Top-down approach to compare the moral theories of deontology and utilitarianism in Pac-Man game setting
The processes underlying important decisions in many areas of our everyday lives are getting increasingly automatized. In the near future, as many decisions would be made by autonomous artificial agents, it would be necessary to ensure that these agents do not cause harm to society. Therefore, artif...
| Autor: | |
|---|---|
| Tipo de recurso: | tesis de maestría |
| Fecha de publicación: | 2019 |
| País: | España |
| Institución: | Universitat Oberta de Catalunya (UOC) |
| Repositorio: | O2, repositorio institucional de la UOC |
| OAI Identifier: | oai:openaccess.uoc.edu:10609/99526 |
| Acceso en línea: | http://hdl.handle.net/10609/99526 |
| Access Level: | acceso abierto |
| Palabra clave: | artificial morality knowledge representation information processing moral artificial representación del conocimiento procesamiento de la información representació del coneixement processament de la informació Knowledge representation (Information theory) -- TFM Representació del coneixement (Teoria de la informació) -- TFM Representación del conocimiento (Teoría de la información) -- TFM |
| Sumario: | The processes underlying important decisions in many areas of our everyday lives are getting increasingly automatized. In the near future, as many decisions would be made by autonomous artificial agents, it would be necessary to ensure that these agents do not cause harm to society. Therefore, artificial agents need to be furnished with a way of acknowledging the moral dimension of their actions. In this study, we use a top-down approach to implement and compare two common moral theories, deontology and utilitarianism, in the same setting. While deontology focuses on the intention behind an action and the nature of an act, utilitarianism emphasizes that an action should be judged solely by the consequences it has and that it should maximize overall good. The differences between both theories need to be captured differently when implementing an artificial moral agent. Inspired by the famous Pac-Man game, we computationally model two moral Pac-Man agents based on top-down rules: a deontological one and a utilitarian one. Besides, we also model an amoral Pac-Man agent that does not take into account any ethical theory when guiding its actions. According to the theory of dyadic morality, every moral or immoral act involves an agent and a patient. In our Pac-Man world, we have an agent helping or harming a patient for every moral or immoral act. The amoral Pac-Man agent does not take into account whether its action would help or harm the patient. The deontological Pac-Man agent constrains its behavior depending on a set of prohibited actions and duties. On the contrary, the utilitarian Pac-Man agent evaluates the happiness and pain of the actions at hand to maximize the happiness, while trying to avoid unnecessary evils when possible. After implementing the agents, we compare their behaviour in the Pac-Man world. While the deontological Pac-Man agent may sometimes have to face conflict between succeeding and sticking to its value of always doing the right thing, the utilitarian Pac-Man agent always manages to succeed. In this study, we discuss the conflicts that arise for each moral agent, between their values and the goals of the game in different scenarios. |
|---|