses. 

[00:10:08] Christine Ko: Okay. That's interesting that you say that because that goes along with what a lot of people believe is our epidemic of over diagnosis of skin cancer, or other cancers too, breast cancer, prostate cancer, because the general sense might be that we'd rather not miss one. We'd rather over call something, over diagnose it, rather than miss one. 

[00:10:31] Duje Tadin: Yeah. But there is a cost to false alarms as well, because you still need people to follow up testing, which may not show up anything at all. [Yes.]. So it's a hard decision to make because at some point you had to put a yardstick somewhere, and you're either going to have more misses or more false alarms. It's a hard public health decision to make. 

[00:10:50] Christine Ko: Yeah. 

[00:10:51] Duje Tadin: Of course, this is getting outside of my knowledge area. 

[00:10:54] Christine Ko: No, I love it. Something for me to think about. Can you talk a little bit about perceptual learning, anything that's really surprised you? 

[00:11:00] Duje Tadin: Again, because we take perception for granted, we open our eyes and see, it's easy to forget that even though we have this massive amount of brain real estate that's involved in perception, most of it is not innate. Almost all of it comes through exposure to the world. Our visual diet is so extensive, constant information through your eyes, and you interact with those objects, and the brain is just constantly training itself to see better. There's a lot of studies through development showing a lot of things that babies don't see and adults see. So our brain constantly learning and getting better and better at perception is something that even people in my field in neuroscience sometimes forget, because their daily lives, they just open their eyes and see.

[00:11:43] That plasticity part of our perception, and how important our visual diet is in training our brains to be efficient at seeing it, is something that even I sometimes forget. 

[00:11:54] Christine Ko: What you just said makes sense to me too because sometimes, for a microscopic slide, for example, I can easily take it and show someone else. And sometimes, sad for me, they'll point out, oh, look at this, and they immediately see something that I did not see. As long as I've been doing this, I'll be like, ah, in my head darn it, I didn't see that. 

[00:12:13] Duje Tadin: Of course, everybody's perceptual background experience is different, and that's the beauty of the range of different human experiences, that we will see and notice different things. That's the whole point, to me, the second opinion in medicine. People will have different background experience that may allow her to see things that somebody else didn't. And vice versa, you will see things that somebody else didn't see. So someone's getting two eyes on a slide is probably not a bad idea. 

[00:12:37] Christine Ko: Yes, absolutely. 

[00:12:38] Duje Tadin: And this is because our perception is messy. Going back to the original comment, it's messy. Different people see the world in different ways. Even identical twins, because they're going to have different experiences into their lifetime, are going to see things differently. So to me, that messiness could then be harnessed. By having multiple opinions, get a better, more robust. 

[00:12:56] Christine Ko: I love that. Dr. Kahneman's book, Noise, talks about that. Differing opinions for the same thing, even if it's your own internal opinion, your own opinion that changes. The average of those opinions is usually better than just the one, just like what you just said. 

[00:13:12] I'm gonna move on to this. What do you think everyone should know about visual perception?

[00:13:17] Duje Tadin: Our brains do so many different things. We have memory, we have emotions, we have decision making. And about half of the brain is involved in you seeing the world because humans are visual creatures. Primates are visual creatures. This is our predominant sensory system. It's not the case with all the animals. My dog relies a lot more on a sense of smell than I do. The fact that there's so much brain power dedicated to something that most of us take for granted, is to me, remarkable.

[00:13:46] Don't take your visual perception for granted. It's one of the most sophisticated things that our brains do. And it's not easy. And best way to illustrate that is how poor computer vision systems are. They're getting better, but those AI systems are going to need massive amount of resources and massive amount of training to get anywhere close to what humans already do.

[00:14:06] Christine Ko: Awesome. Do you have any final thoughts? 

[00:14:09] Duje Tadin: I'm really enjoying this. The fact that, somebody who does pathology and dermatology wants to talk to visual scientists. This is the exciting part of the science because we all want to understand how the world around us works, and we all work in our own separate areas. But sometimes the most exciting things happen when people who seemingly have very little in common start talking, realizing that they're working on the same questions, but from a different perspective. 

[00:14:34] Christine Ko: Yes. Thank you so much for spending time with me. I loved it.

[00:14:37] Duje Tadin: Right. Thank you.