Abstract
On January 1, 2017, the Taxation Modernization Act entered into force in Germany. It includes regulations on fully automated taxation procedures. In order to uphold the principle of investigation that characterizes German administrative law, a risk management system can be established by the tax authorities. The risk management system aims to detect risk-fraught cases in order to prevent tax evasion. Cases identified as risk-fraught by the system need to be checked manually by the responsible tax official. Although the technical details of risk management systems are kept secret, such systems are presumably based on artificial intelligence. If this is true, and especially if machine learning techniques are involved, this could lead to legally relevant problems. Examples from outside tax law show that fundamental errors may occur in AI-based risk assessments. Accordingly, the greatest challenge of using artificial intelligence in risk management systems is its control.
This is a preview of subscription content, log in via an institution to check access.
Access this chapter
Tax calculation will be finalised at checkout
Purchases are for personal use only
Notes
- 1.
Law of July 18, 2016 (BGBl I p. 1679); see also the draft bill dated February 3, 2016 of the Federal Government on the modernization of the taxation procedure, BT-Drs. 18/7457; recommended decision and report of the Finance Committee dated May 11, 2016, BT-Drs. 18/8434; second and third sessions of the Bundestag dated May 12, 2016, minutes of the plenary session of the Bundestag 18/170, pp. 16773C–16783D; approval of the Federal Council dated June 17, 2016, BR-Drs. 255/16.
- 2.
Tax Code in the version published on October 1, 2002 (BGBl. I p. 3866; 2003 I p. 61), last amended by Article 6 of the Law of July 18, 2017 (BGBl. I p. 2745).
- 3.
BT-Drs. 18/7457, pp. 48–49 and 69–70.
- 4.
- 5.
BT-Drs. 18/7457, p. 82.
- 6.
BT-Drs. 18/7457, p. 83.
- 7.
BT-Drs. 18/7457, p. 79.
- 8.
BT-Drs. 18/8434, p. 122.
- 9.
BT-Drs. 18/7457, p. 83.
- 10.
BT-Drs. 18/7457, p. 79.
- 11.
BT-Drs. 18/7457, p. 68.
- 12.
BT-Drs. 18/7457, p. 70.
- 13.
See also BT-Drs. 18/7457, p. 70.
- 14.
Cf. Article 108(3) Grundgesetz.
- 15.
Münch (2013), p. 2152.
- 16.
- 17.
See Krumm (2017), p. 2191, who also sees profiling covered by Article 4 no. 4 GDPR and by Section 88(5) AO.
- 18.
Regulation (EU) 2016/679 of the European Parliament and Council of April 27, 2016 on the protection of individuals with regard to the processing of personal data, on the free movement of such data, and on repealing Directive 95/46/EG (General Data Protection Regulation), O.J. L 119 dated May 4, 2016, pp. 1–88.
- 19.
Investigations into RMS based on Article 15 GDPR are impeded by Section 32c(1) no. 1 AO. See Krumm (2017), p. 2194.
- 20.
Krumm (2017), p. 2192.
- 21.
The increasing importance of machine learning is related to the increasing availability of large amounts of data; see Goodfellow et al. (2016), pp. 18 et seq.
- 22.
The deliberations in this section are based on Braun Binder (2019).
- 23.
See, for instance, Krumm (2017), p. 2191.
- 24.
- 25.
Neumann (2016), pp. 5–6.
- 26.
Kaplan (2016), pp. 1 et seq.
- 27.
A distinction can be made, for example, between supervised and unsupervised learning; see Ertel (2016), pp. 191 et seq.
- 28.
Stiemerling (2015), p. 763.
- 29.
Kaplan (2016), pp. 28 et seq.
- 30.
See paras 9 et seq. and 12 et seq.
- 31.
See State Parliament of Baden-Württemberg, communication of the State Government dated December 14, 2011, Drs. 15/1047, pp. 12 and 19.
- 32.
BT-Drs. 18/7457, p. 69.
- 33.
See BT-Drs. 18/7457, p. 70, according to which RMS should conduct more than just plausibility checks. See also Seer (2017), § 88 AO, para 76.
- 34.
