TL;DR:
- Study reveals people’s tendency to conform more to artificial intelligence (AI) in objective tasks and to humans in subjective tasks.
- Participants exhibited higher levels of conformity to AI in counting tasks with a single correct answer.
- Humans were more influential when tasks involved attributing meaning to images.
- The study emphasizes the expanding role of AI as a source of social influence.
- Participants showed a greater inclination to conform to AI in a digital context under specific circumstances.
- For objective tasks, AI agents may be perceived as more credible and accurate.
- However, humans continue to be seen as the most informative source for subjective tasks.
Main AI News:
A recent study has provided valuable insights into the extent to which individuals conform to information delivered by an artificial intelligence (AI) agent compared to information presented by a human. The results demonstrated that participants exhibited higher levels of conformity when receiving information from an AI in tasks involving objective measurements with a single correct answer. Conversely, in tasks requiring the attribution of meaning to images, participants conformed more to information provided by a human. These findings, published in Acta Psychologica, shed light on the dynamics of social influence in the context of AI.
Social influence encompasses various processes through which individuals or groups impact the attitudes, beliefs, and decisions of others. One form of social influence is conformity, which entails adjusting one’s attitudes, beliefs, and behaviors to align with social norms or the norms of another individual or group. Compliance, obedience, and persuasion also represent different aspects of social influence.
Throughout history, the primary sources of influence on people’s thoughts, emotions, and behaviors have been other humans. However, the emergence of artificial intelligence and non-human agents, such as chatbots, virtual assistants, and robots, has expanded the spectrum of potential sources of social influence beyond human entities.
To examine the differential impact of information provided by humans versus AI agents, the study’s lead author, Paolo Riva, and his colleagues conducted two experiments, each featuring a distinct task. The researchers hypothesized that the level of influence would depend on the nature of the task at hand. In objective tasks, where participants were required to perform counting exercises, the researchers anticipated that AI would exert a greater influence. Conversely, in subjective tasks involving the attribution of meaning, they expected humans to wield more influence.
Recruiting participants via Qualtrics, the researchers enlisted 177 individuals for the first study and 102 participants for the second study. In the first study, participants were presented with a series of eight images consisting of black backgrounds with white dots. Each image appeared for seven seconds, allowing participants to form a rough estimation of the number of dots present, even though it was insufficient time for an accurate count. After the image disappeared, participants were asked to provide their estimates.
Subsequently, the participants were shown two estimations of the number of dots—one attributed to an AI and the other to a human. The participants were randomly assigned to two groups. In the first group, the AI systematically overestimated the number of dots by approximately 15%, while the human consistently underestimated the count by the same margin. In the second group, the roles were reversed, with the AI underestimating and the human overestimating. Following exposure to these estimates, participants were asked to provide their revised estimations.
In the second study, participants were presented with images derived from the card game Dixit, and there were no time constraints for viewing. Each image was paired with two concepts, both equally suitable for association with the image, according to prior evaluations. The participants were informed that one concept was proposed by an AI, while the other was proposed by a human. The program randomly determined which concept would be ascribed to the AI or the human for each participant. Subsequently, participants were asked to rate the extent to which each concept represented the image they had viewed.
The results of the first study revealed that participants conformed more to the influence of AI. When asked to revise their estimations of the dot counts, they were more likely to align their estimates with those proposed by the AI, irrespective of whether the AI overestimated or underestimated the count. Furthermore, participants explicitly expressed a belief in the greater accuracy of the AI’s estimations.
In the second study, the human influence was found to be more pronounced among participants compared to the influence of the AI. However, when explicitly questioned about which source they perceived as more informative, the number of participants favoring the human as the more informative source was nearly equal to the number of participants favoring the AI.
“The findings of this study indicate that individuals may exhibit a greater propensity to conform to non-human agents, specifically in digital contexts and under specific circumstances. When faced with objective tasks that evoke uncertainty, people may be more inclined to conform to AI agents rather than other humans. However, for subjective tasks, human agents are likely to remain the most credible sources of influence when compared to AI agents,” concluded the study authors.
Conclusion:
This research underscores the influence of artificial intelligence in shaping human behavior, particularly in objective tasks. As AI continues to advance and integrate into various industries, businesses should recognize its potential as a source of social influence. By leveraging AI in areas such as decision-making, customer support, and recommendations, companies can tap into its credibility and accuracy to enhance customer engagement and satisfaction.
However, it is vital to understand that in subjective tasks, human influence remains paramount. Striking a balance between AI-driven objectivity and the human touch will be key in leveraging AI’s potential while maintaining the trust and preferences of customers in the market.
