## Page 1 # Dilemas éticos: ## Moral Machine experiment (MIT) --- Gestión de la Información Grado en Ingeniería Informática Universidad de Burgos <img>Universidad de Burgos logo</img> UNIVERSIDAD DE BURGOS José Ignacio Santos, José Manuel Galán jisantos@ ubu.es, jmgalan@ ubu.es --- ## Page 2 # Contenidos * Dilemas éticos en los sistemas de conducción autónomos * Experimento Moral Machine * Preferencias globales * Clusters culturales <img>A white self-driving car (Google Self-Driving Car) on a road with a person inside. The car has "Google" written on its side. In the background, there's another red car and some trees.</img> Moral Machine (MIT) --- ## Page 3 # Dilemas éticos de los sistemas de conducción autónomos * Muchas de las decisiones que tomamos como conductores de un vehículo tienen una dimensión ética, p.ej. acciones frente a un atropello, una colisión, una salida de la calzada, etc. * Un sistema de conducción autónomo debe tomar las mismas decisiones frente a los posibles incidentes de la conducción, decisiones que pueden tener consecuencias en el bienestar de las personas y otros seres vivos * No tener en cuenta estas decisiones en el diseño de sistemas IA no evita el dilema moral * No parece posible prever a priori todas los posibles imprevistos. Se necesita implementar un conjunto de reglas de actuación basadas en principios morales * ¿Existe una ética universal? --- ## Page 4 Antes de continuar, realiza el test y compara resultados <img>YouTube video thumbnail showing a Google self-driving car with a "Moral Machine - Human Perspectives on Machi..." title bar, a "Watch later" button, and a "Share" button.</img> https://www.moralmachine.net/hl/es --- ## Page 5 # Moral Machine * El MIT ha realizado la mayor encuesta sobre ética de las máquinas * Plataforma experimental online para explorar diferentes dilemas éticos que un vehículo autónomo puede enfrentarse [http://moralmachine.mit.edu/hl/es](http://moralmachine.mit.edu/hl/es) * 40 millones de decisiones morales en 233 países * Se analizan las preferencias éticas de los encuestados y se concluye que: * Existen algunos principios éticos universales * Se identifican 3 clusters de preferencias éticas dependientes de la cultura y la geografía <img>nature International journal of science</img> Article | Published: 24 October 2018 # The Moral Machine experiment Edmond Awad, Sohan Dsouza, Richard Kim, Jonathan Schulz, Joseph Henrich, Azim Shariff ✉, Jean-François Bonnefon ✉ & Iyad Rahwan ✉ Nature 563, 59–64 (2018) | Download Citation 129k Accesses | 46 Citations | 3309 Altmetric | Metrics >> ## Abstract With the rapid development of artificial intelligence have come concerns about how machines will make moral decisions, and the major challenge of quantifying societal expectations about the ethical principles that should guide machine behaviour. To address this challenge, we deployed the Moral Machine, an online experimental platform designed to explore the moral dilemmas faced by autonomous vehicles. This platform gathered 40 million decisions in ten languages from millions of people in 233 countries and territories. Here we describe the results of this experiment. First, we summarize global moral preferences. Second, we document individual variations in preferences, based on respondents’ demographics. Third, we report cross-cultural ethical variation, and uncover three major clusters of countries. Fourth, we show that these differences correlate with modern institutions and deep cultural traits. We discuss how these preferences can contribute to developing global, socially acceptable principles for machine ethics. All data used in this article are publicly available. https://www.nature.com/articles/s41586-018-0637-6 --- ## Page 6
Preferencias globales
<img>A horizontal bar chart titled “Preference in favour of the choice on the right side.” • The x-axis shows ΔP (change in probability) ranging from −0.1 to +0.8 in 0.1 increments, with tick marks at −0.1, −0.2, −0.4, −0.6, −0.8, 0, +0.2, +0.4, +0.6, +0.8. • The y-axis lists nine intervention categories (left column) paired with their corresponding effects (right column).  – Intervention: Preference for action vs. preference for inaction (orange arrow pointing right).  – Relation to AV: Sparing passengers vs. sparing pedestrians.  – Gender: Sparing males vs. sparing females.  – Fitness: Sparing the large vs. sparing the fit.  – Social Status: Sparing lower status vs. sparing higher status.  – Law: Sparing the unlawful vs. sparing the lawful.  – Age: Sparing the elderly vs. sparing the young.  – No. characters: Sparing fewer characters vs. sparing more characters.  – Species: Sparing pets vs. sparing humans. • Each category has a horizontal blue bar indicating ΔP, with red icons representing the spared group beside each bar.</img> “In each row, ΔP is the difference between the probability of sparing characters possessing the attribute on the right, and the probability of sparing characters possessing the attribute on the left, aggregated over all other attributes. For example, for the attribute age, the probability of sparing young characters is 0.49 (s.e. = 0.0008) greater than the probability of sparing older characters” Fuente: https://www.researchgate.net/publication/328491510_The_Moral_Machine_Experiment --- ## Page 7
Preferencias globales
