Ethics and Governance of Decision Algorithms in Social Systems
Shermin Voshmgir and Michael Zargham:
"Assumptions that are programmed into the cryptoeconomic protocols might be biased and will be subject to a line of ethical studies covering how the associated cryptoeconomic systems behave over time. When automated systems are envisaged as closed systems they are tantamount to complete contracts–supposing that a prespecified algorithm suffices to judge all possible eventualities. All algorithms presuppose some representation or model of reality; models are always reductions of reality based on some assumptions, and therefore must be judged by their usefulness to some ends[Box 1976][Green and Viljoen 2020]. This places the focus on the assumptions embedded in the models and the effects those assumptions have on people. By acknowledging the subjective and performative aspects of models, and more broadly the synergies between human and machine decision makers, the domain mirrors that of incomplete contracts more closely. The policy-making, machine learning and cryptoeconomic systems design communities share a common need to address ethical questions about the social-systemic effects of algorithm design and implementation [Orlikowski and Scott 2015] [Loukides, Mason and Patil 2018]. To design or govern algorithms which make decisions requires a theory of fairness such as Rawl’s Veil of Ignorance [Rawls 1958] [Heidari et al. 2018]. Fairness cannot be expected to emerge from purely self-interested agents because fairness provides a constraint on profit seeking behavior [Kahneman, Knetsch and Thaler 1986]. As a result, a code of ethics for algorithm designers, as found in other engineering disciplines [Pugh 2009], is required.
Furthermore, it is important to note that data governance [Soares 2015] is not equivalent to protocol governance.
Data governance relates to the management of rights to read, write or manipulate data. Emerging data economies must respect regulations such as General Data Protection Regulation [Voigt and Von dem Bussche 2017] and therefore one cannot simply store private or sensitive information in a public ledger where it cannot be deleted. However, data governance can be addressed through business process automation [Ter Hofstede et al. 2009] using smart contracts [Christidis and Devetsikiotis 2016], which encode the aforementioned rights to read, write or manipulate data which is stored using other cryptographic technologies such as a content addressable distributed hash tables [Benet 2014]. Federated machine learning [Bonawitz et al. 2017][Geyer, Klein and Nabi 2017] is a growing area of research, but practical implementation is hindered by the ethical and regulatory requirement that there are guarantees of privacy preservation [Ahmad, Stoyanov and Lovat 2019]."