When Detective Inspector Tim Thomas from the Western Australia Police Force (WAPF) watches popular TV crime shows he has always felt a twinge of envy and resentment towards his pretend peers on the screen.
The technology wizardry called upon to instantly track down crooks from the most fleeting glimpses of disparate evidence has a name among real life cops … “The CSI effect”. It leads to detectives coming up against disgruntled members of the public, perplexed that crimes are not solved already, and even juries in cases basing their thoughts around what they believe is possible, rather than the reality.
However Thomas, who is in charge of covert online operations, digital evidence operations and cyber-crime investigations, now has the pleasant feeling that he is the guy with the tech surprises up his sleeve.
His team has just completed a pilot project of a new artificial-intelligence-based investigation platform, which they called Söze after the elusive villain in the 1990s movie The Usual Suspects.
They believe it has the potential to dramatically reduce the time it takes to crack cases, by having virtual eyes ploughing through unfathomable amounts of digital evidence around crime scenes and pointing real-life sleuths in the direction of leads they hadn’t come across yet.
“The tyranny of language means you’ll describe it using a set of words, and someone somewhere will say that they already do this … but they aren’t.
“There is a very big capability gap in policing, and it’s all to do with management and information … There has been an explosion of information and data across society because technology is just so ubiquitous, and while we could access it, we have previously had no way to analyse it all efficiently and this changes that.”
Söze has been built for WAPF in partnership with a technology company called Modis on Microsoft’s Azure cloud computing infrastructure.
To describe it to the general public it talks in terms of crimes leaving digital footprints, which can be many and diverse. Things like social media posts, text messages, CCTV images, digital photographs and emails which together can build up a picture of a crime like a jigsaw puzzle, but which until now have been hugely time consuming and difficult to piece together.
Rather than this being a case of AI bots using higher intelligence than human detectives, Thomas says the system looks like being hugely beneficial in spotting patterns, linking human relationships and speeding up tasks significantly.
He uses an example of a recent investigation, which involved tracking messages and conversations accessed in investigations of an organised crime syndicate.
There were three different languages and a huge volume of chatter to sort through to get to the valuable information. Amid the normal ebb and flow of police work, Thomas estimated that it would typically take a year for analysts to work through, and get the relevant bits translated and understood in context.
Because Söze is built on a platform that incorporates Microsoft Azure’s translation tools, the conversations and messages could be understood to a sufficient standard to get the job done in five minutes.
“It’s a fairly substantial paradigm shift … in the past an investigator would find something and think ‘I wonder if that’s interesting,’ and then send it to get translated,” Thomas says.
“Then there would be some down time before the result would come back and some momentum could be lost. With Söze it is immediate and accurate enough for the purpose, because we can go back and listen to relevant parts of a conversation to confirm it for evidence.”
Anthony Doig, director of innovation for Modis and his colleague Jeremy Dennis, who is its national analytics lead, demonstrated the software, which has initially focused mainly on data derived from smartphones in the initial trial.
The software can build up a map of communications and evidence, showing all messages between devices, data on photographs, such as where they were taken and what device was used, based on the identifying EXIF metadata stored on each picture.
It provides detective insight into the data they already possessed, without necessarily realising the significance of it.
“We can start to very quickly visualise all of the people that are important within the investigation, by the way they are communicating. This could easily be expanded to include transfers of funds between bank accounts, or the location of people or vehicles,” Dennis says.
“For evidence purposes the system can identify where all of the imagery has been taken, and plot that on a map, so that the movement and meetings of people of interest can be shown globally and within Australia, proving that meetings have taken place.”
Thomas says he understands that technology like Söze will raise privacy concerns for some citizens, but says the software does not hoover up any data that the police does not already have the right to access.
He says he believes citizens would have any fears allayed if they actually saw the hurdles police had to go to in order to access certain data, and said there was certainly no scope for police to frivolously snoop on any citizens.
“There’s two ways of seeing this. You could see it as a very intrusive thing, or you could see it as police simply making very good use of the information we already have, lawfully, and that really what it goes to is public safety,” Thomas says.
“Our interest is solely in solving crime and improving public safety. We are looking to get to grips with the data that’s already available to the police and make better use of it. So it’s not about reaching out to find new sources.”
Thomas says the system is also not about putting human detectives out of work with an army of AI cops. He says that while there will be some cultural and technical lessons for cops to learn in dealing with the new system, the initial reaction has been one of enthusiasm for how much more effective it makes their work.
“Cutting straight to the chase, this won’t do away with people’s jobs … there are a very large quantity of human beings employed in police, who are occupied with data management tasks of one form or another,” he says.
“This will reduce the quantity of those data management tasks, and put those people onto what their intended, preferred activities are, which is information analysis.”
Thomas says he feels the initial trials have barely scratched the surface of what will be possible when the system is fully embedded and expanded to incorporate more types of data and evidence.
He admits some quizzical conversations have already occurred between WAPF and Modis about just how far the technology could evolve.
While it is a bit early to start talking about the ethical questions posed by movies like Minority Report where criminals are arrested before they have committed a crime, the idea that the system could identify patterns of behaviour and groupings of people to suggest trouble may be brewing is not off the table.
“Modis has already started thinking about this, and it’s essentially looking at common patterns that are predictive,” Thomas says.
“If you think about re-applying those patterns to ongoing situations then we are talking about the art of the possible. This is not part of the current brief, but they are very enthusiastic about things like this, because it’s never been done.”