A Computational Model of Commonsense Moral Decision Making

Abstract

We introduce a new computational model of moral decision making, drawing on a recent theory of commonsense moral learning via social dynamics. Our model describes moral dilemmas as utility function that computes trade-offs in values over abstract moral dimensions, which provide interpretable parameter values when implemented in machineled ethical decision-making. Moreover, characterizing the social structures of individuals and groups as a hierarchical Bayesian model, we show that a useful description of an individual's moral values -as well as a group's shared values -can be inferred from a limited amount of observed data. Finally, we apply and evaluate our approach to data from the Moral Machine, a web application that collects human judgments on moral dilemmas involving autonomous vehicles. 1

Your visual system provides all the information you need to make moral judgments about generic visual events

“…The suggestion that causal perception informs moral judgments is not just an abstract metaphysical point, but also has direct bearing on how we might reverse engineer moral judgments about visual events, for instance, when taking an artificial intelligence approach. Some prominent papers have suggested that biologically inspired models of the visual system, such as hierarchical convolutional neural networks (HCNNs), are effectively too limited to solve aspects of social cognition like moral judgment (Kim et al, 2018;Lake et al, 2017). However, the current results suggest that high-level visual inferences may serve non-perceptual systems like moral judgment, and so we should expect HCNNs to do so too.…”

“…Yet, surprisingly, almost all models of moral cognition do not suggest a role for causal perception, instead concerning themselves with cognitive processing that takes place after perception has already occurred. Where models of visual processing are occasionally mentioned, it is typically with the aim of pointing out that these perceptual models are too limited to explain moral judgment (Kim et al, 2018) or other aspects of intuitive psychology more generally (Lake et al, 2017). Relatedly, almost all studies of moral judgment are conducted using short verbal vignettes, many involving philosophical conundrums such as the trolley dilemma (Foot, 1978).…”

Ethical and Statistical Considerations in Models of Moral Judgments

“…For example, if an individual values infancy highly they may also value the elderly highly. The covariance matrix allows the learner to quickly infer moral principles of one dimension after inferring those of a highly correlated dimension (Kim et al, 2019). Let w w w = {w 1 , ..., w i , ...w n } be the set of moral principles of N respondents and the vector = { 1 1 , ..., t i , ..., T N } represent the resultant states of i respondents over T scenarios.…”

“…using φ to represent the vector of individual i's deontological beliefs. Figure 6 shows that by incorporating deontological statements into prior distributions the model achieves its FIGURE 6 | Box plot showing predictive accuracy of the hierarchical model with priors specified by Kim et al (Kim et al, 2019) when compared to the model initialised with deontological priors.…”

Differences on Utilitarian and Moral Decision Between Male and Female

“…The moral machine is a website application that was designed to analyze the moral performance of participants through hypothetical situations with autonomous vehicles and pedestrians (Bonnefon et al, 2016;Kim et al, 2018). Although the current approach used a website application, this procedure followed the recommendations on moral dilemmas by including the participant in a situation with a finite number of answer options (Christensen & Gomila, 2012), but specifically focused on features of pedestrians preferred by participants.…”