New AI technology trained to aid child abuse investigations
Today, Griffeye release new AI technology for Analyze DI Pro that will help investigators filter out relevant information, speed up investigations and free up time to prioritize investigative work such as victim identification. The technology has been trained on child sexual abuse material at Taskforce Argos, Queensland Police in Australia.
The Griffeye AI program, called Griffeye Brain, aims to innovate how AI is applied to law enforcement work processes. The first outcome of Griffeye Brain is a CSA classifier that scans through previously unseen footage and suggests images that it believes depicts child sexual abuse content. The AI outputs a score that can be used to tell whether a file is pertinent to the investigation or not.
“These first tests show very promising results”, said Johann Hofmann, Director and Head of Griffeye. “They show that AI performs well in some of the toughest application and that it has great potential. However, the results from the AI technology doesn’t by any means take the human investigator out of the loop. Instead, it aids the investigator in making qualified assessments. What it does do is automatically group and filter material and it helps investigators prioritize.”
“The next step is to see how the technology works in different settings”, he continued. “We know it works on live data managed within Taskforce Argos’ database in Queensland Australia, but we still need to find out how it works on data in other countries that is possibly classified based on other jurisdictions. We hope to work with our users to continue develop the technology and improve it even further.”
The version released today is the first beta version, that will be further developed and fine-tuned in several stages during 2018.
A help to prioritize information
AI technology opens up whole new possibilities for investigators. One of the big challenges that law enforcement face today are the large volumes of data that investigators have to deal with. Finding clues is often like looking for a needle in a haystack, and information is often found too late, or not at all.
“Technology can help solve some of those problems and our hope is that the AI technology produced in the Griffeye Brains programme will provide important relief to strained investigators. Instead of getting bogged down in trying to manually go through the data, it will help them prioritize and free up time to spend time on analysis and solving crime”, said Hofmann.
Trained on quality-assured data
The Griffeye Brain CSA classifier has been trained on real CSA case data at Taskforce Argos, Queensland Police in Australia. Training the algorithm on real data is necessary for it to work and it is absolutely crucial how the training is conducted and what data it is being trained on.
“The performance of the AI algorithm is directly related to the quality of the data that it is trained on. Taskforce Argos’ database is one of the best, quality assured databases in the world”, said Hofmann.
Jon Rouse, Detective Inspector at Taskforce Argos, said that AI brings new and interesting opportunities to law enforcement:
”Harnessing technological evolutions like AI is the only way forward. We have spoken previously about the spiralling volume of child exploitation material that is being seized and intercepted by LEA globally and now with some sectors of industry self-reporting we are also looking for new and innovative ways to risk assess referrals and prioritize our active investigations.”
“Beta testing of the AI module has demonstrated that computers can remove, with a high level of accuracy, data that investigators do not need to focus on,” Rouse continued. “The finite resources we have for victim identification can be centered on the images and video that really matters, with an end goal – identify and remove from harm more child victims faster.”
The first step of many
The version released today is a beta version of Analyze DI Pro. A first official release is planned for August when the module will support GPU instead of CPU to speed up data ingestion significantly. Following releases will also include support for video and various ways to interact with the output of AI. A refined version of the beta release will also be integrated into Analyze DI Core during 2018.
“The technology can, of course, be trained on other types of data as well. Further down the line, we see possibilities in applying AI to help investigators reach better results in a variety crime types”, said Johann Hofmann.