These seems to be for the most part on the same level as some good police sketches, what surprised me is that the few renders where they showed a tooth gap that matched the photo quite well, the seem also to be pretty good at predicting some facial features like the nose but overall fail quite poorly at eyebrows (i'm not sure if it's because of the neutral expression or the algo itself).
Dystopian indeed but if these examples are actually representative I'm almost astounded by the resutls.
While I don’t disagree quite a few of their predictions didn’t match that well so it’s hard to tell how much selection bias there is because we don’t know how many uses they have, the posters seem to be all the cases they worked that allowed the photos to be released, some of the media ones weren’t that close either.
Marketing department smart enough to make the advertisong material pique the interest of the target market, rather than have them, and you, think "they've obviously selected the best possible results".
Considering the method (and generously assuming these are not cherry-picked) I would call these pretty impressive. However, it seems that for practical purposes dispersion is large enough for it to be not that useful. I mean, if you have a reasonable set of suspects, this would help immensely, but basically to a degree that identifying race/ethnicity, age & sex would. Otherwise, it's a pretty generic GTA 3 character model, it would be hard to look for "somebody like that", when you know that the nose/face shape could turn out to be quite different.
This is why police sketches don't work well on people with close relationship to the subject of the sketch but work rather well for law enforcement and the general public.
When you rely on a sketch you don't rely on it being a photocopy of the subject but on specific features matching the actual appearance of the subject.
For example eye distance, brow height, nose shape, chin shape matching up 2-3 sketch features with a suspected target tend to lead to fairly high likelihood of a correct match.
Do police sketches actually work well? I doubt it. No one has ever done a controlled experiment to measure the rate of cases closed with or without police sketches.
Not everyone is born with the same length of telomeres and not everyone ages at the same rate so it’s not a good predictor unless you have multiple samples over a sufficiently long enough period.
> what surprised me is that the few renders where they showed a tooth gap that matched the photo quite well
So seeing this I had a thought and went back to look. It looks like the renders either have a tooth gap or a closed mouth (and in one case the actual photo had a closed mouth) - could their model just add a tooth gap regardless, or did I just miss one without?
This is the slippery slope that privacy advocates have sometimes been characterized as 'paranoid' for pointing out - "The database is just for ____ crimes!": national security, terrorism, heinous sexual assaults, child exploitation etc. But there is almost no world in which we do not slowly and methodically expand the scope of 'bad enough' crimes that the data gets used for. In this case, it slid from sexual assault to just plain assault, but there is rarely a line like this that gets kept firm.
I understand why would you think that but you underestimate how little data cases can have and just getting basic sketch of race, gender and average facial features that you can use in the field and modify as new information gets available is big deal. But I can see movies presenting this as magic device with 100% exact face. I see big benefit in increasing speed of dna sample analysis and reducing price even more.
That would be like a layman hearing someone say "general AI isn't possible, their claims are bollocks" and someone responding with "maybe that's true, but do you have a background in general AI?".
And this attitude causes things like police to arrest the wrong person because they don't understand the limits of machine learning.
I don't need to be an expert to simply say this is problematic and shouldn't be used. YOU need an expert to explain why it CAN be used.
We need more skepticism, not less.
But to answer your question, I understand biology well enough to understand the challenges involved here. But you don't even need that to understand the problem. In their marketing photo's, they've attempted to replicate the _HAIRSTYLE_. That by itself is all you need to understand how low the applicability of this is.
Just imagine a police officer arresting someone who was bald and sort of looked like the criminal because this algorithm chose to render them as bald. People aren't NEARLY good enough at being able to identify a _STRANGER_ with a radically different hairstyle to be trusted with this.
If you say something, be it positive or negative, you need to be able to back it up, so no it's not the wrong attitude.
> YOU need an expert to explain why it CAN be used.
Trying to offload onto ME the validating of YOUR claim is wrong. You claim, you back it up. Not me.
> We need more skepticism, not less.
Oh sure! But skepticisim isn't denial. You denied, calling it 'magic'. You back it up, with stats and science, not namecalling.
> they've attempted to replicate the _HAIRSTYLE_. That
I'm not stupid, I noticed that too. And I'm not stupid, I'm not going to put any faith into this until some value is shown, scientifically not via the marketing department. Don't assume I do.
In England sexual assault requires deliberate physical contact; the contact has to be sexual; and there has to be no consent or belief of consent. Common assault doesn't require any of that, and you can be convicted of assault without physically touching the victim.
Sorry, the phrasing there is unfortunate - what I'm trying to say is that there was an already strange moral 'ordering' to crimes (that I agree is artificial) that was slowly eroded.
Maya Forstater wasn't "fired" for "denying basic biology", and this is all entirely civil law (an argument between two parties settled by the courts) and has nothing at all to do with criminal law (an argument between the state and an individual).
1) she wasn't fired, her contract was not renewed.
2) This had nothing to do with biology (which she got wrong), but was to do with her choice to harass people in the workplace, and to harass people on social media using an account tied to her workplace.
I wish people would read the Maya Forstater tribunal case notes, because it's clear that there's a lot of misinformation.
