Home › Techno › How DNA and artificial intelligence can recreate a killer's face
How DNA and artificial intelligence can recreate a killer's face
01.02.2024
American law enforcement officials claim that 3D facial models created from DNA from crime scenes can be used to solve old cases. How the method works and why it is called controversial - in the retelling of the material of Wired.
In 2017, detectives with the East Bay Regional Park Police Department were poring over a long-standing unsolved case. They had an idea that might help them solve the murder of Maria Jane Waidhofer, whose body was found in Tilden Regional Park in 1990.
Nearly 30 years later, the department sent genetic material from the crime scene to Parabon NanoLabs, a company that claims it can recreate a person's face from their DNA.
The data was fed into a machine learning model, and soon detectives had the face of a potential suspect.
The company assumed that the killer was a man. He had fair skin without freckles, brown eyes, brown hair, and thick eyebrows.
It wasn’t a photograph, but a 3D render, a bridge across the uncanny valley between reality and science fiction. The algorithm suggested what a person with the genetic characteristics found in the DNA sample might look like. The artist who draws the photoworks added a vague short haircut and mustache in Photoshop—details that aren’t found in DNA, but that eyewitnesses mentioned.
In 2017, the department made a controversial decision to release the resulting photo. In 2020, one of the detectives did something that seems even more problematic — and also violates Parabon NanoLabs’ user agreement: he asked the Northern California Regional Analytical Center to run the image through a facial recognition program.
This was revealed in a hacked police archive published by the group Distributed Denial of Secrets. It is the first known case of police attempting to use facial recognition technology on an image generated by a DNA-based algorithm from a crime scene. However, it is unlikely to be the last.
It sounds like a dystopian plot, but according to facial recognition experts and privacy advocates, it's entirely predictable.
As Jennifer Lynch, legal counsel for the Electronic Frontier Foundation, points out, the method is more likely to produce incorrect results than useful clues. “First of all, there is no evidence that Parabon can accurately recreate a face. It is very dangerous because it creates the risk of a person becoming a suspect in a crime they did not commit,” Lynch believes.
It is not yet clear whether the detective's request was granted. The think tank did not respond to requests for comment. However, in 2021, the agency's executive director, Mike Sena, said that every time the center receives a request, it conducts a search.
For Parabon NanoLabs, this is not just a violation of the user agreement, but also a nightmare idea in general.
The company was founded in 2008 and primarily provides uk number data forensic genealogy services to law enforcement — comparing DNA data with profiles in genealogical databases to find potential criminals or victims.
the U.S. Department of Defense to study DNA phenotyping, which is the prediction of appearance based on genetic data alone. The method was planned to be used to identify individuals who create improvised explosive devices from their remains.
Ellen Graytak, director of bioinformatics at Parabon NanoLabs, says the company uses machine learning to create predictive models for each part of the face. The process involves more than a thousand volunteers, comparing their DNA data with 3D scans of their faces to train them.
According to Graytak, each face has 21,000 phenotypes — visible physical features — which the models process to understand how regions of a DNA sample affect appearance.
Parabon claims it can reliably predict a person's hair, eye and skin color, as well as the number of freckles and overall facial shape. These phenotypes are used to generate facial renderings that the company creates for law enforcement.
Parabon's methods have not been peer-reviewed, and scientists question how useful they are at determining face shape. The company says its work has been presented at conferences and the technology has been tested on thousands of samples.
When Parabon NanoLabs launched its service in 2015, its agreement did not explicitly prohibit the use of predictions for facial recognition. However, law enforcement clients soon began asking whether the technology would be suitable for this scenario.