Randomized Classifiers vs Human Decision-Makers: Trustworthy AI May Have\n to Act Randomly and Society Seems to Accept This

2021/11/15 by Gábor Erdélyi, Erdélyi, Gábor, Olivia Johanna Erdélyi +4
Computer Science · Social Sciences · #Adversarial Robustness in Machine Learning #Artificial Intelligence (cs.AI) #Computers and Society (cs.CY) #Ethics and Social Impacts of AI #Explainable Artificial Intelligence (XAI) #FOS: Computer and information sciences #Machine Learning (cs.LG)

paper · pdf · doi:10.48550/arxiv.2111.07545

openalex publication_date 2021/11/15 · openalex created_date 2023/02/18 · openalex updated_date 2026/07/28

Abstract

As \artificial intelligence (AI) systems are increasingly involved in\ndecisions affecting our lives, ensuring that automated decision-making is fair\nand ethical has become a top priority. Intuitively, we feel that akin to human\ndecisions, judgments of artificial agents should necessarily be grounded in\nsome moral principles. Yet a decision-maker (whether human or artificial) can\nonly make truly ethical (based on any ethical theory) and fair (according to\nany notion of fairness) decisions if full information on all the relevant\nfactors on which the decision is based are available at the time of\ndecision-making. This raises two problems: (1) In settings, where we rely on AI\nsystems that are using classifiers obtained with supervised learning, some\ninduction/generalization is present and some relevant attributes may not be\npresent even during learning. (2) Modeling such decisions as games reveals that\nany -- however ethical -- pure strategy is inevitably susceptible to\nexploitation.\n Moreover, in many games, a Nash Equilibrium can only be obtained by using\nmixed strategies, i.e., to achieve mathematically optimal outcomes, decisions\nmust be randomized. In this paper, we argue that in supervised learning\nsettings, there exist random classifiers that perform at least as well as\ndeterministic classifiers, and may hence be the optimal choice in many\ncircumstances. We support our theoretical results with an empirical study\nindicating a positive societal attitude towards randomized artificial\ndecision-makers, and discuss some policy and implementation issues related to\nthe use of random classifiers that relate to and are relevant for current AI\npolicy and standardization initiatives.\n

Citations

• Power to the People: The Role of Humans in Interactive Machine Learning
• The Problem of Abortion and the Doctrine of the Double Effect
• On a Test of Whether one of Two Random Variables is Stochastically Larger than the Other
• Are Individual Differences Germane to the Acceptance of New Information Technologies?
• Productivity and Selection of Human Capital with Machine Learning
• Equality of Opportunity in Supervised Learning
• Inherent Trade-Offs in the Fair Determination of Risk Scores
• Clinical course and risk factors for mortality of adult inpatients with COVID-19 in Wuhan, China: a retrospective cohort study
• Clinical and immunological features of severe and moderate coronavirus disease 2019

Related

• From computational ethics to morality: how decision-making algorithms can help us understand the emergence of moral principles, the existence of an optimal behaviour and our ability to discover it
• Assessing AI Utility: The Random Guesser Test for Sequential Decision-Making Systems
• Decision-aid or Controller? Steering Human Decision Makers with Algorithms
• Does AI help humans make better decisions? A statistical evaluation framework for experimental and observational studies
• The Morality of Probability: How Implicit Moral Biases in LLMs May Shape the Future of Human-AI Symbiosis