Machine learning and biometrics: How AI is becoming more humanMachine learning and biometrics: How AI is becoming more human
Machine learning and biometrics: How AI is becoming more human
December 16, 2019
by Susan Randford 16 December 2019
"[AI] is going to change the world more than anything in the history of mankind. More than electricity." - AI specialist and investor Dr. Kai-Fu Lee, 2018
Artificial Intelligence (AI) tools are becoming better at understanding humans by the day. Behavioral biometrics is a prime example of this new approach to technology.
What is behavioral biometrics?
Behavioral biometrics is used as an identification mechanism: it identifies people according to how they interact with online applications and devices. It works passively in the background, making account credentials impossible to steal or duplicate, unlike passwords. It is principally used to block the use of synthetic or stolen identifications while applying for credit online, as well as blocking account hacking.
Artificial Intelligence (AI), Deep learning (DL), and Machine Learning (ML) are all associated with behavioral biometrics. The behavioral biometric technologies of today can obtain more than 2,000 parameters from a mobile device alone. Some of these parameters include:
The pressure someone uses when they type;
The way a person holds the phone;
The way they scroll and toggle between fields;
How a person responds to different stimuli that have been presented by online applications.
Today's behavioral biometrics
The use of behavioral biometrics has been expanded to identity proofing in order to present a unique dimension for judging the validity of online applications in light of the extreme data breaches, as well as enabling risk-based authentication in payment apps.
Physical biometrics technologies such as iris, finger, and facial recognition are based primarily on a static procedure of measuring points obtained from fixed images. Meanwhile, behavioral biometrics is governed by a dynamic approach driven by machine learning and deep learning, which involves gathering and processing very large data sets.
Using primary device sensors such as accelerometer, gyroscope, and touch, hundreds, possibly thousands of behavioral patterns can be used to authenticate users continuously. The tap duration, fingerprint area, device acceleration, swipe speed, and session duration are among some of the behavioral parameters that are captured and profiled.
Machine learning and behavioral biometrics
Machine learning makes it conceivable to drive the decision-making processes that are needed to maintain a large number of parameters and data sets that must be analyzed. It is being utilized in all aspects of behavioral biometrics. ML is capable of learning from human behavior and continuously enhancing user profiles that can be used to authenticate sessions or transactions.
For example, during a bank transfer, behavioral biometrics can analyze the keystrokes by looking at type speed and which fingers are being used to type, and within 10 minutes, it can create a profile that is strong enough to validate a user.
However, as time goes on, and a person uses the device more often, their behavior changes and adapts. Machine learning aids in breaking through the confusion of different signals and locates the consistencies in the behavioral patterns over time, regardless of the changes.
Today, 100% of the fraud comes from authenticated sessions and happens when a legitimate user logs in, but the account is taken over by malware, bots, social engineering, and other types of remote access attacks.
In the beginning, AI-led behavioral biometrics was used for account hacking prevention in financial services. However, today, behavioral biometrics has the ability to provide passive, continuous, and frictionless authentication. Over the next five to eight years, machine learning and behavioral biometrics is set to restructure the authentication landscape.