Cf. see Buchholtz, paras 11, 24, 30. Risk assessment software has been in use in the US judiciary for some time and in various areas. For an overview, see Kehl et al. (2017).
- 35.
A sample questionnaire, including 137 questions, is available online at https://www.documentcloud.org/documents/2702103-Sample-Risk-Assessment-COMPAS-CORE.html. Accessed 30 September 2018.
- 36.
- 37.
- 38.
Burgess (2018).
- 39.
- 40.
See Knaus (2017).
- 41.
In a joint study, MIT and Stanford University, for example, were able to demonstrate that three different commercially developed computer programs for image recognition, each based on neural networks, systematically determined the gender of light-skinned men more precisely than that of dark-skinned persons and/or women. See Buell (2018).
- 42.
Cf. also paras 24 et seq.
- 43.
Cf. paras 9 et seq.
- 44.
Tutt (2017), pp. 101 et seq.
- 45.
Kirn and Hengstenberg-Müller (2014), pp. 228 et seq.
- 46.
Seer (2017), para 77.
- 47.
Cf. Bundesrechnungshof (Federal Court of Auditors) (2012), pp. 25 et seq.
- 48.
Martini and Nink (2017), p. 12.
- 49.
Martini and Nink (2017), p. 12.
- 50.
Cf. Tischbirek, passim and esp. paras 31 et seq.
- 51.
Cf. Wischmeyer, paras 44 et seq. See also the activities of algorithmwatch, https://algorithmwatch.org/de. Accessed on 30 September 2018.
References
Ahrendt C (2017) Alte Zöpfe neu geflochten – Das materielle Recht in der Hand von Programmierern. Neue Juristische Wochenschrift:537–540
Angwin J, Larson J, Mattu S, Kirchner L (2016) Machine bias – there’s software used across the country to predict future criminals. And it’s biased against blacks. https://www.propublica.org/article/machine-bias-risk-assessments-in-criminal-sentencing. Accessed 30 Sept 2018
Braun Binder N (2016) Ausschließlich automationsgestützt erlassene Steuerbescheide und Bekanntgabe durch Bereitstellung zum Datenabruf. DStZ:625–635
Braun Binder N (2019) Algorithmisch gesteuertes Risikomanagement in digitalisierten Besteuerungsverfahren. In: Unger S, von Ungern-Sternberg A (eds) Demokratie und Künstliche Intelligenz. Mohr Siebeck, Tübingen (in press). https://www.mohrsiebeck.com/buch/demokratie-und-kuenstliche-intelligenz-9783161581892
Buell S (2018) MIT researcher: artificial intelligence has a race problem, and we need to fix it. Boston Magazine, 23 February 2018. https://www.bostonmagazine.com/news/2018/02/23/artificial-intelligence-race-dark-skin-bias. Accessed 30 Sept 2018
Bundesrechnungshof (2012) Bericht nach § 99 BHO über den Vollzug der Steuergesetze, insbesondere im Arbeitnehmerbereich vom 17.1.2012
Burgess M (2018) UK police are using AI to inform custodial decisions – but it could be discriminating against the poor. WIRED, 1 March 2018. http://www.wired.co.uk/article/police-ai-uk-durham-hart-checkpoint-algorithm-edit. Accessed 30 Sept 2018
Ertel W (2016) Grundkurs Künstliche Intelligenz, 4th edn. Springer Vieweg, Wiesbaden
Flores AW, Bechtel K, Lowenkamp CF (2016) False positives, false negatives, and false analyses. http://www.uscourts.gov/federal-probation-journal/2016/09/false-positives-false-negatives-and-false-analyses-rejoinder. Accessed 30 Sept 2018
Goodfellow I, Bengio Y, Courville A (2016) Deep learning. The MIT Press, Cambridge
Haunhorst S (2010) Risikomanagement in der Finanzverwaltung – ein Fall für die Finanzgerichte? DStR:2105–2110
Kaplan J (2016) Artificial intelligence. What everyone needs to know. Oxford University Press, Oxford
Kehl D, Guo P, Kessler S (2017) Algorithms in the criminal justice system: assessing the use of risk assessments in sentencing. Responsive Communities. https://dash.harvard.edu/handle/1/33746041. Accessed 30 Sept 2018
