COPKIT is developing data-driven policing technologies to support Law Enforcement Agencies (LEAs) in preventing, investigating, and mitigating crime and terrorism. In this blog, Thales Nederland summarises the project’s approach to identify standards and certification opportunities for AI technologies in the Fight against Crime and Terrorism (FCT) domain.
European Research projects developing new technologies are encouraged to take into account existing standards and certifications and to explore their values for project results, as well as the opportunities for standardisation and certification. These activities are encouraged for efficiency and, most importantly, to increase the impact of the project.
The COPKIT project is no exception and exploring the potential of and opportunities for standardisation and certification of its results has been an important aspect of the tools’ development in order to facilitate their use and evaluation by Law Enforcement Agencies (LEA) and to facilitate their uptake.
While the advantages of using standards and certification are clear, a good balance between flexibility and standardisation is critical for the COPKIT tools, as well as a careful assessment of the maturity of the various areas and of potential changes that might make specific implementations obsolete.
The COPKIT tools make use of a large range of relatively new and immature data types and technologies, such as Data science, AI, and Machine Learning, which are relatively recent and for most applications, no consensus on approaches or very few standards exist. Still, several standards (or dominating formats) have been identified and implemented in the COPKIT tools. Our analysis therefore extends the scope to de-facto standards and dominating data formats.
The COPKIT strategy has been to choose formats that balance popularity and practicality and that provide a flexible implementation. Beyond the format, agreed-upon semantics are required to be able to exchange data, especially in the case of (criminal) intelligence and knowledge.
Our analysis is not limited to technical aspects but encompasses processes, methodology, training etc., and is largely based on the responses of the project’s LEA partners to a questionnaire aimed at identifying their “pains” and areas in which the lack of standardised approaches negatively impacts the execution of their tasks.
To ensure an efficient exploration, the project used a bottom-up approach, with technical partners assessing the situation for each component, with a special focus on input / output formats as an enabler for interoperability. In parallel, LEAs provided inputs to identify their main priorities. The combination of these inputs resulted in a manageable list of focal points that could be subjected to in-depth analysis and the development of realistic actions.
COPKIT’s report on “Standardisation and Certification opportunities” (to be published soon on the EC website) summarises this analysis and serves as an exploratory overview of the state of maturity of formats and standards in (some areas of) AI. The report discusses the formats for AI models, or graph data and puts them in perspective with the balance between dynamic innovation, maturity, reusability, and interoperability.
Overview of the process followed to analyse opportunities for Standardisation and Certifications and produce action plans
Application of standards in the FCT domain
The analysis uncovered a particularly interesting insight regarding the representation of domain knowledge (ontologies, taxonomies, etc.) in the context of fighting crime and terrorism. The survey carried out by COPKIT confirmed that many H2020 projects carry out activities aiming at representing domain knowledge pertaining to specific types of crime or focusing on different angles (e.g., the criminal’s view or the investigator’s view).
While the developments generally do not overlap, the COPKIT project concludes that increased cross-exploitation of knowledge representation could benefit the applications in the FCT domain.
Among other advantages, reusing structured upper-level knowledge (beyond the type of crimes for which they were developed) and technical solutions (e.g., for continuous knowledge management and updating) would help to reduce the effort required to develop intelligent tools (the infamous knowledge acquisition bottleneck).
These shared resources are also an important step towards supporting investigations spanning across multiple criminal activities. For this reason, the COPKIT project, together with other H2020 projects (MAGNETO and PREVISION, among others), is currently exploring views on the increased re-use of developed domain knowledge representation, the means to improve the availability of the resources beyond the life of the projects and to manage the stakeholders’ ownership. The COPKIT project presented this approach during the MAGNETO final event (the presentations will be made available on the website of the MAGNETO project (http://www.magneto-h2020.eu/) shortly).
Share your perspective
Is your project carrying out activities to represent FCT domain knowledge? Are you interested in existing work on this topic? Do you believe that the re-use of FCT domain knowledge models can be improved? Please contact us at firstname.lastname@example.org!