Yikes! A Sketch Artist Explains What Happened Here.

Unless you just so happen to reside in southeastern England, you probably don’t think too much about what the police in Kent are up to—not to mention the rozzers assigned to the humble town of Royal Tunbridge Wells. Recently, though, the tiny municipality’s law enforcement has earned a fair amount of international attention thanks to its bizarre, perhaps overcomputerized attempts to pursue some petty criminals.

On April 11, a 90-year-old Tunbridge Wells resident heard an intruder prancing about his house and went upstairs to confront the stranger, who pretended to be a cop and then fled the premises. Actual police in the area responded to the senior’s complaint and took down a standard description of this wannabe thief:

Using a system known as Electronic Facial Identification Technique, or E-FIT for short, the Kent police mocked up a computer-generated composite of the suspect’s identifying features, which, uh, came out looking like this:

I mean, yes, as several online jokesters pointed out, this guy looks like a visually degraded Mii from the old-school Wii universe—or like a much uglier Pete Buttigieg or James Blunt, or like someone you may have fought off while playing GoldenEye on your old Nintendo 64. But, more to point, almost nothing about this “sketch” helps with the description, does it? No missing tooth, no characteristic tee, not even any actual short hair, dark though it may be. (Side note: Why does his forehead look like it’s melting?)

This isn’t even the sole E-FIT mockup to bring mockery onto the Kent fuzz: After a group of men attempted to steal a pedestrian’s watch in the town of Maidstone last month, the Kent Police produced another that spurred far more celebrity-lookalike comparisons—this time to radio broadcaster Rylan Clark—than any actually helpful leads. As far as any of us know, all these attempted thieves remain at large.

In a time when policing methods are almost fully digitized, helped by recent advances in automated image generation, the traditional forensic sketch artist is a less common part of the PD than she used to be: Only a couple dozen such artisans are employed full time with police branches across the U.S., and many others work on a freelance basis or helm their own firms. But as the Kent E-FIT debacle reminds us, such pencil pushers still have a lot to teach our modern-day crime stoppers, and even the fanciest image model can’t beat a trained sketch artist just yet.

To get an expert’s evaluation of the Kent computer image and a lesson in what really goes into an effective drawing of a suspect, I called up Carrie Stuart Parks, an FBI-trained forensic sketch artist with decades of experience who now runs the Idaho-based Stuart Parks Forensic Associates firm with her husband, which she described to me as “the largest instructor of forensic art in the world.” Our conversation has been edited and condensed for clarity.

Nitish Pahwa: What role has computer generation and automatic image generation software played in forensic art, especially with the current rise of “artificial intelligence” tools?

Carrie Stuart Parks: If you look at doing a composite sketch, you’ve got four elements. First, you have the element of the victim or witness of a crime. A lot of it has to do with modern tech: surveillance footage, Ring cameras, all of that has made a major impact. [In cases where] we used to be called in a lot, that doesn’t happen so much anymore because they have a surveillance camera that caught something. But you still have cases where there isn’t one. Most rape cases, for example, there’s no surveillance.

So you have the victim or witness of a crime, and there are things within that. What was the lighting condition? How long did they see the person? From what angle did they see the person? How is the person’s ability to articulate or to work with the artist? And how good is their memory?

The second element is your artist, who’s either freehand drawing the information or using a hybrid. They may be using an iPad or some type of tablet, but they’re still generating it through their own [drawing] skills. The third element would be the viewing public, the people who see it because it’s put out there for identification purposes: How are they viewing these images, and what are they seeing?

The final element is if the person taking the information is not an artist, then you’ve got two additional limitations. Is this person actually trained in the human face, or do they just know how to manipulate the computer program? And then, with the program itself, what are the elements and limitations of that program?

People don’t think about the viewing public when they produce these composites, along with how we recognize and identify people. The closer the image is to a photograph, the more the viewing public will look for that particular person, not for somebody who looks like that. The more exact an end result you have, the more it has to be absolutely perfect or you’re not going to get an identification. The more it looks like a sketch, a drawing, or a caricature even, the more likely you are to get an identification.