Kirn S, Hengstenberg-Müller CD (2014) Intelligente (Software-)Agenten: Von der Automatisierung zur Autonomie? Verselbständigung technischer Systeme. MMR:225–232
Kleinberg J, Mullainathan S, Raghavan M (2016) Inherent trade-offs in the fair determination of risk scores. https://arxiv.org/abs/1609.05807. Accessed 30 Sept 2018
Knaus C (2017) Internal Centrelink records reveal flaws behind debt recovery system. The Guardian, 13 January 2017. https://www.theguardian.com/australia-news/2017/jan/13/internal-centrelink-records-reveal-flaws-behind-debt-recovery-system. Accessed 30 Sept 2018
Krumm M (2017) Grundfragen des steuerlichen Datenverarbeitungsrechts. DB:2182–2195
Martini M, Nink D (2017) Wenn Maschinen entscheiden… – vollautomatisierte Verwaltungsverfahren und der Persönlichkeitsschutz. NVwZ-extra 10:1–14. http://rsw.beck.de/rsw/upload/NVwZ/NVwZ-Extra_2017_10.pdf. Accessed 30 Sept 2018
Münch L (2013) Außergerichtlicher Rechtsschutz bei Korrekturen gemäß § 129 AO und § 173 AO nach faktischer Selbstveranlagung. DStR:2150–2156
Neumann L (2016) Stellungnahme des Chaos Computer Club. 13 April 2016. http://goo.gl/2Qm2uR. Accessed 30 Sept 2018
Oswald M, Grace J, Urwin S, Barnes G (2017) Algorithmic risk assessment policing models: lessons from the Durham HART Model and ‘Experimental’ Proportionality. https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3029345. Accessed 30 Sept 2018
Pasquale F (2017) Secret algorithms threaten the rule of law. MIT Technology Review. https://www.technologyreview.com/s/608011/secret-algorithms-threaten-the-rule-of-law. Accessed 30 Sept 2018
Präsident des Bundesrechnungshofes in seiner Funktion als Bundesbeauftragter für Wirtschaftlichkeit in der Verwaltung (2006) Probleme beim Vollzug der Steuergesetze. https://goo.gl/92gu6r. Accessed 30 Sept 2018
Schmidt E (2008) Moderne Steuerungssysteme im Steuervollzug. DStJG 2008:37–57
Seer R (2017) § 88 AO. In: Tipke K, Kruse HW (eds) AO/FGO. Kommentar, 150th edn. Otto Schmidt, Köln
Stevenson (2017) Assessing risk assessment in action. George Mason University Legal Studies Research Paper Series, LS 17–25. https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3016088. Accessed 30 Sept 2018
Stiemerling O (2015) “Künstliche Intelligenz” – Automatisierung geistiger Arbeit, Big Data und das Internet der Dinge. CR:762–765
Tutt A (2017) An FDA for algorithms. Adm Law Rev 69:83–123
Urwin S (2018) Durham Constabulary written evidence submitted to the Common Science &Technology Committee inquiry into algorithms in decision-making. 20 February 2018. http://data.parliament.uk/writtenevidence/committeeevidence.svc/evidencedocument/science-and-technology-committee/algorithms-in-decisionmaking/written/78290.pdf. Accessed 30 Sept 2018
Author information
Authors and Affiliations
Corresponding author
Editor information
Editors and Affiliations
Rights and permissions
Copyright information
© 2020 Springer Nature Switzerland AG
About this chapter
Cite this chapter
Braun Binder, N. (2020). Artificial Intelligence and Taxation: Risk Management in Fully Automated Taxation Procedures. In: Wischmeyer, T., Rademacher, T. (eds) Regulating Artificial Intelligence. Springer, Cham. https://doi.org/10.1007/978-3-030-32361-5_13
Download citation
DOI: https://doi.org/10.1007/978-3-030-32361-5_13
Published:
Publisher Name: Springer, Cham
Print ISBN: 978-3-030-32360-8
Online ISBN: 978-3-030-32361-5
eBook Packages: Law and CriminologyLaw and Criminology (R0)