Discussion paper: International data ethics frameworks

Published on 07 June 2020.

Discussion paper - International data ethics frameworks - March 2020 [PDF, 296KB]

Purpose and scope

This paper has been prepared on behalf of the Government Chief Data Steward for the Data Ethics Advisory Group (DEAG). It discusses the current landscape of data ethics frameworks, which includes artificial intelligence, and explores common themes which are particularly relevant to the function of the DEAG. The focus is on the benefits and considerations of frameworks. This paper is not official New Zealand government policy.

This paper is intended to support the general discussion of the benefits and considerations of data ethics frameworks for the DEAG. The issues canvassed should not be considered reflective of the position of any specific government agency (including Stats NZ).

Executive summary

Data ethics frameworks have proliferated from a variety of stakeholders, primarily in Europe and North America. They tend to focus on artificial intelligence and provide high-level principles or ‘deontological ethics’ which guide professional cultures and narratives in non-binding ways. They are considered more flexible and can be applied more rapidly than laws or professional codes of conduct. Some frameworks include self-assessments or certifications for compliance.

  • While generally non-binding, some frameworks can be mandated under specific conditions such as during the of procurement for AI systems (e.g. UK Government Draft Guidelines for AI Procurement).
  • In most cases frameworks are framed for ‘ethics-by-design’ and they target new projects during the procurement, development or deployment stages. There is limited discussion on a retro-active application of frameworks.

UK Government Draft Guidelines for AI Procurement

Four common themes of ‘privacy’, ‘transparency’, ‘bias and discrimination’ and ‘accountability’ emerge from analysis of these frameworks, however, divergences tended to arise in relation to the actors which have formulated them, how they are framed and how they are interpreted. The common themes tend to overlap with areas where technical fixes can be or have already been developed.

  • Recent frameworks [1] expand on the four common themes with aims for social benefit and they tend to incorporate aspects or refences to human rights. Key themes include the balance of beneficence and non-maleficence (‘proportionality’), ‘autonomy’ or ‘self-determination’ and ‘justice’.
  • It should be noted that this exploratory work did not find prominent indigenous themes per se for the majority of data ethics frameworks, however, the CARE Principles indicate three key indigenous ethical components: ‘proportionality’, ‘justice’ and ‘future use’.

CARE Principles

While there is some consensus at a high level, this has not been achieved at a detailed level. In particular, there are unresolved issues around how these principles should be interpreted, why they should be deemed important, what issue, domain or actors they should pertain to, and how they should be implemented [2].

There have been calls to consolidate some existing ethical frameworks, particularly where frameworks are frustrating attempts to achieve industry-level compliance and streamlined processes. [3] In this sense, frameworks should be designed to complement professional codes of conduct or standards (e.g. IEEE) which provide practical guidance for data practitioners and laws which provide procedures to manage and remediate non-compliance.

IEEE - Institute of Electrical and Electronics Engineers

References

[1] E.g.: Cowls, J. and Floridi, L., 2018. An ethical framework for a good AI society, URL.

[2] Jobin, A., Ienca, M. and Vayena, E., 2019. The global landscape of AI ethics guidelines. Nature Machine Intelligence, 1(9), pp.389-399, URL; Daly, A., Hagendorff, T., Hui, L., Mann, M., Marda, V., Wagner, B., Wang, W., Witteborn, S., 2019, Artificial Intelligence Governance and Ethics: Global Perspectives, URL.

[3] See page 56: Australian Human Rights Commission, 2019. Human Rights and Technology Discussion Paper, URL; and Mittelstadt, B., 2019. Principles alone cannot guarantee ethical AI. Nature Machine Intelligence, pp.1-7, URL.

Remaining content of the paper

Following the executive summary, the PDF version of the paper also contains the following content.

  • There has been a proliferation of international principles
  • Common theme: privacy
  • Common theme: transparency
  • Common theme: bias and discrimination
  • Common theme: accountability
  • Considerations for data ethics frameworks
    • The uptakes and usages of data ethics frameworks are not well understood
    • Similarities have formed between data ethics and medical ethics
    • Differences between frameworks arise from the actors involved, framing and interpretations
    • Data ethics frameworks do not provide a ‘catch-all’ for ethical practices
    • Indigenous themes are not prominent in data ethics frameworks
    • There are unresolved tensions which arise from framework ambiguity and different values
    • There are preconceived narratives which exist in the area of data and AI ethics
  • Appendix 
    • Table 1 - Simplified principles from literature analysis.

 

Contact us

Please email datalead@stats.govt.nz.

Top