In this week’s blog, Business Development Associate Max Margolis deep dives into Facial Recognition and emphasizes how this Machine Learning technology is assisting investigations and augmenting data analytics.
Facial Recognition is a powerful, new emerging technology that is being rapidly adopted across law enforcement agencies. But as with any new piece of software, it has its kinks, and numerous challenges that need to be dealt with. Praescient is well-versed in handling these challenges through years of experience working with other types of software in the industry. An example is using H2O Model Analyzer to minimize the known biases associated with FR. Praescient is prepared for these challenges, but is everyone else?
Breaking Down Facial Recognition
It is becoming clearer every day that data science and analysis are going to be the primary problem solving methods in the 21st century and beyond. At Praescient Analytics, catching the bad guys is what we do, and we use data to do so. But what happens when a crime occurs, and we need the data quickly? We can’t wait an entire day to receive data of a suspected crime, the suspect will be long gone. This is where facial recognition, or FR, software comes into play. Let’s say a crime occurs in a private company (ex: a casino). It would take tons of manpower, and time, to go through each individual aspect associated with the crime, identify suspects, and create leads for the investigation. However, if the casino embedded cameras implemented with FR software around the area, investigators would know exactly who the perpetrators were, almost instantly, as well as who they associated with and who they harmed. FR brings in expedited information, which in turn, leads to crimes being solved swiftly. Using an algorithm to pick out specific details of a person’s face, the software compares these unique details to a large facial recognition database of faces. Some FR software attempt to positively identify individuals directly, others calculate a ‘probability of match’ score between the suspect’s face and other faces in the database. Either way, the database being used is so massive, that whoever is employing the software will almost certainly find out everything about their subject, quickly.
Facial Recognition Shaping Current Events
During a Dallas Mavericks NBA playoff game, a 15 year old girl was taken from her family, and trafficked into Oklahoma City overnight. Surveillance footage showed the girl leaving with an unknown man, seemingly willing to police. However, her parents knew better. Dallas police could not help the parents, due to Texas Family Code prohibiting them from acting unless the circumstances appeared completely involuntary. So, the family instead worked with a nonprofit counter-trafficking initiative to find their daughter. The initiative used FR software on every known adult site, to see if the traffickers had posted any ads or videos of the girl. Within a few days, the initiative got a hit on an ad of a girl matching the missing girl’s face. The parents then confirmed that it was in fact their daughter. The initiative geolocated the ad and contacted local police. The girl was saved from a hotel room after being missing for 10 days. FR software saved this girl’s life. What in the past would have taken years to analyze, took only a few days to achieve. FR software brings speed to data analysis, and expedites the process of data collection and investigation. It offers the capability of almost real-time analysis, which in the case of this girl, was required to save her life. If FR software wasn’t used, by the time someone figured out she was in Oklahoma City (if they ever did), the girl would have almost certainly been gone, completely engulfed in the human trafficking system, never to be seen again by her family or loved ones.
The Controversies of Facial Recognition and Machine Learning
FR software is still far from being ubiquitous. In its current state, the software is extremely technical, mistake-prone, and not fully realized. To clarify, there are many different FR software applications one can purchase, which results in a wide array of issues to discuss that don’t apply to each piece of software. The average deployment cost of one of these applications is nearly $500,000 USD, which does not include the monthly maintenance costs to keep the software operational. A doable amount for many law enforcement firms and police departments, yet, these groups typically do not have staff with the specialized expertise required to operate this software. When you look at a company like Praescient, their analysts work with similar software on a regular basis, and possess numerous years of experience working in the field as well. Relaying the important findings from these types of software in simple terms to the inexperienced is an enormous part of the day-to-day life as an analyst at Praescient. Inexperience is one of the main reasons FR software is not in every police department or private business, but it only takes a few months training from someone like Praescient Analytics to dismantle this road block.
Besides necessitating experienced staff members to operate properly, FR software, as of now, is poor at identifying people of color, leading to numerous known cases of misidentification. Plus, the more faces that get added to the database used to search for matches, the less accurate the software gets, due to many people’s faces being very similar, adding more potential for misidentification. The software is also known to be biased toward darker-skinned people during the initial scanning of the database. Law enforcement does not want to lock up innocent individuals, but as of now, FR software isn’t helping in that matter. Praescient has proven capabilities tackling these biases through its use of H20 Analyzer. H20 Analyzer allows Praescient analysts to simulate what-if statements within the software in order to continuously explore and evaluate a predictive model’s behavior and limitations with changes in the real world. Analysts are able to probe machine learning models to discover and validate prior beliefs with minimal effort; this can illuminate potential biases in the model. Praescient has an answer to one of the key present issues with FR software at the moment, indicating that it’s only a matter of time before the other issues are dealt with as well.
As exemplified above, FR software also saves lives. It enables analysts and law enforcement to be proactive instead of reactive. Instead of waiting for an individual to engage in nefarious activities, and later analyze what happened after the fact, FR software brings the possibility of analyzing what is going on as soon as that person walks through the door. The future of security could look very different if FR software is further developed; and Praescient will be ready for whatever it brings.