Ref. Ares(2022)8519438 - 08/12/2022
eu-LISA Working Group on Artificial Intel igence
Workshop – prioritisation of AI use-cases

Short stay visa issuance
2

Identification of irregular travelling patterns – use case description
Problem statement
To avoid suspicion, some travellers take convoluted routes to avoid attention from authorities (e.g. going
from Egypt to Belgium through Japan). Due to the complexity, these routes are not always identifiable by
the authorities. This can lead to TCNs entering the Schengen Zone without having undergone appropriate
assessment regarding risks such as overstaying.
Description of the use case
AI base approach to monitor, search and combine data from dif erent sources such as ETIAS and EES
(but also PNR data collected by airlines) to detect possible “irregular travelling patterns. The AI could
detect irregularity either at the 'node' level of a sequence of stops ("is it uncommon to travel from point A to
point B?") or at the aggregate travel pattern level ("does the overall length and shape of this travel pattern
appear to be dif erent to usual behaviour?"). The outputs of this analysis could either prompt a human to
investigate further, or feed into an “Engagement” based AI solution to ask the applicant for further
information/documentation.
Benefits
Better insight into irregular travel patterns; Faster identification of possible high-risk individuals regardless
of the country of departure; Systematic check for possible irregular travel patterns which can prompt
further investigation.
3

Questions for discussion
• Do you agree that this particular use case should be prioritised?
• If not, which of the use-cases would you prioritise?
• Are there additional use-cases that you might propose at this stage?
• What are the main reasons to prioritise this particular use case?
• Would you be interested in developing/piloting this use-case?
• How would you see this in practice, what would be needed in this respect? What
would be the preconditions/prerequisites for the development/piloting of this use
case?
• What are the main limiting factors/constraints?
• Would you be wil ing to collaborate in a pilot with authorities from other MS? If
no, why?
• Would EU funding be necessary to develop/pilot such a use case?
• Do you see an active role for eu-LISA in piloting such a use case? Other
Agencies?
• If yes, what role could eu-LISA play? Other Agencies?
4

Issuance of long-stay or residence permits in the Schengen Area
5

Automatic triaging of applications – use case description
Problem statement
Requests for long-term stay or residence are high and rising in volume and variety, and it is a complex
task to appropriately assess categories of cases as it requires certain "local knowledge" (e.g. documents
evidencing family links or professional qualifications from India compared to Egypt). Furthermore,
governments are struggling to keep up with the number of requests and timely response regarding the
enquiry.
Description of the use case
An AI triaging system could be implemented to automatically and rapidly identify standard applications and
classifying more complex applications for human review. Classifying means grouping similar applications
(e.g. from a certain country or reason for travel) for review by an expert experienced at dealing with a
certain type of application. By grouping the applicants, the responsible officers would take save time in the
‘basic’ ones, and would only spend more time, in those that are classified as complex.
Benefits
Increase in the efficiency and accuracy of the process; improve the user experience for applicants;
decrease the dependency on a human assessor to classify applications.
6

Asylum
8

Vulnerability assessment – use case description
Problem statement
The first step when the authorities come in contact with undocumented migrants at border crossings is to perform a 'vulnerability check'.
This is to judge whether a person is in immediate danger or has special needs, and this requires expertise and experience. There may
be times where the individual does not want to provide the relevant information, for example if they are fearful. The process being quite
subjective sometimes is a challenge in itself, but also the capacity of these experts (e.g. trained social workers or caseworkers) is a
constraint on this sub-process which would be mitigated with ef iciency improvements.
Description of the use case
AI to perform real-time analysis of an applicant's facial movements, spoken language and body language (plus any other potential
aspects) to detect signals which can better inform decision-making by a human social worker/specialised expert (e.g. if the person
should be granted special procedural guarantees). Techniques would seek to notice and assess the emotional cues implied by both what
the applicant says/does and the way that they do it, either in terms of modelling apparent emotion types or by detecting fluctuating or
unusual behaviour. This latter approach, where the solution would be used to detect deviations from a baseline, as opposed to directly
detecting cues of vulnerability, is the more likely and feasible method for usage. the asylum context, applicants commonly experience
negative emotions when recal ing traumatic past experiences. An intel igent AI seeking to identify specific issues such as
torture/traf icking (possibly by attempting to link emotional cues with corresponding usage/sequencing of specific words and phrases)
would be a more advanced use case than one providing more general information (identified and quantified emotional signals) or more
generic signs of distress or abnormal patterns in communication.
Benefits
• Reduced capacity pressures on the trained social workers/experts who are usual y required to make these kind of assessments.
• An additional method for ensuring vulnerable asylum seekers receive the appropriate procedural guarantees.
9

