The author of “Race in a Bottle” discusses the risks and pitfalls of a technology that uses genetic samples to create virtual mugshots.
In October of this year, police officials in Edmonton, Canada were struggling to solve a 2019 sexual assault case in which a woman was left unconscious and almost fully unclothed in minus 16-degree Fahrenheit weather. There were no witnesses, no CCTV footage, nor any DNA matches in criminal databases, and the assailant was wearing heavy winter clothing, meaning the victim could only provide a vague description. Desperate for a break, the local police department resorted to a controversial technology for the first time in its history: forensic DNA phenotyping, which predicts a suspect's physical features directly from their DNA.
The police posted a computer-generated mugshot of a young Black male of primarily East African ancestry. But just two days later, they removed it after the image was criticized on social media and in Edmonton's Black community for its “broadness” and exacerbating racial stereotypes relating to criminal behavior. In the U.S., African Americans are about 7 times more likely to be wrongfully convicted of murder than White people. Black men are also about 2.5 times more likely than White men to be killed by police during their lifetime. When a questionable technology like DNA phenotyping gets used in Black neighborhoods, the risks of misuse are potentially fatal.
The value of DNA phenotyping is widely disputed. It’s based on how certain genes influence eye, skin, and hair color, as well as face shape, but scientific evidence for the technique is limited and some people believe it perpetuates the myth that race is rooted in biology. The images produced are predictions with dubious accuracy, critics say, that may fuel racial profiling and stigmatization.
While basic versions of DNA phenotyping are only about 20 years old, the use of forensic science in ways that discriminate against different racial groups began as early as the late 19th and early 20th centuries, within the origins of forensic anthropology. The field was created by anatomists who analyzed human skeletons with the seemingly innocuous goal of determining their ancestral origin. But the practice was far from neutral: Scientists tended to use these observations to classify races in ways that promoted the superiority of people of European descent. More recently, scientists have shown that forensic pathologists often fall into racial and ethnic biases when analyzing evidence, including assuming that Black children die from homicide rather than an accident, compared to White children.
Yet DNA phenotyping continues to be used around the world, including in Australia and the United States, and is permitted by law and practiced in several European countries. First used in Europe in the Netherlands in 1999 to determine the biogeographic ancestry of a rape and murder suspect, companies like Parabon NanoLabs now create snapshots of suspects based on DNA samples. The Virginia-based company, which specializes in DNA technology, claims to have helped police enforcement close nearly 250 cases since 2018, using genetic genealogy paired with DNA phenotyping that predicts physical features and ancestry, and may include a computer-generated snapshot.
Undark recently discussed the use of forensic DNA phenotyping with Jonathan Kahn, a professor of law and biology at Northeastern University and a leading authority on how biotechnologies shape our ideas and rights, in particular regarding race and justice. His influential 2012 book, "Race in a Bottle," explored the inappropriate use of race in biomedicine, and he has long been a critic of what he has described as the weaponization of "racialized DNA" across the political spectrum. Our interview was conducted recently over Zoom and has been edited for length and clarity.
Undark: What are your first impressions of this technology? On the one hand, it’s been touted by police and law enforcement as this breakthrough technology that has helped solve cases that had hit a dead end. On the other hand, you have people saying it is potentially discriminatory technology and might actually be very bad for communities that are marginalized.
Jonathan Kahn: My general sense, aside from the issue of the possible discriminatory impacts — which I think are significant — but on its own terms I’m very skeptical of the practical efficacy of this technology. I think there's a lot of hype around it because it's sort of the mystique of DNA and the idea of its great power. It’s been around for a long time. The evidence that it's actually been primarily responsible for having this sort of breakthrough, so far as I can tell, is very scant. When Parabon is saying things like “Oh, we have contributed to solving X numbers of cases,” well, what does “contribute to” really mean? What we don't see is how often it's used and it doesn't lead anywhere. At least I haven't seen that data.
More generally, my other feeling is, yes, of course it's going to happen just the way it did in Edmonton, which is you get the presentation of this proposed phenotype. And the idea that you can turn DNA into a specific picture, I think, is well — I guess the word that comes to mind is “ludicrous.” I’m not an expert in that technology, per se. But I'm sort of familiar generally with how it works and it seems like it's just way overclaiming.
UD: One of the questions that critics have raised is whether these ambiguous descriptions and images open up possibilities for discrimination and racial profiling, given U.S. law enforcement’s history of police violence and brutality.
JK: Well, I mean, I guess it’s sort of yes and no. It's not uncommon in any sort of crime for the victim to use racial descriptors in characterizing the perpetrator and say, “Oh, yes, the person who attacked me was and looked to be Black, Latino, Asian, White,” whatever. So the genetic technology here doesn't add a whole lot to that. What it does is, where the victim can't speak for whatever reason, it tries to provide some information — and that in itself is not unreasonable to try to fill that in, because racial descriptors are used in crime all the time.
Where the racism comes in is oftentimes not the fact that you're identifying a potential suspect by their race, but by how the law enforcement authorities act on that information. There are histories of doing things like this based on DNA, like “racial dragnets” or more particularly there are just innumerable stories, especially for African Americans in the U.S., of being pulled over for driving while Black, where they say, “Oh, yes, we had a report that there was a Black person and a Black male in the neighborhood doing X, Y, or Z” and they just leave out the fact that the report was of a Black man who was like, 6’4’’ and looked to be around 20 years old and they’re pulling over some 45-year-old Black man who’s 5’5’’ and weighs 400 pounds and saying, “Oh, we had a report of a Black man.”
And those sorts of occurrences are daily things happening that have nothing to do with the DNA technology. But the question is whether or not this technology is feeding part of that world. Is it augmenting that tendency?