I have a test that I’ve shown people since 1988. If I show them the image and it’s a photographic representation, I have a less than .005 percent identification rate. If it is a caricature or a sketch, I have an identification closer to 98 percent. And the only reason why there’s any variation there is because sometimes people don’t know the person.

I have to ask about what you make of the computer-generated image from the Kent Police. Suppose a student submitted that to you in class. How would you evaluate it?

Well, they wouldn’t have submitted that because that’s computer-generated: It’s using actual facial features to represent the person. Anytime you start using a cut-up photograph, you’ve got problems with it. But let’s say we keep everything the same, only instead of those nasty-looking photograph cut-ups, it was a simple drawing. When I look at it, what do you notice about it?

I notice the hair upfront, plus the jaw line.

OK, short chin?

Yeah, the outline of the mouth, and the space between the lips and the nose. It is a very oddly shaped head.

Yeah, head shape or chin, funky hair.

The ears aren’t really noticeable. I can’t really tell what the eyes are supposed to be. And there are a lot of weird skin waves on the forehead?

So if a student brought this to me, I would ask, “What did the witness tell you?” You have to draw what the witness says, no matter how weird it is. And we have forms that they fill out, so I can actually look at what they say.

Let’s say they’re using my interview form. One of the questions we ask is, “What is the overall impression of this guy?” If they said he’s a weird-looking dude, I expect it to be strange. My expectation is this is not going to be an average face. This drawing is an odd-looking drawing, so they match. So I know that the problem at this point is not necessarily with the artist.

But if the witness said, “Oh, he was an average-looking guy.” I look at that, I’ll see that it’s not an average face. So the problem is not the witness, the problem is the artist. So what is the problem? What were the features that were chosen?

A big problem with this one is in the facial proportions. If the face is “average-looking,” there is a place for the eyes, nose, mouth, chin to be. Do you happen to know where the mouth goes between the nose and the chin? Chances are not. If you don’t know, then the “mistake” is the person doing it, not in what they were told. So you have to determine how close is this to their memory?

The final thing is the proportions of the face, which are odd enough that it makes you go, “Oh, he’s got a really low mouth. He’s got a really short chin. You know what that looks like? That weird-looking guy up the street.” So it does achieve the goal if that had been the witness’s point. If you look at the weirdness, it would remind you of another person that had, maybe not quite that weird of facial features, but maybe close to that.

What kind of lesson do you think this particular mess demonstrates for other police departments trying to get ahold of these digital systems, and then ending up with some output that may not necessarily be the most helpful?

A number of years ago, I attended a conference where they were showing off one of the older systems. And they said, “We have sold 700 computers with this on it.” Let’s say that each department that bought one did 10 images. That’s 7,000 images out there. They said, “We think we finally have a successful identification.” And I thought, “Do you realize what the identification rate is with that program?”

In a perfect world, we would love to be able to sit down and tweak here and twiddle there and doodle, doodle, doodle, and spit out this incredible photographic image of what the guy looks like and it’s perfect and it’s wonderful. That’s where the computer people, whose goal is to sell the programs, say, “Wouldn’t you want to have a photograph of the guy? We can make this. You don’t even have to be an artist. You can just sit down and push a few buttons. And here it is.” Basically, it’s a sales tactic.

But what you have to do is bring people back to what the end product requires: knowledge of victims or witnesses, what they’re going through, knowledge of the human face, and knowledge of the public that’s going to be viewing what it is that’s turned out.

And there are always trends. They always say, “Oh, we’ve got the newest and the latest.” They’ll stop sending students for a while because they say, “No, we’ve got this great fantastic thing.” They use it once or twice, realize it doesn’t work, and then they have to save up to get money to actually get their person trained.

Right now there’s nothing that any computer system can do that an artist with a pencil and piece of paper can’t do. And it’s still that knowledge that that artist has, regardless of what they use to put the image down. I use an iPad and pencil, and it still requires that knowledge. You still have to have the idea and know how to do it.