We live in times when “intelligent machines” are not an element of science fiction, but a reality that surrounds us. AI/ML technologies are altering the way we exist, making our lives easier, rooting to our phones, cars, finances, and education, to name a few. Criminal justice is one of the fields where artificial intelligence is widely used to ultimately improve public safety. The ecosystem is continuously improving, allowing specialists to analyze criminal activity more profoundly. Thus, with AI/ML it becomes possible to analyze DNA, gunshots, and audio & visual materials faster and more efficiently.
Today, Robin Eklund, AI/ML expert at Sigma Technology Insight Solutions, tells us more about his current assignment at Griffeye, a software helping to solve the crimes in child sexual abuse cases. He explains how such technology helps in crime detection and how it works in practice.
Tell us about yourself and your professional background.
In 2017 I finished my Master’s degree in Applied Physics and majored in nanotechnology, specifically experimental nanoplasmonics, which is quite different from what I am working with today. I have been in the automotive industry at CEVT and Volvo AB for the last four years, working with data analysis/machine learning. I started my assignment at Griffeye this year as my second assignment at Sigma Technology.
What is your current assignment with Griffeye? Can you please share some details about the projectand your role there?
Griffeye develops software solutions to help law enforcement worldwide solve crimes in child sexual abuse cases. The product provides intelligence by processing a large volume of sensitive data in photos and videos and reducing the mental strain that the investigators are exposed to when faced with these cases.
My assignment at Griffeye is to develop AI and computer vision algorithms to help investigators collect evidence from images and videos. One of our projects in my team is to develop a robust video matching algorithm that finds similar clips in a video compared to already seen videos. It will benefit investigators to identify either a complete copy of an already seen video or shorter clips from a video compilation.
Robust video matching uses a particular kind of hashing algorithm called perceptual hashing that generates a hash, or fingerprint, of the frame based on visual features. It aims to give a quantifiable description of how our brain perceives an image. Then you can compare the hashes to determine if the frame is in the database of already seen frames. My role in this project is to improve the similarity search algorithm in the database. Since the database of known hashes is huge, we need to optimize the memory-speed-accuracy tradeoff by researching different nearest-neighbor implementations.
We also have several ongoing projects to aid the investigators in their daily work, like automated pixelation of genitalia and speech-to-text from videos so the investigators don’t have to listen to the sound, which can be very distressing to their mental health.
How does AI/ML help investigate crimes? Are there any successful cases?
The Griffeye Analyze platform provides several state-of-the-art technology solutions to investigators. However, today AI is not built to solve a case on its own but rather, in combination with different technologies, provides key intelligence to the investigation. For example, estimating a geographical position of an image based on certain objects in the photo, such as local brands and logos, street signs or determining which hotel chain the victim has been photographed in based on the particular interior design like wall hangings or bed sheets.
The main purpose of using AI in sexual abuse cases is to find the needle in the haystack and reduce investigators’ workload. For example, a photograph was found in seized material of a perpetrator that had his company logo on his shirt. It turned out that the company only had a handful of employees, and the suspect could be identified.
AI technologies are a scorching topic now. People bring the question of the moral side of AI. What do you think about it?
I share the concern about the future usage of AI and machine learning when we use it to solve problems that are not suitable for computers, such as juridical decisions. Although I believe with time, it will be more acceptable to provide personal data about ourselves, such as surveillance, in exchange for the promise of living in a safer society.
How do you think AI and ML technologies will be used for crime detection in the future?
AI and ML will play a vital role in future criminal investigations like DNA profiling is developed. The increasing rate at which we collect data today requires a tremendous number of working hours to process that manually, and that is what machine learning excels in. In other cases, the data is not in a human-readable format, such as in cybercrime detection, in which algorithms find deviations from ‘normal’ usage and can in real-time detect potential threats.
For sexual abuse cases, I believe that AI will, in the future, provide connected patterns between several data types, such as photos, text messages, bank transactions, etc., that we as humans would not realize without spending a very long time searching.