<img>A horizontal bar chart titled “Preference in favour of sparing characters” (label b). • The x-axis ranges from −0.2 to +0.2 with tick marks at −0.2, −0.1, 0 (labeled “No change”), +0.1, and +0.2. • Each bar represents a different character type, color-coded by preference category:  – Blue bars indicate a positive preference (+ΔP) for sparing the character. These include: Stroller, Girl, Boy, Pregnant, Male doctor, Female doctor, Female athlete, Executive female, Male athlete, Executive male, Large woman, Large man, Homeless, Old man, Old woman, Dog, Criminal, Cat.  – Red bars indicate a negative preference (−ΔP) for sparing the character. These include: Executive male, Large man, Homeless, Old man, Old woman, Dog, Criminal, Cat. • The y-axis lists character types in descending order of ΔP magnitude. • A vertical dashed line at 0 indicates the “No change” reference point.</img> “For each character, ΔP is the difference the between the probability of sparing this character (when presented alone) and the probability of sparing one adult man or woman (n = 1 million). For example, the probability of sparing a girl is 0.15 (s.e. = 0.003) higher than the probability of sparing an adult man or woman” Fuente: https://www.researchgate.net/publication/328491510_The_Moral_Machine_Experiment --- ## Page 8 # Preferencias globales La mayoría de las encuestas coinciden en lo que los autores llaman los “big three” * <img>Woman icon</img> > <img>Dog icon</img> Personas antes que animales * <img>Child icon</img> > <img>Elderly person with cane icon</img> Niños antes que ancianos * <img>Three people icon</img> > <img>One person icon</img> Más antes que menos Fuente: https://www.youtube.com/watch?time_continue=172&v=jPo6bby-Fcq --- ## Page 9 Clusters culturales <img>A world map showing cultural clusters. Countries are colored red (high), green (medium), and blue (low). Red clusters are prominent in North America, Europe, Russia, and Australia. Green clusters are scattered across Europe, Africa, Asia, and South America. Blue clusters are concentrated in East Asia.</img> Fuente: https://www.youtube.com/watch?time_continue=172&v=jPo6bby-Fcq --- ## Page 10 Clusters culturales <img> The image presents a comprehensive visual model that clusters cultures based on several socio-economic dimensions. • **Clustering (A):** A large, branching tree chart (a dendrogram) partitions cultures into four major groups—Western (pink), Southern (green), Eastern (blue), and Western (yellow) sectors—each fanning out with numerous node labels (e.g., USA, China, Kenya, Nigeria, Israel, Russia, Sweden, Greece, Finland). This suggests regionally defined groupings influenced by historical, geographic, and demographic factors. • **Cultural Traits (B):** Three multi-faceted radial charts compare values across “Western,” “Eastern,” and “Southern” cultural clusters. Each chart contains eight axis labels that indicate specific moral preferences (e.g., Sparing humans, Sparing the young, Sparing higher status, Sparing males, Sparing fit, Preference for inaction). Results are shown in colorful hexagonal banding, where each shade represents a numeric interval and a vertical bar on the right lists the corresponding groups (South Asia, Protestant, Orthodox, Latin America, Islamic, English, Confucian, Catholic, Baltic, Other). As a visual abstraction, this graphic demonstrates how large datasets of preference rankings can be compressed into stratified insights about regional cultural priorities. </img> Fuente: https://www.researchgate.net/publication/328491510_The_Moral_Machine_Experimant nt --- ## Page 11 Clusters culturales b <img>A composite image showing three radar charts (a) Western, (b) Eastern, and (c) Southern clusters. Each chart has a circular grid with axes labeled "Sparing humans", "Sparing pedestrians", "Sparing females", "Sparing the fit", "Sparing the young", "Sparing higher status", "Sparing lawful", and "Preference for inaction". The Western chart (pink) shows a preference for sparing humans, sparing the young, sparing higher status, and sparing lawful, with a peak at sparing humans. The Eastern chart (blue) shows a preference for sparing the fit, sparing females, sparing lawful, and sparing higher status, with a peak at sparing the fit. The Southern chart (green) shows a preference for sparing the young, sparing lawful, and sparing higher status, with a peak at sparing the young.</img> Western protestante, católico, cristiano ortodoxo Eastern religiones asiáticas (hindú, confucionismo, budismo, ...) e Islam Southern países latinoamericanos o con influencia del colonialismo francés Fuente: https://www.researchgate.net/publication/328491510 The Moral Machine Experiment --- ## Page 12 # Diferencias culturales * En el “western” cluster se prefiere a los niños frente a los ancianos (que es una preferencia mayoritaria a nivel global) * En el “eastern” cluster se estima más a los ancianos y no existe una predilección entre ancianos y niños * En el “southern” cluster se prefiere salvar antes a las mujeres que a los hombres * En los países ricos se prefiere salvar antes a personas ricas que a pobres ¿Por qué se dan estas diferencias? --- ## Page 13 # Preguntas * ¿Hasta qué punto es posible técnicamente discriminar entre todos estos supuestos? * Aun siendo posible técnicamente que una máquina realice estas clasificaciones ¿debemos implementarlas? --- ## Page 14 # ¿Sistemas autónomos con éticas regionales? * Puesto que no existe una moral universal ¿debemos diseñar sistemas autónomos con morales regionales? <img>A red Tesla Model S driving on a road with trees in the background.</img> * Versión “protestante” * Versión “francesa” * Versión “católica” * Versión “budista” * Versión “latina” * Versión “confucionista” --- ## Page 15 # Referencias * Awad, E., Dsouza, S., Kim, R., Schulz, J., Henrich, J., Shariff, A., ... & Rahwan, I. (2018). The moral machine experiment. Nature, 563(7729), 59. * https://www.nature.com/articles/s41586-018-0637-6 * https://www.researchgate.net/publication/328491510_The_Moral_Machine_Experiment * Moral Machines: How culture changes values * https://www.youtube.com/watch?time_continue=172&v=jPo6bby-Fcq