State of use of AI tax systems
The first publicly available mention of the use of tax machine-learning algorithms by the tax administration of the Republic of Slovenia dates back to 2017, when the administration integrated a risk-scoring algorithm developed by SAP.
What functions are performed with AI?
Based on publicly available data, the only function performed by a tax machine-learning algorithm is:
- External risk-management (risk-scoring): the tool predicts the risks of tax fraud or tax non-compliance associated with individual taxpayers, to subquently rank and (pre-)select taxpayers for further audits by tax officials. Lenart J. Kučić of AlgorithmWatch reports that the machine-learning model runs on approximately 158 risk-factors, derived from taxpayer historical data and new risk-factors are immediately transmitted to the system and to all tax officials.
There has also been incidental reports that the Slovenian tax administration is experimenting with nudging tools, which adapt the language of standard communication to taxpayers in order to nudge them towards compliance without the use of more coercive means of enforcement.
What data can be processed by these systems?
The data collected and processed for the development and use of the tax machine-learning algorithm has not yet been specified.
Are these systems regulated by specific norms?
The use of the tax machine-learning algorithm is not regulated by specific legal norms.
References:
- Lenart J. Kučic, ‘Slovenia’ – AlgorithmWatch ‘Automating Society 2020 Report’, available at:https://automatingsociety.algorithmwatch.org/report2020/slovenia/ - last accessed July 2022.
- Doerrenberg & Schmitz, ‘Tax compliance and information provision. A field experiment with small firms’ (2017), Journal of Behavioural Economics for Policy, vol. 1, issue 1, 47-54.