External processes (MS): SIS/SIRENE
11

Automatic form completion – use case description
Problem statement
Fil ing in SIRENE forms requires to be done in a consistent and correct way, to ensure that the communication between
Member States happens smoothly. The officer needs to choose which form to use and then fil it with the correct
information. However, in practice it is observed that not only the officers use the wrong forms, but also that these are
not fil ed in accurately and require a lot of human effort and further interaction to correct them.
Description of the use case
A SIRENE form can be completed using the information from the original alert alongside a report from an officer
(highlighting the action taken). An AI system could be used to automatically recognise key information from the alert,
and to collect and structure information coming from an officer’s report. Specifically, using natural language processing
(NLP), the system would work as a virtual assistant to interact with the officer and gather the necessary information to
automatically complete the SIRENE form determined to be correct. Then, this information would be sent to the other
Member State (perhaps with automatic translation to the correct new language).
The technical approach would use techniques like entity recognition and ‘spellcheck’-type functionalities as well as a
chatbot interface, as described in VISA-1 and VISA-9.
Benefits
Ensure the use of the right form for a specific alert; Grant correct fil ing of the forms, improving the communication
between Member States; Reduce the number of interactions needed between Member States.
12

Generic: transversal processes
1
2
14

Identification of fraudulent supporting documents – use case
description
Problem statement
The process to detect fraudulent documents during a risk assessment is performed manually by a case worker,
requiring significant expertise and experience. However, the manual process is subject to human error and can
be very time-consuming (especially if reliant upon very specific local knowledge of certain documentation types).
Description of the use case
AI to detect forged supporting documents such as birth certificates or bank statements (note: this use case excludes travel passports).
The model would attempt to identify documents, which appear fraudulent in various ways: for e.g. if the layout/format of a document is
not of the expected type, if the contents of the document are not internal y consistent, if the contents of the document are not consistent
with other information provided. Examples of these types could be misplaced logos and sections (with respect to known examples),
mismatched data in dif erent locations (such as two dif erent spel ings), mismatched data in dif erent documents (such as mentioning a
dif erent education background in a CV vs. in an education qualification document).
To analyse documents in this way, text-based models and image-based models could be used together. The model would seek to flag
potential cases for human review, for example by highlighting which areas of documents should be investigated further. Advanced
approaches here could include investigating the activations in a Generative Adversarial Networks (focused on the discriminator
component) to see which pixels of an input image look the ‘most fake’ (i.e. contributed most significantly to the classification as fake)
Benefits
• Improved identification of fraudulent documents irrespective of expertise or experience by means of ensuring
minimum quality level performed by an AI engine;
• Faster identification of fraudulent documents freeing up time for more value-added tasks by case workers.
15

Detection of forged travel documents – use case description
Problem statement
Border guards are trained to identify stolen or forged documents The references of stolen travel documents are reported in
databases like SIS operated within the EU and SLTD (Stolen and Lost Travel Documents) operated by Interpol. However
identifying forged documents requires expertise and experience, as the border guards must know the documents from all
around the world and be able to distinguish a forged/falsified document from an original. Moreover, humans are subject to error
and therefore it is possible that a traveller is able to pass the border with a forged/falsified travel document. This is even more
problematic when considering the potential inconsistency in decisions made by dif erent officers. Specifically, this could result in
a situation where two forged documents are treated dif erently for the
same procedure (request for a permit, border crossing) and where one officer sees that the documents are forged and the
other one does not.
Description of the use case
By using computer vision, an AI system can detect the use of forged travel documents. To do so, it would analyse the captured
image of the provided documents and assess if the physical characteristics of the document matches an original one, if the
information provided in the documents is accurate, and the person providing the documents corresponds to the person in the
document and is not a lookalike.
The technical approach would be similar to CROSS-6 which is focused on supporting documents.
Benefits
Increase the security in the Schengen Area; Reduce the dependency on the officers’ expertise and experience; increase the
consistency of decision making.
17

Other use-cases suggested by the respondents
• Transcription (speech to text)
• Categorization of criminal data for further analysis
• Pattern identification in visa application supporting documents
(including pattern identification based on document numbers and
issuing authority)
• Identification of travel documents produced using stolen blanks
• Forensic analysis of graphic/video evidence (e.g., child sexual
abuse) using AI tools
19

Questions for discussion
• In addition to the use cases mentioned above would you like to suggest any other use
cases?
• Would you prioritise the development of the use cases suggested above in addition to the
ones discussed in the study?
• What are the main reasons to prioritise this particular use case?
• Would you be interested in developing/piloting this use-case?
• How would you see this in practice, what would be needed in this respect? What would
be the preconditions/prerequisites for the development/piloting of this use case?
• What are the main limiting factors/constraints?
• Would you be wil ing to collaborate in a pilot with authorities from other MS? If no, why?
• Would EU funding be necessary to develop/pilot such a use case?
• Do you see an active role for eu-LISA in piloting such a use case? Other Agencies?
• If yes, what role could eu-LISA play? Other Agencies?
20

Thank you !
eu-LISA
European Union Agency for the Operational
Management of Large-Scale IT Systems in
the Area of Freedom, Security and Justice
facebook.com/agencyeulisa/
www.eulisa.europa.eu
twitter.com/EULISA_agency
linkedin.com/company/eu-lisa/
youtube.com/c/euLISAagency

Document Outline