AI Centre of Excellence
In our previous issue we talked about Artificial Intelligence (AI) initiatives at eu-LISA – now let's take a look at the broader context.
While there is agreement around the idea that AI should become a practical support tool for law enforcement practitioners, helping bolster Europe's internal security, an important question has so far remained unanswered:
How do you find the right balance between AI and data protection, data privacy and fundamental rights?
In April 2021, the European Commission published a 'package' concerning AI, seeking to address some of these concerns, but also laying down a set of harmonised rules and amending legislation. The AI package is of particular interest to eu-LISA, as it covers the areas of law enforcement, border management, migration and asylum.
According to the reviewed Coordinated Plan on AI (started in 2018), the Commission intends to focus on funding networks of AI centres of excellence and to set up a European Partnership on AI, Data and Robotics under the Horizon Europe research project.
The Commission's proposal also covers restrictions on the use of remote biometric identification systems and, based on a proportionate risk-based approach, breaks down AI applications/ systems into the following categories:
- Minimal risk: not regulated;
- With specific transparency obligations: humans must be notified they are interacting with AI systems;
- High-risk: mandatory requirements to ensure the AI systems will be trustworthy and will protect citizens’ fundamental rights;
- Prohibited: any kind of action resulting in physical/ psychological harm, social scoring, remote biometric identification in public spaces.
According to the proposed regulation, EU Agencies, such as eu-LISA, can have their own regulatory sandbox, that would allow the development and testing of AI systems in an environment completely separated from the systems they already have in use.
Similar to parameters and clauses foreseen in the field of data protection, the European Data Protection Supervisor EDPS shall act as the market surveillance authority, having the right to impose administrative fines on EU Institutions, Agencies and Bodies for infringements or non-compliance.
eu-LISA has also been active in the domain this year. In a workshop on AI and law enforcement that took place in April, eu-LISA's top management stressed that, ultimately, AI is just a tool that helps human decision-making and does not replace it.
While it is a well-known fact that humans have biases, and so will the algorithms created by them, there are a number of measures that can be taken in order to reduce those biases, namely:
- Training the algorithms with big large data-sets and ensuring the quality of the data: AI algorithms are only as reliable as the data on which they are trained; training using synthetic data or small data-sets could yield higher error rates;
- Avoid black-boxes: the basis for the AI algorithms' decision making needs to be apparent and can be subject to external audit; this is also important for the end-users to understand and ultimately trust the algorithmic decision proposals;
- Continuous training of the AI algorithms: AI is not just a simple software tool, its real nature must be recognised and taken into account by constantly teaching AI to act in a responsible manner;
- Training the people who will be using the algorithms: by assessing the decisions that law enforcement officers make on the basis of AI suggestions, it will be possible to assess those decisions and any inherent biases in their decision-making are identified and addressed.
In May, the Agency hosted the first meeting of a new Working Group on Artificial Intelligence, that will identify use cases for implementing AI solutions in the large-scale IT systems entrusted to eu-LISA, to maximise the added value of the services the Agency provides to its stakeholders.
“AI creates many opportunities to enhance and strengthen internal security, border management, migration management and to increase efficiency of practitioners”, said eu-LISA Executive Director Krum Garkov during the meeting. “The Working Group will make a direct contribution to the implementation of the new Security Union strategy, and will help the EU to develop and keep its strategic independence in the area of AI.” Read more
The November edition of our Industry Roundtable also focused on AI and its many implications. Read more
Early next year, eu-LISA and fellow Agency Eurojust will publish a joint report on AI supporting cross-border cooperation in criminal justice.