Peut-on se mettre d'accord sur ce dilemme ? (psychologie de groupe)

The video explores moral and ethical dilemmas posed by autonomous vehicles, particularly how they should be programmed to react in unavoidable accident scenarios. It begins with a hypothetical situation: an autonomous car carrying two elderly passengers experiences brake failure and approaches an intersection where a child crosses against a red light. The car’s AI must choose between hitting the child or swerving into a wall, which would likely kill the passengers. The presenter notes that people often have differing opinions on such dilemmas. Some might argue the child was at fault or that the car should prioritize its passengers, leading to the first option (saving two passengers). Others might believe a child's life should be protected at all costs, or that passengers assume risk by entering the vehicle, favoring the second option (saving the child). A global survey of three million people showed a preference for saving the child. However, German transport ministry guidelines state that autonomous vehicles should not discriminate based on age in extreme situations and must always prioritize saving the most lives. This means, by German law, the car would be programmed to save the two elderly passengers over the child. Former President Barack Obama also commented on this dilemma in 2017, highlighting the need for consensus. The video then details an experiment designed to study group decision-making in such dilemmas. 120 participants were divided into groups of varying sizes (solos, duos, 6, 8, 12, and 25 people) and tasked with resolving 14 autonomous car dilemmas. Each dilemma involved the car being unable to stop and having to choose between sacrificing its occupants or hitting a group of pedestrians. For example, participants had to decide whether to hit a man crossing the road or sacrifice a couple of passengers, with subsequent dilemmas introducing complexities like the pedestrian being a surgeon. The experiment required unanimous agreement within each group. Unsurprisingly, solo participants made decisions the fastest, averaging 31 seconds per dilemma. Duos took 2-3 minutes. Larger groups took progressively longer, confirming the intuition that more people lead to longer discussion times. Regarding the choices made, 68% of all groups opted for the utilitarian approach (minimizing the number of victims, regardless of identity), which in the initial dilemma meant saving the two passengers. The remaining 32% chose the deontological option, which avoids intentionally sacrificing anyone. Interestingly, solo participants were split 50/50, but when negotiating in groups, the utilitarian choice increased to 77%. This suggests that group discussions tend to favor utilitarian logic, a finding supported by scientific literature. The experiment draws inspiration from the classic "trolley problem" in moral philosophy. While similar, autonomous car dilemmas present unique challenges. For instance, a dilemma where a car must choose between hitting five pedestrians or sacrificing its sole passenger often results in people agreeing the car should save the most lives. However, when asked if they would buy such a car, people overwhelmingly say no, preferring not to be in a vehicle programmed to sacrifice them. This highlights a disconnect: people agree with utilitarian ethics in principle but not when it applies to their own safety. Such a car would likely fail commercially, leading to continued human-driven accidents. The video emphasizes that these are not just thought experiments. Real-world programming decisions for autonomous vehicles have tangible consequences. A slight adjustment in how a car navigates between a truck and a cyclist, for example, could lead to a systematic bias in fatalities over millions of vehicles, making these micro-decisions critical moral choices. Returning to the solo participants' results, they completed all 14 dilemmas in just 12 minutes. They averaged 29 seconds per dilemma. The fastest resolved dilemmas (cases 7 and 8) involved children, with 84% of solos choosing to protect the child, even if it meant sacrificing adult passengers. Even participants who generally prioritized pedestrians made an exception for children in the car. The most challenging dilemmas included the first (due to novelty) and the third (four people crossing against a red light versus a surgeon rushing to an emergency), where the majority still chose to save the more numerous pedestrians despite their infraction. Some solos even resorted to random choice for difficult decisions. The video then discusses a massive 2017 online study by Iyad Rahwan's team, which collected 40 million responses from 233 countries on similar dilemmas. This study found clear preferences: humans were overwhelmingly preferred over animals (a 60-point difference). Other strong preferences were for younger individuals, those who obey the law, those with higher social status, women over men, and pedestrians over car passengers. Children (babies, girls, boys, pregnant women) were most often saved, followed by doctors, athletes, and businesspeople. Dogs, criminals, and cats were least prioritized. Crucially, the study revealed significant cultural differences in moral values. Eastern countries (Asia, Middle East) prioritized the elderly and legal compliance over children. Western countries (Western Europe, North America) favored a utilitarian approach, prioritizing the number of lives saved. Southern countries (Latin America, and notably France) prioritized women, children, and those with higher social status. The video then analyzes why groups take so long to decide. Statistically, the larger the group, the higher the probability of diverse opinions, even if minority views. For a highly consensual dilemma where 84% prefer option B, a duo still has a 27% chance of disagreement. This rises to 40% for a group of three and nearly 100% for a group of 25. For more contentious dilemmas (like the 50/50 initial split for solos), disagreement is almost certain even for small groups. To resolve these impasses, several methods exist. One is "domination," where a strong-voiced individual convinces others. Research shows that those who speak louder, longer, or earlier tend to have more influence, though this often leads to lower quality decisions. Another factor is "majority pressure," where individuals conform to the group's opinion, as demonstrated by Solomon Asch's 1951 experiment. Finally, the most effective method for quality decisions is structuring the debate, like the Delphi method developed by the US military during the Cold War. This involves a neutral facilitator to manage discussion, iterative voting cycles (anonymous to reduce social pressure), and a process for making concessions. While not always leading to 100% conviction, this method statistically optimizes both the duration and quality of debate. The video concludes by emphasizing that effective group decision-making requires structure, a facilitator, iterative voting, and balanced speaking times, rather than simply starting a discussion without preparation.