The Artificial Intelligence Act in an actuarial context
Abstract
The report deals with questions relating to anonymization and pseudonymization procedures and their application and evaluation in the context of actuarial tasks.
Motivated by the legal framework and the risks of disclosing sensitive data, tools for anonymization are presented. The report presents a procedural model for the anonymization of data that is usually tabular in the insurance industry. In addition to the actual anonymization procedures, it is important to understand how these methods influence anonymity and the associated loss of information and data utility. When publishing data, it should not be possible for potential attackers to disclose the protected information. Concepts for generating synthetic data are introduced as a further option for anonymization. In addition to the simple and unrelated examples included in the report, separate notebooks in the statistical software R and in Python are presented.
Reports on findings are summaries of the results of work carried out by DAV committees or working groups,
- where their application can be freely decided upon within the framework of the code of conduct,
- that should inform discussion of the current opinion among actuaries or also among the broader public.
As working results of a single committee, they do not, for the time being, represent any recognised position within the DAV and do not comprise any actuarial standards of practice. In this respect they are clearly distinguishable from any standards of practice.