She tried to make the case that she held a philosophical belief and that her belief was thus protected. She failed because her belief was incoherent and because her absolutionist approach requires her to cause harm to other people.
> However, I consider that the Claimant's view, in its absolutist nature, is incompatible with human dignity and fundamental rights of others.
[...]
> I conclude from this, and the totality of the evidence, that the Claimant is absolutist in her view of sex and it is a core component of her belief that she will refer to a person by the sex she considered appropriate even if it violates their dignity and/or creates an intimidating, hostile, degrading, humiliating or offensive environment. The approach is not worthy of respect in a democratic society.
To what extent do genes encode the facial features that make them distinctly recognizable to us? Is there a way to say it as a "%"?
I can certainly see how your basic head size/shape and position of nose/mouth/eyes, etc is baked in there somehow.
But I'd love to know, the fundamentals of why you look like your parents, what is that? Certain (75%?) placement of facial features locked in by genes, so of course you look like them, but then the last 25% up to developmental noise? (Example -- fingerprints, almost completely determined by developmental noise, right?)
Identical twins, sharing the same DNA, look rather identical but not truly identical. Even when raised apart.
I am skeptical of reconstructing faces from DNA using current knowledge and technology though. It seems theoretically feasible, but I wouldn't bet on this company being able to do it well.
Maybe so, but fathers can look a great deal like sons and they only share something like half their code and don't share a womb at all. Identical twins share their code and look even more alike.
I think later life matters like obesity probably make a bigger difference. Being fat or gaunt can dramatically change somebody's appearance. Grooming habits as well. Spies deliberately adopting disguises can make themselves look quite different by modifying these sort of things.
Not necessarily. Fitness, sun, sleep, alcohol, drugs, diet, smoking, and stress are some examples of environmental factors that can influence your appearance including your face after birth.
Yeah. In fact, the least I would expect from this article are some samples of "real/reconstructed", but as normally expected from journalists it is just a lot of words talking about some "controversy" and non-descriptive anecdotal poorly written detective stories instead of something of value.
Oh for sure, I fully agree. However, is this company's tech actually that good? It would be interesting to see how well it does with the DNA of real twins; how close can they actually get compared to two people with the same DNA? Could it make a face that really looks like their triplet?
Unfortunately the article doesn't have any comparison images of any sort, so it's hard to evaluate. But the skeptical/cynical part of me suspects the article would contain such comparisons if it worked well.
This assumes that the company is able to produce an accurate image that's on par with a photo of the subject's twin, when it's entirely possible that the images the company produces are no more accurate than noise.
Depending on how you breathe, your posture, how you sit, if you exercised consistently as a child, etc. your face can look completely different. The greatest example of this is how early humans had significantly larger jaws due to their diet and exercise, while humans today who have almost the exact same facial genes have much smaller jaws.
Identical twins having extremely similar faces isn’t exactly a great argument for determining facial characteristics via dna, as identical twins also shared the same environment during fetal growth. It’s entirely possible that genetics and developmental experience combined creates facial structure, and that dna alone is insufficient.
I guess you should compare to non-identical twins. Who I think look like siblings who happen to be the same age. But no doubt someone somewhere has tried to quantify this.
I think "deforming accident" is a lot easier than some might realize. I personally have had an injury that probably changed the shape of my face, and it wasn't super note-worthy. Especially if someone gets something like a broken jaw, you can look almost unrecognizable.
No, as identical twins also share whatever happened in utero.
To draw a strong example, take fetal alcohol syndrome, which often causes facial changes. Two identical twins with FAS will share similar facial changes, but these changes will not actually be encoded in their dna, but instead will be the result of developmental history.
You keep repeating that DNA doesn't have a strong role in determining what a person looks like. Yet everything we know about development including twin-studies seems to suggest that in fact it has strong role.
For example, this scientific article published in Nature opens with this:
"The human face is a complex trait under strong genetic control, as evidenced by the striking visual similarity between twins."
Probably pretty predictive. Identical twins have different (but correlated) developmental noise and look anything from very similar to totally identical.
Imagine when people who don't know any better start plugging this into facial-recognition cameras, and arrest based on the chaining of two flawed/imperfect algorithms. Not if, but when.
I particularly like how it can predict hair styles. Not only how a barber would do dreads but also how they would color it and how they happened to brush their hair on a given day. Very impressive!
Would be interesting to put a confidence "interval" on the predicted face. Perhaps show a range of faces that are consistent with the model. That might help to alleviate some concerns.
https://snapshot.parabon-nanolabs.com/examples (TV/Media volunteers)
https://snapshot.parabon-nanolabs.com/posters (Actual cases from the looks of it)
These seems to be for the most part on the same level as some good police sketches, what surprised me is that the few renders where they showed a tooth gap that matched the photo quite well, the seem also to be pretty good at predicting some facial features like the nose but overall fail quite poorly at eyebrows (i'm not sure if it's because of the neutral expression or the algo itself).
Dystopian indeed but if these examples are actually representative I'm almost astounded by the resutls.