If you’re into true crime, you’ll know how often the wrong person is selected in a police line-up. Well, a group of psychologists argue it’s because current police line-up procedures don’t help witnesses choose from suspects effectively – but new AI technology could better jog their memories.
Unlike in Hollywood movies, current police line-ups usually don’t have a victim choose from a group of suspects paraded around a physical room. Instead, a victim only views static photos of them, which offers a limited view of possible culprits.
A new interactive system developed by researchers at the University of Birmingham wants to change this. Using 3D models, a perpetrator can be viewed from multiple angles – including the one the victim first saw them at.
The results? The researchers found that interactive viewing improves the accuracy of witness selection by 42 per cent. Even compared to video line-ups – when the suspects are filmed turning – the technology improved selection accuracy by 20 per cent.
“We need to do better to increase the odds that guilty people are identified while decreasing the odds that innocent people are selected from line-ups,” Prof Heather Flowe, who has been working on this project for over nine years, tells BBC Science Focus.
Existing line-up methods that police forces in the UK and USA use are ineffective and cause errors, says Flowe. In America, the suspects’ photos do not even have to be consistent – ranging from driver’s license photos to people in prison uniforms. “So you have this hodge-podge line-up,” Flowe says. “It’s just not fair.”
These line-ups are missing tricks to help jog memories, she says – tricks which Flowe and her team are investigating. While her current technology allows witnesses to interact with the suspects’ images, the team are testing ways to include dynamic facial movements, emotional expression, changes in lighting, and accessories such as masks using AI.
The ethos behind Flowe’s research is that the more of these you can provide a witness, the more accurate their choice will be.
But for now, Flowe is focusing on getting the interactive line-up technology rolled out. Her team is in discussion with the US and UK police forces before it starts being tested in the field.
“It's a good opportunity for police to get on board and start using better tech, which really hasn't changed in 100 years – we're still using just photographs when we could do so much better,” Flowe says.
Better line-up technology could have caught killers
So how does the new procedure work? The technology turns video into an interactive 3D image that witnesses can click and drag to change the perspective. This means they can look at the suspects from different angles, including above and below.
To build these interactive images, the research team – comprised of experts in psychology, computational modelling, and artificial intelligence (AI) – used the extensive video library of the UK police force along with existing research in memory and facial recognition.
It’s the “ultimate line-up procedure”, says Flowe proudly. “It's about giving the witness the opportunity to reinstate how they saw the perpetrator.”
Flowe cites the Malkinson case as an example of where better line-up procedures could have led to better justice. In 2003, Andrew Malkinson was wrongfully convicted for the rape of a victim who mistakenly chose him in a suspect line-up, despite sharing few physical characteristics with the victim’s description of the actual perpetrator.
“There are lots of cases where we just haven't thought about how witnesses saw the perpetrator,” Flowe says.
In fact, she recently wrote a paper about the Ted Bundy case. One of the survivors saw Bundy from the side during his attack, then later had to pick his photo from a front-facing line-up. Bundy’s defence challenged the victim on this, claiming that her choice must have been influenced by the familiarity of his image in newspapers.
There’s exasperation in Flowe’s voice as she recounts this: “We should show the witness at test what they actually saw.”
Advancing justice with AI
In the future, Flowe hopes that technology will help to bring static, front-on photos of suspects – like those drivers license pictures in the USA – to life.
Her next project is investigating the impact of showing witnesses different emotional expressions on suspects’ faces during line-ups. The team has already received a research grant to work on this with researchers from the Max Planck Institute in Germany, the University of Victoria in Canada, and the University of Stirling in Scotland.
But how would the police make suspects act out emotions? If someone has been accused of a crime – guilty or not – the last thing they would want to do is put on a smile for the benefit of the line-up, right? After all, Flowe says she’s seen mug shots where police are literally holding the perpetrator down because they don’t want to have their photo taken.
That’s where the AI comes in. Flowe’s group are testing the use of AI to create line-ups that generate photorealistic facial expressions – though she acknowledges that they need to explore caveats like AI falsely recreating someone’s emotions.
Later, AI could even be used in what’s known as contextual reinstatement: superimposing suspects into the scene with all the evidence in question. “There’s strong theoretical theory to support that notion”, Flowe says.
Read more: