MattI was playing around with ChatGPT last year and thought to myself: the number of photos with ages attached to them that this AI has probably consumed is likely a massive number and therefore would probably be good at estimating people’s age based on their appearance. So I tried a few people first, including my old photos when I was a lot younger, and I noticed it was incredibly accurate for most people, unless they look unsually young or old.
Unfortunately, I don’t have the best quality photos when I first started taking picture to track my progress, but even so, ChatGPT was very close to estimating my actual age still, at least before and when I started CR.
As the AI gets better, I assume these results may change, so I will try it again with few models and take better qualtiy pictures going forward. But for now, I have to use what I have.
I asked ChatGPT what it looks at to determine age and this what it said:
Some of the strongest cues are things like:
Overall facial proportions that subtly change with age
Fine lines around the eyes or forehead
Skin texture and elasticity
Facial fat distribution
Jawline definition and soft tissue changes
Hair density, hairline, and greying
Eye area shape/tiredness
You’ve probably noticed yourself that even though some people have incredibly good skin, they still tend to look their age. One of the big reasons is to do with facial proportions that change with age. It’s not just about whether or not you have wrinkles, expression lines, acne or whatever. Even in poor quality videos from 20 years ago, you can still tell which age range the person is in. We are very good at pattern recognition and have been exposed to so many people that we instinctively know how old someone is within a pretty small range. Unless of course they look unusually young or old for their age.
So let’s get to my experiment. I admit it’s not the most scientific experiment but it was just for a bit of fun while I was trying out this AI.
I’ve included all of the photos I used for the experiment. All of these photos are ones that I have previously posted on my blog or social media accounts before. The final photo I did again today because I couldn’t find the screenshot for it. It still guessed 25 anyway, so no change there. And this is using an improved image model.
Chronological age: 19
Chronological age: 20
Chronological age: 24
Chronological age: 25
Chronological age: 26
Chronological age: 27
Chronological age: 28
Chronological age: 30
Chronological age: 33
Chronological age: 35
Chronological age: 37
Chronological age: 41
After I asked ChatGPT to guess my age, I then asked it to create some graphs and tables for me so we can see how much my perceived age (according to ChatGPT) has diverged away from my chronological age.
How accurate is ChatGPT at estimating age based on appearance?
Based on my testing, I have found it to quite accurate most of the time. Is it perfect? No, it can sometimes give different ages from one day to the next, like my most recent photo. I was comparing against someone else and it guessed his age perfectly, and it compared my photo and ChatGPT said I looked in my late teens to early 20s. Well, I guess 23 is early 20s, as it gave that number before, but I probably don’t look like a teeanager anymore. 🙂
Ideally, you would use the same lighting, same angle, and same setup every time with a decent quality camera. But even without all that, it’s quite remarkable how accurate it is when I was testing photos of family and public photos of people.
Give it a try yourself!
