Are your fingerprints really unique?

A new AI tool says it can detect similarities in fingerprints that humans can't.

Fingerprints have long been known to be completely unique. They also don’t change their pattern over your lifetime, making them an extremely useful biometric for identification. Their uniqueness largely comes from how they form in the womb: as waves of skin cells growing in random patterns of ridges under the top layer of skin in our hands and feet.

Fingerprints are so unique that it is considered impossible to match two different fingerprints from the same person — the only way to know for sure is to match a fingerprint to the exact finger. But a new AI tool developed by students at Columbia University says there are more similarities in intra-person, or same person, fingerprints than we’ve previously known.

Sources and further reading:

“Unveiling intra-person fingerprint similarity via deep contrastive learning,” by Gabe Guo et al.

“The developmental basis of fingerprint pattern formation and variation,” by James D. Glover et al.

One of the original studies of fingerprints, Francis Galton’s 1892 publication “Finger Prints”

"Accuracy and reliability of forensic latent fingerprint decisions," by Bradford T. Ulery et al.

Related Vox videos:

Kim's video explaining how your voice is like a fingerprint

Joss's classic documentary on forensics, False Positive

[Vox]

Thanks for stopping by. This is a new format we’re trying out – what would you like to see us talk about next? -Coleman

[MrBitflipper]

I kept waiting for you to bring up the case of the Oregon man arrested in connection with a terrorist train bombing in Spain. That was based on fingerprint matching, even though the suspect couldn't have possibly been the guy.

[johnchessant]

4:41 Alan Turing?!? Not the name I expected to pop up here lol, that's incredible

[censoredviews]

This video has such a relaxed and inviting vibe—almost like chatting with friends. We need more content like this!

[주-hi]

I'm here trying to watch the video and look at my own thumb at the same time

[justCallMeOmid]

The fact that an AI model couldn’t predict results based on minutiae doesn’t necessarily mean that minutiae are useless! If you read the article, you’ll see that the model learns to perform using a vectorized representation of the ridges. This means the model doesn’t truly understand minutiae at all! Even if you only provide it with minutiae, it will still attempt to create some form of ridge vectors and process them. That’s why the model performs only slightly better than a coin flip!

It’s similar to me trying to read Russian. Since I don’t know the language or its characters, I would try to map them to sounds or symbols I recognize (like English letters). As a result, I might get some mappings right but completely misinterpret others. The fact that "I" can’t read Russian based on my limited knowledge doesn’t mean Russian is unreadable!

The same principle applies here. A model that doesn’t understand minutiae can’t effectively use them, leading to poor performance. But that doesn’t mean minutiae are useless! Nor does it prove that ridge orientation vectors are inherently superior. It simply shows that, "for this model", ridge orientation vectors are the best source of information.

[Cleath]

How did they not know that the prints on different fingers are different? Have they never owned a phone with a fingerprint reader?

[AmrouBouaziz]

I like the style of the video. It feels like a casual convertiation that I could be a part of. we need more of this :)

[lijmoo]

Great video format. Can imagine coffee with this guy: "How was your weekend?" "Let me show you this 120 page dossier on the science of weekends."

[baluvideo]

I'm sorry, but how on Earth is it surprising that all 10 of your fingers have different fingerprints? They are different fingers! Americans never cease to amaze me.

[josenadie]

I loved the video and was very charmed by the format. Thank you!

On the other hand, I know it's a very relevant topic regarding pattern analysis and such, but I can't help feeling a bit weary when it started discussing AI. Is anyone else here also a bit fatigued by the AI conversation?

[mister_ed]

1:09 Why would anyone think all of a person's fingers had the same print?

[ranpergames]

I like this format! Its as if someone who dove into a rabbit hole and can finally tell another person about it and I like that

[part-timegamer]

Am i the only who is watchin all the videos just for the editing style and techniques?

[aliengeo]

While obviously the friction ridges are set, your fingerprints DO change over time, they just don't change pattern. Fingerprints wear away, some people's faster than others. I found this out when I was being fingerprinted for one of those exact cards. Apparently my natural fingerprints look like a "bad print." I have to update my fingerprint auth on my phone every couple of years because of it.

[JWentu]

2:57 again... are you talking of 60% of PEOPLE or 60% of FINGERS?? these are the details that make me judge a video as low quality (and I usually appreciate Vox videos very much). Honestly, So far I would rate this video a 4 out of 10 so far...

[thecatspajamas]

Two things:
1. I DO like this style of video. I like hearing someone else learning and asking questions along with me.
And 2. Minutiae has 3 different pronunciations. All of the ways you said it were correct. (I had to look it up after the last bit 😊)

[KanameYuuki]

I liked this format, asking a simple question and having a bit of a lecture without going too deep on a topic.

[AnonymousOG]

Such an overused quote: "The AI is not going to replace the [profession], its going to replace the [profession] that doesn't use AI."

[elhassanelnasir]

I really loved this new format you have adopted here! It's just so much fun! Also the video is really incredible!