User login
This transcript has been edited for clarity.
My oldest daughter is at sleepaway camp for a couple of weeks, and the camp has a photographer who goes around all day taking pictures of the kids, which get uploaded to a private Facebook group. In the past, I would go online every day (or, okay, several times a day) and scroll through all those pictures looking for one that features my kid.
I don’t have to do that anymore. This year, I simply uploaded a picture of my daughter to an app and artificial intelligence (AI) takes care of the rest, recognizing her face amidst the sea of smiling children, and flagging just those photos for me to peruse. It’s amazing, really. And a bit scary.
The fact that facial recognition has penetrated the summer camp market should tell you that the tech is truly ubiquitous. But today we’re going to think a bit more about what AI can do with a picture of your face, because the power of facial recognition is not just skin deep.
What’s got me hot and bothered about facial images is this paper, appearing in Cell Metabolism, which adds a new layer to the standard facial-analysis playbook: facial temperature.
To understand this paper, you need to understand a whole field of research that is developing various different “clocks” for age.
It turns out that age really is just a number. Our cells, our proteins, our biochemistry can be analyzed to give different numbers. These “clocks,” as distinct from the calendar we usually use to measure our age, might have more predictive power than the number itself.
There are numerous molecular clocks, such as telomere length, that not only correlate with calendar age but are superior to calendar age in predicting age-related complications. Testing telomere length typically requires a blood sample — and remains costly. But we can use other sources to estimate age; how about a photo?
I mean, we do this all the time when we meet someone new or, as a physician, when we meet a new patient. I have often written that a patient “appears younger than their stated age,” and we’ve all had the experience of hearing how old someone is and being shocked. I mean, have you seen Sharon Stone recently? She’s 66 years old. Okay — to be fair, there might be some outside help there. But you get the point.
Back to the Cell Metabolism paper. Researchers report on multiple algorithms to obtain an “age” from a picture of an individual’s face.
The first algorithm is pretty straightforward. Researchers collected 2811 images, all of Han Chinese individuals ranging in age from 20 to 90 years, and reconstructed a 3D facial map from those. 
They then trained a convolutional neural network to predict the individuals’ ages from the pictures. It was quite accurate, as you can see here.
In the AI age, this may not seem that impressive. A brief search online turned up dozens of apps that promised to guess my age from a photo.
I sent this rather unflattering picture of myself to ChatGPT which, after initially demurring and saying it was not designed to guess ages, pegged me at somewhere between 35 and 45, which I am taking as a major victory.
But the Cell Metabolism paper goes deeper. Literally. 
And this is where things start to get interesting. Because sure, the visible part of your face can change depending on makeup, expression, plastic surgery, and the like. But the temperature? That’s harder to fake.
It turns out that the temperature distribution in your face changes as you get older. There is a cooling of the nose and the cheeks, for example.
And the researchers could combine all this temperature data to guess someone’s calendar age fairly accurately, though notably not as accurately as the model that just looks at the pictures.
But guessing your age is not really the interesting part of thermal imaging of the face. It’s guessing — or, rather, predicting — the state of your metabolism. All these study participants had extensive metabolic testing performed, as well as detailed analysis of their lifestyle behaviors. And facial images could be used to predict those factors.
For example, the 3D reconstruction of the faces could predict who ate seafood (they tend to look younger than their actual age) compared with who ate poultry and meat (they tend to look older). The thermal imaging could predict who got more sleep (they look younger from a temperature perspective) and who ate more yogurt (also younger-appearing, temperature-wise). Facial temperature patterns could identify those with higher BMI, higher blood pressure, higher fasting glucose.
The researchers used the difference between actual and predicted age as a metric to measure illness as well. You can see here how, on average, individuals with hypertension, diabetes, and even liver cysts are “older,” at least by face temperature.
It may even be possible to use facial temperature as biofeedback. In a small study, the researchers measured the difference between facial temperature age and real age before and after 2 weeks of jump-roping. It turns out that 2 weeks of jump-roping can make you look about 5 years younger, at least as judged by a thermal camera. Or like the Predator.
Okay, this is all very cool, but I’m not saying we’ll all be doing facial temperature tests in the near future. No; what this study highlights for me is how much information about ourselves is available to those who know how to decode it. Maybe those data come from the wrinkles in our faces, or the angles of our smiles, or the speed with which we type, or the temperature of our elbows. The data have always been there, actually, but we’ve never had the tools powerful enough to analyze them until now.
When I was a kid, I was obsessed with Star Trek — I know, you’re shocked — and, of course, the famous tricorder, a scanner that could tell everything about someone’s state of health in 5 seconds from 3 feet away. That’s how I thought medicine really would be in the future. Once I got to medical school, I was disabused of that notion. But the age of data, the age of AI, may mean the tricorder age is not actually that far away.
Dr. Wilson is associate professor of medicine and public health and director of the Clinical and Translational Research Accelerator at Yale University, New Haven, Conn. He has disclosed no relevant financial relationships.
A version of this article first appeared on Medscape.com.
