Computers are more accurate than humans at detecting faces

A study by ID R&D shows how humans have more difficulty identifying faces than computers that perform the same task. The precision of the machines reaches an error of 0%, essential to detect scams.

Finding out if a face is real or not may seem like a simple task, but it is not. A study of the facial and voice biometrics firm based on Artificial Intelligence (AI), R&D ID, belonging to Mitek Systems, shows that the ability of people to identify false faces presents more difficulties than computers.

Specifically, the margin of error in these tasks performed by humans stands at 30%, while in the case of computers it is 0%. In addition, humans also take longer than computers with a CPU core to detect fake faces, being 4.8 seconds for people and 0.5 seconds for machines.

The study ‘Human vs Machine: Can People Detect Counterfeits Better Than Artificial Intelligence?’ The abilities of people and machines are compared to detect the facial vitality that determines whether a face is real or not. For this, the most widely used impersonation techniques have been used, ranging from printed photographs to videos, digital images or 2D and 3D masks.

Warning, scroll to continue reading

The results of this study clearly show how computers are more efficient in the five techniques, achieving a 0% error rate in the 175,000 images and in all types of attacks.

However, people were found to be less precise in determining the veracity of the faces with each spoofing technique, and 30% errors occurred in printed photo forgeries, one of the easiest attack types for scammers.

However, in the tests carried out with a group of seventeen people, and not individually, who deliberated together whether the faces they were shown were real or not, the results were more accurate than when they did so individually. But even so, team performance didn’t outperform PC performance either.

Face recognition against fraud

The results of this study reinforce the commitment of financial services companies and other sectors that rely on automation technologies, especially for fraud detection. The time savings that allow human resources to be used more effectively in other areas of the consumer life cycle is just one of the many benefits it brings them, and it leads them to continue betting on these techniques.

However, there is something that machines that perform these tasks with such precision lack: adequate sensitivity. There is no doubt that computers have a great capacity to detect counterfeits, but an excess of sensitivity in detecting fraud can compromise the experience of real customers.

This is an area in which these systems still need to improve since many of the current facial recognition systems achieve a low rate of detection of scammers due to many real users being caught in the dynamics of identity verification.

But the ID R&D report also shows in the tests performed, the machines with Artificial intelligence They have only wrongly classified 1% of real faces as fake. A percentage that, in the case of people, rises to 18%, which shows that automation technology offers better results when it comes to keeping authentic users out of the fraud network.

As Alexey Khitrov, CEO of ID R&D, highlights, these results confirm the effectiveness of the technology to guarantee verification experiences and protection against scammers. “Companies can achieve fantastic efficiency using biometric systems”, says the person in charge for whom “there is still work to be done. Technologies must continue to find the balance between security and convenience, reducing fraud to the minimum possible while ensuring that users have a perfectly seamless experience. ”

Related Articles

Leave a Reply

Your email address will not be published. Required fields are marked *