A study conducted by researchers from Imperial College London has uncovered intriguing insights into human interactions with artificial intelligence (AI) agents. The study, led by Jianan Zhou and Dr Nejra van Zalk from the Dyson School of Design Engineering at Imperial, was published in the journal Human Behavior and Emerging Technologies. It explored human behaviour in relation to AI bots during a virtual ball game and revealed a tendency among humans to extend social behaviours typical among humans to AI entities, especially when these AI agents appeared to be excluded or "ostracised" from playtime.
The experiment was structured around a virtual game called 'Cyberball', a simple activity where players on a screen throw a digital ball to one another. A total of 244 human participants, aged between 18 and 62, were involved in the study. During the game, participants observed scenarios where an AI bot, designed to represent a virtual agent, was alternately included or excluded by a human non-participant. Participants then engaged with the game, and their reactions were analysed to assess whether they would choose to rectify perceived unfairness towards the AI bot.
The findings indicated that participants generally displayed empathy towards the AI agents, attempting to include the AI bot by passing the ball to it whenever it seemed isolated or mistreated. Interestingly, older participants exhibited a stronger tendency towards identifying and correcting unfairness. This reaction mirrors typical human-to-human interactions, where individuals are inclined to support those who are unfairly treated and show disdain towards those perpetrating exclusionary behaviour.
These results shed light on a significant behavioural trend whereby humans treat AI agents as social beings. This inclination is particularly relevant as AI agents become more integrated into everyday interactions, from service access to companionship roles. The research suggests designers should be cautious in making AI overly human-like to help distinguish real from virtual interactions, especially given the potential implications for replacing genuine human relationships in some contexts, such as friendship or health advising.
While the study’s setup did involve a unique form of interaction—via a virtual ball game—it acknowledged that real-world human-AI interactions typically occur through conversational interfaces like chatbots and voice assistants. This virtual setup might not entirely reflect natural interaction scenarios, where communication often involves spoken or written language, potentially altering user expectations and perceptions.
In response, the researchers plan to expand their studies to encompass direct, face-to-face conversations with AI agents in diverse settings, from controlled laboratory environments to casual encounters. These future investigations aim to determine the broader application of their findings and further understand the dynamics of human empathy towards AI in more realistic interaction environments.
Overall, the Imperial College London study provides valuable insights into human cognitive processing concerning AI. It underscores the importance of considering psychological factors in the design and implementation of AI technology to ensure beneficial, clear, and appropriate human-machine interfaces.
Source: Noah Wire Services