The €4.5 million ($5.1 million) project is called iBorderCtrl and it involves two stages. The first is a pre-screening step, in which people planning to arrive in the EU upload their passport, visa and proof of funds, then interact with a computer-animated border guard that asks questions and, via the user’s webcam, studies the traveler’s face for “micro-expressions” that might indicate lying.

The second stage is what happens at the airport – in cases of potentially risky travelers, their documents are reassessed, biometrics are taken again and their risk is recalculated.

The European Commission said the trials will initially take place in a lab environment, in order to help border guards get to grips with the system, and will later continue in “realistic conditions along the borders.”

According to CNN, the AI-powered virtual border guard will become more “skeptical” if it thinks the traveler is lying, after it asks questions about name, age, date of birth and the purpose of their trip.

So far, the system has only been tested on less than a few dozen people. According to the New Scientist, the success rate was only 76%, but the project staff hope that will improve to 85% through further testing.

“AI” technologies are supposed to become more accurate with use, but that depends on the data they are fed. If the data is unbalanced, it can lead to algorithms that are biased against certain people, sometimes along ethnicity and gender lines.

CNN reported that the tests will take place only on passengers who have given their consent to participate.