iBorderCtrl: Automated decision making at the EU border
The EU plans to test an artificial-intelligence-based assessment approach for border control in Greece, Hungary and Latvia.
People travelling to the EU will interact with an automated system, answer a set of questions and their interaction with the system will be captured on video and automatically analysed. Micro expressions are used to determine whether a person is lying. On this basis the artificial intelligence based lie detector will assess if a traveller may enter the EU or if further investigation is necessary.
This project raised concerns within the Data Traction team for multiple reasons:
- Methodology and evaluation results:
The approach is based on micro expressions – brief, involuntary facial expressions. These micro expressions are detected in the video data and passed on to a machine learning based approach to determine whether a person is telling the truth.
However, experts in psychology, such as Bruno Verschuere, a senior lecturer in forensic psychology at the University of Amsterdam, state that there is no scientific evidence for the methods employed – micro expressions are not able to express whether a person is lying.
For the sake of the argument, let us briefly forget about this assessment by an expert in psychology and let us assume a liar could be indeed detected using micro expressions. In this case we wonder, how the micro expression detection and recognition process was designed and evaluated. Previous computer vision research for the recognition of micro expressions seems to be highly dependent on constraints regarding the image acquisition – data samples are captured under fixed lighting conditions and for fixed head positions. Moreover, previous experiments report differences in recognition rates depending on racial/ethnic affinity of test subjects [6]. For this reason, high diversity in the trainings dataset that is used to derive the employed machine learning approach plays a critical role.
Early experiments in this project report a 76 percent success rate. Considering that this rate was achieved in controlled experiments for a very small test set of 32 individuals, this is a poor result, in particular when considering that results achieved in controlled experiments almost always outperform the results achieved in an unconstrained, real life environment.
The scientists involved in the project stated that they are confident to improve the system and to achieve an 85 percent success rate. But is this enough? In 2015 more than 520.000 people were granted a Schengen visa and thus with a high likelihood crossed an EU border. Considering this group of travellers nearly 80.000 of these people would have been wrongly detected as liars under an 85 percent success rate. - Automated decision making
Reservations about this project are met by the counter argument that the artificial- intelligence-based lie detector is only one stage in a border crossing process. If a liar is detected by the automatic system, the respective traveller is in a next step questioned by a border control officer. Potentially adding a human in the loop should rectify errors of the automated process.
The automated system may deem a traveller safe to enter the EU or may pass the traveller on to a border control officer for further processing. In both cases an automated decision is made, either for the grant to enter the EU or for further processing. - Collection of biometric data
In the case of further processing additional data of the traveller is considered in the risk assessment. This data includes biometric data such as fingerprints, facial features and measurements or palm vein scans. Biometric data is considered a special category of personal data according to the GDPR. The European Union Agency for fundamental rights published a report on the data collection and the interoperability of EU information systems. Data protection advocates expressed concerns regarding the EU-wide collection and processing of biometric data.
sources:
[1] The Verge | The EU plans to test an AI lie detector at border points
[2] The Guardian | EU border ‘lie detector’ system criticised as pseudoscience
[3] Gizmodo | An AI Lie Detector Is Going to Start Questioning Travelers in the EU
[4] orf.at | Umstrittenes Projekt Lügendetektor für EU-Grenze geplant
[5] European Union Agency for Fundamental Rights | Under watchful eyes – biometrics, EU IT-systems and fundamental rights
[6] ACII 2011 International Conference on Affective Computing and Intelligent Interaction | The Machine Knows What You Are Hiding: An Automatic Micro-expression Recognition System
[7]Eidgenössisches Justiz – und Polizeidepartement | Visa Monitoring