This transcript has been edited for clarity.
My oldest daughter is at sleepaway camp for a couple of weeks, and the camp has a photographer who goes around all day taking pictures of the kids, which get uploaded to a private Facebook group. In the past, I would go online every day (or, okay, several times a day) and scroll through all those pictures looking for one that features my kid.
I don’t have to do that anymore. This year, I simply uploaded a picture of my daughter to an app and artificial intelligence (AI) takes care of the rest, recognizing her face amidst the sea of smiling children, and flagging just those photos for me to peruse. It’s amazing, really. And a bit scary.
The fact that facial recognition has penetrated the summer camp market should tell you that the tech is truly ubiquitous. But today we’re going to think a bit more about what AI can do with a picture of your face, because the power of facial recognition is not just skin deep.
What’s got me hot and bothered about facial images is this paper, appearing in Cell Metabolism, which adds a new layer to the standard facial-analysis playbook: facial temperature.
To understand this paper, you need to understand a whole field of research that is developing various different “clocks” for age.
It turns out that age really is just a number. Our cells, our proteins, our biochemistry can be analyzed to give different numbers. These “clocks,” as distinct from the calendar we usually use to measure our age, might have more predictive power than the number itself.
There are numerous molecular clocks, such as telomere length, that not only correlate with calendar age but are superior to calendar age in predicting age-related complications. Testing telomere length typically requires a blood sample — and remains costly. But we can use other sources to estimate age; how about a photo?
I mean, we do this all the time when we meet someone new or, as a physician, when we meet a new patient. I have often written that a patient “appears younger than their stated age,” and we’ve all had the experience of hearing how old someone is and being shocked. I mean, have you seen Sharon Stone recently? She’s 66 years old. Okay — to be fair, there might be some outside help there. But you get the point.
Back to the Cell Metabolism paper. Researchers report on multiple algorithms to obtain an “age” from a picture of an individual’s face.
The first algorithm is pretty straightforward. Researchers collected 2811 images, all of Han Chinese individuals ranging in age from 20 to 90 years, and reconstructed a 3D facial map from those.
They then trained a convolutional neural network to predict the individuals’ ages from the pictures. It was quite accurate, as you can see here.
In the AI age, this may not seem that impressive. A brief search online turned up dozens of apps that promised to guess my age from a photo.
I sent this rather unflattering picture of myself to ChatGPT which, after initially demurring and saying it was not designed to guess ages, pegged me at somewhere between 35 and 45, which I am taking as a major victory.
But the Cell Metabolism paper goes deeper. Literally.
And this is where things start to get interesting. Because sure, the visible part of your face can change depending on makeup, expression, plastic surgery, and the like. But the temperature? That’s harder to fake.
It turns out that the temperature distribution in your face changes as you get older. There is a cooling of the nose and the cheeks, for example.
And the researchers could combine all this temperature data to guess someone’s calendar age fairly accurately, though notably not as accurately as the model that just looks at the pictures.
But guessing your age is not really the interesting part of thermal imaging of the face. It’s guessing — or, rather, predicting — the state of your metabolism. All these study participants had extensive metabolic testing performed, as well as detailed analysis of their lifestyle behaviors. And facial images could be used to predict those factors.
For example, the 3D reconstruction of the faces could predict who ate seafood (they tend to look younger than their actual age) compared with who ate poultry and meat (they tend to look older). The thermal imaging could predict who got more sleep (they look younger from a temperature perspective) and who ate more yogurt (also younger-appearing, temperature-wise). Facial temperature patterns could identify those with higher BMI, higher blood pressure, higher fasting glucose.
The researchers used the difference between actual and predicted age as a metric to measure illness as well. You can see here how, on average, individuals with hypertension, diabetes, and even liver cysts are “older,” at least by face temperature.
It may even be possible to use facial temperature as biofeedback. In a small study, the researchers measured the difference between facial temperature age and real age before and after 2 weeks of jump-roping. It turns out that 2 weeks of jump-roping can make you look about 5 years younger, at least as judged by a thermal camera. Or like the Predator.
Okay, this is all very cool, but I’m not saying we’ll all be doing facial temperature tests in the near future. No; what this study highlights for me is how much information about ourselves is available to those who know how to decode it. Maybe those data come from the wrinkles in our faces, or the angles of our smiles, or the speed with which we type, or the temperature of our elbows. The data have always been there, actually, but we’ve never had the tools powerful enough to analyze them until now.
When I was a kid, I was obsessed with Star Trek — I know, you’re shocked — and, of course, the famous tricorder, a scanner that could tell everything about someone’s state of health in 5 seconds from 3 feet away. That’s how I thought medicine really would be in the future. Once I got to medical school, I was disabused of that notion. But the age of data, the age of AI, may mean the tricorder age is not actually that far away.
Dr. Wilson is associate professor of medicine and public health and director of the Clinical and Translational Research Accelerator at Yale University, New Haven, Conn. He has disclosed no relevant financial relationships.
A version of this article first appeared on Medscape.com.
