Although political decisions deeply affect our personal and social lives and determine the future of any society, their quality is often disputable. This is due to elements of poor information and irrationality that conventional decision-making involves, like insufficient documentation, biased evaluation of the data collected, absurd preferences, or even corrupting practices of the responsible politicians or administrative personnel. This article explores the possibility to improve the quality of political decisions in general, and more specifically of governmental decisions concerning legal regulation, by introducing artificial intelligence applications into the process of decision-making. Three topics deserve special attention: 1) the question on compliance with constitutional prerequisites that characterize every modern democracy, 2) the problem of personal data collection and processing, and 3) the question related to political accountability.
TopIntroduction
The expanded use of artificial intelligence in decision-making nowadays involves the most critical sectors of social activity, from economic life, communications, and crime prevention, to education, health, and scientific research. With the term “artificial intelligence” (AI), we mean, here, “machine-based systems that can, for a given set of human-defined objectives, make predictions, recommendations, or decisions influencing real or virtual environments” (OECD, 2019).
In a complex world requiring rapid, accurate, transparent, and unbiased decisions, the benefits are obvious; the systems of artificial intelligence make it possible to avoid obstacles inherent to human decision-making, due to various reasons, like limited access to relevant data, bureaucratic restraints, waste of time, and even unethical motivation. On the other hand, the involvement of algorithms in decision-making presents a major risk which may be described as a fundamental lack of adaptability to specific conditions, unforeseen at the time of the algorithm development. That problem affects directly the responsibility of those entrusting decisions to AI systems and, in legal terms, their liability. So far, no alternatives to the traditional responsibility of humans exist to capture decisions escaping from the direct human control as those of AI systems; humans remain responsible for these decisions too, and this may discourage the use of AI systems.
The above remarks make inevitable a permanent work of risk/benefit balancing regarding the use of AI in specific sectors of social activity. However, the sector of political decision-making illustrates, here, a striking exemption. Indeed, the world of politicians, persons undertaking critical decisions with massive influence at the scale of populations, looks unattainable considering the presence of AI applications in policy- and- law-making at any level. This happens although the relevance of AI is extensively investigated, over the last thirty years (Duffy & Tucker, 1995; der Voort et al., 2019, p. 27; Hochtl et al., 2016, pp.154-156 et seq.; Rubinstein et al., 2016; Poel et al., 2018; Castelluccia & Le Métayer, 2019, p. 19 et seq.), to the extent that examples of “digital politicians” and “virtual embassies” (a Swedish initiative) (Efthymiou et al., 2020, pp. 49-50) are also discussed.
On the one hand, this seems reasonable if we take into account the specific nature of political decisions requiring direct relation to the will of political representatives for ensuring their accountability before the people, at least in modern democracies. On the other, it is evident that numerous (and often crucial) political decisions either fail to regulate or even, they provoke damage in terms of public interest they are supposed to serve due to problems of misjudging, delaying, corruption, or just personal incapacity, all related to the human nature of decision-makers.
In general, we can explain such problems as a) lack of information crucial for a certain decision, b) biased evaluation of the relevant data (even if sufficient), c) incapacity of distinguishing and calculating important data in complex cases. Assuming that prevention of problems belonging to any of the three categories could be feasible if we entrusted applications of AI, it looks reasonable to explore a possibility of removing a portion of political decision-making from the direct will of our political representatives or officers, following the example of other sectors of social life.