Alan Kelly, chief creative officer at Rothco, part of Accenture Interactive, tells LBB’s Alex Reeves about how tech helped his agency team and Identv to become detectives
Alcatraz is probably the most famous high-security prison the world has ever known. 36 people are thought to have attempted escape from the island penitentiary in San Francisco Bay, but it’s commonly believed that none succeeded. There is one case, however, that calls that into question.
On the night of June 11th, 1962, three prisoners - Frank Morris, along with brothers John and Clarence Anglin - broke free from their cells and made it to the shoreline of the island. The trio, made famous by Clint Eastwood in the 1979 film Escape from Alcatraz, used sharpened spoons to bore through the prison walls and left papier-maché dummies in their beds. Most have assumed they failed, drowning as they fled the island on a raft fashioned from raincoats. But in 2015 a grainy old photo surfaced, allegedly showing the Anglin brothers in Brazil in 1975 - 13 years after their escape.
Nobody has been able to confirm the identities of the men in the photo, but recent advances in technology have allowed the Irish agency Rothco, part of Accenture Interactive, to work with machine learning and AI experts Identv to unlock the hidden data in the scratchy image. Using a facial matching system based on a type of machine learning algorithm called a deep neural network, ‘The Long Shot’ project appears to have made progress in the case.
But what does this mean for the long-cold case? Could this technology help to crack other mysteries like it? LBB’s Alex Reeves asked Rothco chief creative officer Alan Kelly about the ambitious project.
LBB> Where did the first seed of this idea come from? Was it a technology insight or more led by a brief?
Alan> The genesis for this idea originated with a totally different project. About 18 months ago, a photo surfaced of what purported to be Amelia Earhart sitting on the dock of a Japanese harbour with a plane that looked a lot like hers being winched out of the sea. This story was huge for about 48 hours but was debunked when it came to light that the photo was from a popular Japanese book and was taken before Amelia Earhart disappeared. So unfortunately that project died, but the idea stayed with us.
LBB> How did you hear about the story of the Alcatraz escapees? And what were your initial thoughts?
Alan> I’ve known about this story ever since Clint Eastwood appeared in ‘Escape from Alcatraz’. I remember that movie poster appearing all over Dublin when I was a kid. It showed a steely eyed Clint peering out of a crack in the prison wall – an instant classic.
LBB> Why do you think so many people remain intrigued by this story?
Alan> People love a good old mystery. I think this story appeals to the Agatha Christie in all of us.
LBB> When you saw the photo of the brothers in Brazil, what were your initial thoughts on what might be possible given the technology we now have access to?
Alan> I thought… ‘Well the beards and sunglasses might be a problem’ but the amazing thing is - the tech is now so advanced, it can take those factors into account.
LBB> What was the journey like finding the right methodology to give the most conclusive result?
Alan> At Rothco we love ‘what if’ ideas, and now that we are part of Accenture Interactive, we can match those ‘what if’ ideas with ‘here’s how’ solutions. We now have access to very smart people in the Dock in Dublin and in Accenture R&D in France. They quickly told us how to solve this.
LBB> What was the search for the original image like? How did you go about it?
Alan> That took a while… We had to turn part-creative, part-private detectives. I got in-touch with a producer who worked on an Alcatraz documentary for the History channel. He knew a man, who knew a man, who put me in touch with a woman, who knew another producer, who knew a man who had the high res photo. It took me a couple of months to track it down.
LBB> Can you explain the process of turning an old photograph into a CG model of a face?
Alan> That was pure serendipity. Our producer Mags had MPC over giving a presentation on their post skills (which are amazing). So we literally ambushed them. We told them about the mystery, they loved it and went away and just did it. Legends!
LBB> How did the training process work? What did you train it on?
Alan> In a nutshell, Identv’s process is to feed an algorithm many images of a single person's face along with many images of different people, they repeat this process millions of times over and internally they use mathematical models that are tweaked during the training phase to learn how to differentiate one person's face from another.
LBB> In the simplest way (if that's possible!) what exactly is a ‘convolutional neural network’?
Alan> A convolutional neural network (CNN) is a deep learning algorithm which can take in an input image, assign importance (learnable weights and biases) to various aspects/objects in the image and be able to differentiate one from the other.
LBB> How long did it take to turn up a result once the network had been trained and set to task on the image?
Alan> Once it was set up, not long at all. Minutes.
LBB> What is the consensus around this application of technology? What would it take for it to be admissible in court for example?
Alan> I think the consensus is this is as far as technology can prove something right now. But you’d still want to match it with DNA proof etc. in a court of law.
LBB> How have relevant parties (FBI, the relevant families, news media etc.) reacted to the results so far?
Alan> Too soon to say.
LBB> Where do you think / hope the story will go next?
Alan> I hope they are all still alive. They would be in their 80s now. If they are alive - I hope to see them on Ellen in the very near future.