Abstract
Public organisations have adopted AI into their public service aiming to tap into the promised potential for society, such as increasing efficiency and effectiveness of current processes. Recent studies from the European Commission share, however, that critical issues of AI use only tended to surface when they were already in operation and thus had already affected citizens. To prevent negative impact to citizens, we propose public organisations to use random sampling as a safe, yet valuable practical evaluation step before considering a pilot. This safe pre-pilot evaluation step enables evaluation of the AI system without applying it in any decisions or actions that already affect citizens. We pose six arguments on the added value of random sampling in the evaluation step of AI systems: 1) it provides high quality data for evaluation and validation of assumptions; 2) it supports gathering input for fairness evaluation; 3) it creates a benchmark to compare AI to alternatives; 4) it enables challenging assumptions in the organisation and the AI development; 5) it supports a discussion on the limitations of AI 6) and it provides a safe space to evaluate and reflect. In addition, we discuss limitations and challenges for random sampling in the evaluation, such as temporary loss of efficiency, class and representation imbalances, organizational hesitancy and societal experiences. We invite the participants of this workshop to reflect with us on the potential benefits and challenges, and in turn distill the practical requirements where using a random sample for evaluation is safe and useful.
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
Similar content being viewed by others
References
Bethlehem, J.: Applied Survey Methods, a statistical perspective. John Wiley and Sons Inc (2009)
Clemmensen, L., Kjærsgaard, R.: Data representativity for machine learning and AI systems (2022)
Dannenberg, E.: Factsheet overtredingen van de inlichtingenplicht - meer maatwerk en eenvoudigere regels (2021). https://www.divosa.nl/publicaties/factsheet-overtredingen-van-de-inlichtingenplicht/factsheet-overtredingen-van-de
EUR-Lex Access to European Union Law: glossary proportionality. https://eur-lex.europa.eu/EN/legal-content/glossary/principle-of-proportionality.html
European Parliament: amendments adopted by the European parliament on 14 June 2023 on the proposal for a regulation of the European parliament and of the council on laying down harmonised rules on artificial intelligence (artificial intelligence act) and amending certain union legislative acts (2023). https://www.europarl.europa.eu/doceo/document/TA-9-2023-0236_EN.html
Gerards, J., Schäfer, M., Muis, I., Vankan, A.: Fundamental rights and algorithms impact assessment (FRAIA). Utrecht University, Tech. rep. (2021)
High-Level Expert Group on Artificial Intelligence: ethics guidelines for trustworthy AI. Tech. rep, European Commission (2019)
Hoekstra, M., Chideock, C., Veenstra, A.: Quick scan AI in de publieke dienstverlening ii. Tech. rep. (2021)
Martínez-Plumed, F., et al.: CRISP-DM twenty years later: from data mining processes to data science trajectories. IEEE Trans. Knowl. Data Eng. 33(8), 3048–3061 (2021). https://doi.org/10.1109/TKDE.2019.2962680
Mehrabi, N., Morstatter, F., Saxena, N., Lerman, K., Galstyan, A.: A survey on bias and fairness in machine learning. ACM Comput. Surv. (CSUR) 54(6), 1–35 (2021)
Molinari, F., Van Noordt, C., Vaccari, L., Pignatelli, F., Tangi, L.: AI watch beyond pilots: sustainable implementation of AI in public services (KJ-NA-30868-EN-N (online)), 14 (2021). https://doi.org/10.2760/440212(online)
Reventlow, N.J.: Data collection is not the solution for Europe’s racism problem (2020). https://www.aljazeera.com/opinions/2020/7/29/data-collection-is-not-the-solution-for-europes-racism-problem
Schröer, C., Kruse, F., Gómez, J.M.: A systematic literature review on applying CRISP-DM process model. Proc. Comput. Sci. 181, 526–534 (2021). https://doi.org/10.1016/j.procs.2021.01.199, https://www.sciencedirect.com/science/article/pii/S1877050921002416
Stahl, B., et al.: A systematic review of artificial intelligence impact assessments. Artif. Intell. Rev. (2023)
Steen, M., Timan, T., Vethman, S.: Using an extended error matrix to promote transdisciplinary collaboration and jointly work towards social justice (2022). https://marcsteen.nl/docs/ESDiT_2022__Error_Matrix.pdf
Valizadegan H, Amizadeh S, H.M.: Sampling strategies to evaluate the performance of unknown predictors. In: Proceedings SIAM International Conference Data Mining (2014)
Wirth, R., Hipp, J.: CRISP-DM: towards a standard process model for data mining (2000)
Xia, B., Lu, Q., Perera, H., Zhu, L., Xing, Z., Liu, Y., Whittle, J.: Towards concrete and connected AI risk assessment (C2AIRA): a systematic mapping study. In: 2023 IEEE/ACM 2nd International Conference on AI Engineering - Software Engineering for AI (CAIN), pp. 104–116. IEEE Computer Society, Los Alamitos, CA, USA (2023). https://doi.org/10.1109/CAIN58948.2023.00027
Acknowledgement
We would like to thank all our colleagues in the AI Oversight lab\(^{2}\), our external partners, as well as all other public and private organisations that have facilitated transparency on this urgent yet sensitive topic such that the lessons described in this paper could be learned.\(^{2}\)https://appl-ai-tno.nl/projects/ai-oversight-lab/
Author information
Authors and Affiliations
Corresponding author
Editor information
Editors and Affiliations
Rights and permissions
Copyright information
© 2024 The Author(s), under exclusive license to Springer Nature Switzerland AG
About this paper
Cite this paper
Vethman, S., Schaaphok, M., Hoekstra, M., Veenman, C. (2024). Random Sample as a Pre-pilot Evaluation of Benefits and Risks for AI in Public Sector. In: Nowaczyk, S., et al. Artificial Intelligence. ECAI 2023 International Workshops. ECAI 2023. Communications in Computer and Information Science, vol 1948. Springer, Cham. https://doi.org/10.1007/978-3-031-50485-3_10
Download citation
DOI: https://doi.org/10.1007/978-3-031-50485-3_10
Published:
Publisher Name: Springer, Cham
Print ISBN: 978-3-031-50484-6
Online ISBN: 978-3-031-50485-3
eBook Packages: Computer ScienceComputer Science (R0)