W3: Araujo et al. (2020) Flashcards
(14 cards)
Paper’s purpose
Explores how personal characteristics influence perceptions of automated decision-making by AI across media, health, and judicial contexts, drawing from social science theories and emerging studies on algorithmic appreciation. Findings reveal mixed opinions about the fairness and usefulness of AI-driven decisions influenced by individual characteristics. AI decisions were often evaluated on par or even favourably compared to those made by human experts in specific scenarios
Automated decision-making (ADM)
Involves algorithms or AI systems processing data to make automated decisions, with feedback loops for system improvement. It is a socio-technical concept evolving through societal contexts. It can range from AI-powered recommendation systems to fully autonomous decision-making processes. The level of human involvement varies, with some systems leavin autonomy to users while others operate without human input. While it offers potential benefits, there are growing concerns about the biases and limitations. Much is unexplored about what influences people’s perceptions of the fairness, usefulness, and risks of ADM, especially as these play a critical role in the acceptance and evaluation of such systems
Attitudes towards AM
Influenced by factors including perceptions of objectivity and rationality compared to human decision-makers. There’s a tendency for people to trust expert systems more than human advisers, often based on the belief that statistical methods outperform human judgement. This idea is reinforced by the machine heuristic. Attitudes may also be influenced by factors such as the system’s transparency, forgiveness towards mistakes, and the context in which decisions are made. Trust in ADMs also depends on the type of decision-maker compared (e.g. non-expert vs. expert), and the nature of the decision being evaluated (e.g. objective vs. subjective)
Machine heuristic
Suggests that users perceive algorithmic decisions as less biased when the interface is less anthropomorphised
Knowledge
Previous research shows mixed findings regarding the impact of knowledge on perceptions of ADM. While comfort with mathematics and education level generally correlates positively with favourable attitudes, higher levels of computer programming knowledge may lead to perceptions of less fairness in algorithmic decisions
Online privacy concerns and self-efficacy
ADM is closely linked to data-driven decisions, raising concerns about personal data privacy. Higher levels of privacy concern tend to lead to more critical evaluations of ADM, while greater online self-efficacy results in more positive perceptions
Demographics
Age and gender may also impact perceptions of ADM, although evidence is mixed. Older individuals may prefer human decision-makers over algorithms, while gender influences attitudes toward algorithmic fairness
Belief in equality
The perception of ADM as unbiased and objective, despite evidence of bias and risks, may be influenced by personal beliefs about equality. Positively associated with perceptions of usefulness and fairness
General knowledge (education)
Positively associated with perceptions of usefulness
Domain-specific knowledge
About AI, algorithms, and computer programming. Positively associated with perceptions of both usefulness and fairness
Online privacy conerns
Negatively associated with perceptions of usefulness and fairness, but positively associated with perceived risk
Online self-efficacy
Positively influencing perceptions of usefulness and fairness, while negatively influencing perceived risk
Age
Negatively associated with perceived usefullness, but positively associated with risk
Gender
Females perceived ADM as less useful than males