UD: So forensic DNA phenotyping will create an image and provide confidence estimates about whether the suspect has brown skin or black hair or other. Then the police will have a face that they can use as a reference to narrow down their suspect list. Is that feeding into the problem in any way?
JK: The thing about presenting the picture, something that actually looks like a specific human being, is it potentially places dozens, hundreds, thousands of people under unwarranted suspicion. Especially because it's rendered almost photographically, which is even different from a crime sketch that you get from a witness.
It's not going to place people specifically under suspicion because of their race, per se. But people within that racial group who bear any phenotypic resemblance to this produced photograph become subjects of surveillance and suspicion in a way that, I think, is very problematic.
UD: The argument from law enforcement and companies that provide this service is that forensic DNA phenotyping helps them in cases where they have no lead whatsoever. You don't have a witness who saw what happened. Or they have a really poor description of what happened. And so they say that this provides information — imperfect information — that is predictive, not certain, but at least gives them something to lean on. I'm wondering what you think about this philosophy, which I think is the most compelling to police when they are trying to solve a case.
JK: So there are a few things. In something like the Edmonton case, it's not that “Oh well, they shouldn't use this information.” But it's what really is this information? And how should it be used? I think the idea Parabon is producing these suggested traits with confidence estimates, I think as a practical matter is ridiculous. I mean, what police officer — especially if they’re releasing it to the public — is going to be processing this idea about statistical reliability of, you know, X, Y, or Z?
You might be able to get some information about: “OK, well, we know the DNA tells us the sex,” so we go on that. If you want to try to use ancestry informative markers or something that seems to clearly indicate that this person has blue eyes, OK — that’s fine. Keeping it sort of one-to-one — we have information on this trait, this trait, this trait. The idea that you can then create an overall portrait of the person, a sort of mugshot, that's the step that I think is really problematic.
UD: And making it public.
Making it public is really problematic but also just sort of sticking to the actual information you have rather than extrapolating from it. I guess instead of the devil being in the details, the devil's in the extrapolation.
UD: So it's about how we interpret this information, how we weaponize it, use it in society. And what are the risks, particularly in relation to marginalized communities, of misusing this technology in the way that you described?
JK: I think the most obvious difference is it just reinforces and feeds into existing sort of racist dynamics in policing, which are pervasive. It's not being used in a vacuum, it's being fed into a system that is already deeply suffused with highly problematic racist assumptions and racist practices. And I think just reinforcing that but under the veneer of scientific objectivity.
What's going on here is that the characterization of the genes may have a measure of objectivity to it. But even then, the genes that we've chosen to look at involved human choice. Again, the extrapolation is supremely not scientific, it's supremely political and involves choices that are not dictated by science, but are choices made by Parabon and by law enforcement — about what they think is relevant.
The idea that you can then create an overall portrait of the person, a sort of mugshot, that's the step that I think is really problematic.
UD: One thing that I think is affecting the way we're having this conversation is that we don't actually know exactly how their system, their programming, their algorithms and so forth, are used to create these images to help law enforcement. And I think that's not a matter of coincidence — I think they've actively decided not to share that. It’s not open source.
JK: It’s proprietary, yes, which is extremely problematic, especially when you see the recent information about sentencing algorithms and things of that sort, and the kind of racist assumptions built into the various sentencing algorithms in the criminal justice system.
UD: I guess my question is specifically about the secrecy or the sort of hoarding of this algorithm, this information — and then providing this service to police enforcement. The fact that it’s not open source, is that a problem?
JK: If they're making scientific claims, it should be subject to scientific review and their method should be subject to scientific review.
UD: One thing we haven't spoken about — we’ve been quite bleak — is what we can do about this kind of technology for communities or police enforcement or for other relevant stakeholders. What can they do to either make it useful in the way we use it or actually just try to push against it in ways that can be constructive?
JK: I guess for me the main concern is the overclaiming and the practical utility. Like how much of a difference does it really make? How much of it is just sort of the bells and whistles of this fancy new technology? Is it really improving law enforcement? I see very little evidence of that.
But the other thing is simply to regulate or push back against the overclaiming. And I think the big thing is to push back against the production of these pseudo-mugshots and say, “OK, this is the information we have” and just do it in terms of words: brown eyes, straight hair, fairly complected, darkly complected — keeping the claims humble, not overextending the claims themselves. To my mind, the main thing is you've got to provide justification and explanation for the claims and I don't see that forthcoming.
UD: I know you're concerned with biotechnologies more generally and their relationship to injustice. Are there precedents that can help shed light on this current technology in terms of how it's used or misused and help us understand what the future might look like if we don't have regulations?
JK: Historically, I'd say the early precedent — it's not in a law enforcement setting — would be with the ancestry testing companies. Especially in the early days, they were making wildly overstated claims about the accuracy and specificity of what they could tell you. It has gotten somewhat better and more refined.
Unfortunately, it still tends to sort of collapse people into “Oh, you're Black, White, Red, Brown” — traditional racial categories, and so in that sense it tends to reinforce the idea that race is genetic, which is highly problematic. And this technology does the same thing because it produces the impression, right? That, “Oh, yes, the genes are telling us it was a Black man.”
One of the dangerous things is this, like so many other uses of genetic technology in relation to race, is used in a way that implies that genes tell us race. And that's a broader, more general issue that's very problematic, sort of a broad cultural issue. Because genes don't tell us race.
UPDATE: An earlier version of this article omitted that Parabon NanoLabs provided genetic genealogy alongside DNA phenotyping to police enforcement in the nearly 250 cases they claimed to have helped close. It has also been updated to clarify that such DNA phenotyping always includes physical features and ancestry.