This transcript has been edited for clarity.
My oldest daughter is at sleepaway camp for a couple of weeks, and the camp has a photographer who goes around all day taking pictures of the kids, which get uploaded to a private Facebook group. In the past, I would go online every day (or, okay, several times a day) and scroll through all those pictures looking for one that features my kid.
I don’t have to do that anymore. This year, I simply uploaded a picture of my daughter to an app and artificial intelligence (AI) takes care of the rest, recognizing her face amidst the sea of smiling children, and flagging just those photos for me to peruse. It’s amazing, really. And a bit scary.
The fact that facial recognition has penetrated the summer camp market should tell you that the tech is truly ubiquitous. But today we’re going to think a bit more about what AI can do with a picture of your face, because the power of facial recognition is not just skin deep.
What’s got me hot and bothered about facial images is this paper, appearing in Cell Metabolism, which adds a new layer to the standard facial-analysis playbook: facial temperature.
To understand this paper, you need to understand a whole field of research that is developing various different “clocks” for age.
It turns out that age really is just a number. Our cells, our proteins, our biochemistry can be analyzed to give different numbers. These “clocks,” as distinct from the calendar we usually use to measure our age, might have more predictive power than the number itself.
There are numerous molecular clocks, such as telomere length, that not only correlate with calendar age but are superior to calendar age in predicting age-related complications. Testing telomere length typically requires a blood sample — and remains costly. But we can use other sources to estimate age; how about a photo?
I mean, we do this all the time when we meet someone new or, as a physician, when we meet a new patient. I have often written that a patient “appears younger than their stated age,” and we’ve all had the experience of hearing how old someone is and being shocked. I mean, have you seen Sharon Stone recently? She’s 66 years old. Okay — to be fair, there might be some outside help there. But you get the point.
Back to the Cell Metabolism paper. Researchers report on multiple algorithms to obtain an “age” from a picture of an individual’s face.
The first algorithm is pretty straightforward. Researchers collected 2811 images, all of Han Chinese individuals ranging in age from 20 to 90 years, and reconstructed a 3D facial map from those.
They then trained a convolutional neural network to predict the individuals’ ages from the pictures. It was quite accurate, as you can see here.
In the AI age, this may not seem that impressive. A brief search online turned up dozens of apps that promised to guess my age from a photo.
I sent this rather unflattering picture of myself to ChatGPT which, after initially demurring and saying it was not designed to guess ages, pegged me at somewhere between 35 and 45, which I am taking as a major victory.
But the Cell Metabolism paper goes deeper. Literally.
And this is where things start to get interesting. Because sure, the visible part of your face can change depending on makeup, expression, plastic surgery, and the like. But the temperature? That’s harder to fake.
It turns out that the temperature distribution in your face changes as you get older. There is a cooling of the nose and the cheeks, for example.
And the researchers could combine all this temperature data to guess someone’s calendar age fairly accurately, though notably not as accurately as the model that just looks at the pictures.
But guessing your age is not really the interesting part of thermal imaging of the face. It’s guessing — or, rather, predicting — the state of your metabolism. All these study participants had extensive metabolic testing performed, as well as detailed analysis of their lifestyle behaviors. And facial images could be used to predict those factors.
For example, the 3D reconstruction of the faces could predict who ate seafood (they tend to look younger than their actual age) compared with who ate poultry and meat (they tend to look older). The thermal imaging could predict who got more sleep (they look younger from a temperature perspective) and who ate more yogurt (also younger-appearing, temperature-wise). Facial temperature patterns could identify those with higher BMI, higher blood pressure, higher fasting glucose.
The researchers used the difference between actual and predicted age as a metric to measure illness as well. You can see here how, on average, individuals with hypertension, diabetes, and even liver cysts are “older,” at least by face temperature.
It may even be possible to use facial temperature as biofeedback. In a small study, the researchers measured the difference between facial temperature age and real age before and after 2 weeks of jump-roping. It turns out that 2 weeks of jump-roping can make you look about 5 years younger, at least as judged by a thermal camera. Or like the Predator.
Okay, this is all very cool, but I’m not saying we’ll all be doing facial temperature tests in the near future. No; what this study highlights for me is how much information about ourselves is available to those who know how to decode it. Maybe those data come from the wrinkles in our faces, or the angles of our smiles, or the speed with which we type, or the temperature of our elbows. The data have always been there, actually, but we’ve never had the tools powerful enough to analyze them until now.
When I was a kid, I was obsessed with Star Trek — I know, you’re shocked — and, of course, the famous tricorder, a scanner that could tell everything about someone’s state of health in 5 seconds from 3 feet away. That’s how I thought medicine really would be in the future. Once I got to medical school, I was disabused of that notion. But the age of data, the age of AI, may mean the tricorder age is not actually that far away.
Dr. Wilson is associate professor of medicine and public health and director of the Clinical and Translational Research Accelerator at Yale University, New Haven, Conn. He has disclosed no relevant financial relationships.
A version of this article first appeared on Medscape.com.