Researchers at Imperial College London found that individuals exhibited empathy towards and demonstrated protectiveness of AI robots that were excluded from playtime, suggesting a level of anthropomorphism in their perception of artificial intelligence entities.
Researchers conducting a study utilizing a digital ball simulation found that individuals subconsciously treat AI intermediaries like social entities, a phenomenon that should be taken into account when developing artificial intelligence systems.
The research is revealed in.
Lead creator Jianan Zhou, from Imperial’s Dyson College of Design Engineering, notes that this project offers a unique insight into human-AI collaboration, with profound implications for the design of AI systems and our understanding of human psychology.
As human interaction with artificial intelligence (AI) digital intermediaries becomes increasingly prevalent, individuals are now commonly expected to collaborate with these systems when engaging with businesses, while many also utilize them as companions for social communication. Despite these discoveries suggesting that humans are better at recognizing and responding to subtle social cues, builders should refrain from creating agents that are too similar to humans.
According to Senior Creator Dr. Nejra van Zalk of Imperial’s Dyson College of Design Engineering, “A growing yet inconsistent body of research suggests that individuals may have mixed views on treating AI digital agents as social entities.” What implications do these relationships have for the way people collaborate with intermediaries?
Individuals treated AI-powered agents as social entities when they perceived exclusion by attempting to involve them in a ball-tossing game if the AI seemed to be left out, thereby suggesting a desire for social interaction with artificial intelligence. The phenomenon of familiar patterns emerging in human-computer conversations has been consistently observed, with individuals exhibiting this trait even when interacting with artificial intelligence agents, as demonstrated by their participants’ consistent responses despite being aware they were communicating with a digital entity. Apparently, this impact was significantly more pronounced among elderly populations.
People’s innate inclination to empathize and take corrective action against injustice is a deeply ingrained trait. Previous studies revealed that when people excluded others from social interactions, they would often make up for it by throwing the ball to the ostracized individual more frequently, and concurrently, express dislike towards the perpetrator while harbouring a sense of pity and compassion towards the target.
Researchers investigated how 244 humans reacted when an AI digital agent was deliberately excluded from participation by another player in the game “Cyberball”, where players pass a virtual ball to each other on screen, to test the theory that exclusion would elicit empathetic responses towards the AI and increased cooperation among human participants. Participants ranged in age from 18 to 62 years old.
Players occasionally tossed the ball with moderate frequency to AI opponents, while in other instances, they deliberately excluded the digital players, targeting their human teammates exclusively.
Researchers surveyed contributors about their reactions when the ball was handled unfairly, inquiring if they supported a bot receiving possession afterwards, and the reasons behind their opinions.
Participants frequently attempted to compensate for perceived bias by deliberately targeting the bot with passes, thereby attempting to rectify the perceived unfairness. Individuals older in age were more apt to perceive unfairness.
As artificial intelligence-driven brokers become increasingly popular for collaborative tasks, increased human interaction may foster greater recognition and trigger automatic processing. Can customers naturally perceive digital agents as colleagues and engage with them in a social manner?
While this notion presents potential benefits for collaborative work, it more likely pertains to digital intermediaries acting as facilitators of interpersonal connections or health consultants in physical or mental wellness spheres.
By eschewing the creation of overly anthropomorphic agents, developers may facilitate a clearer distinction between digital and genuine interaction. They may further customize their designs to cater to specific age groups, taking into consideration how different human characteristics influence our perception.
Researchers acknowledge that Cyberball’s cooperative gameplay may not accurately represent human collaboration in everyday scenarios, where interactions often occur through written or verbal communication with AI-powered tools like chatbots and voice assistants. The potential experimentation may have conflicting outcomes, potentially disrupting consumers’ preconceived notions and sparking emotional reactions, ultimately influencing respondents’ reactions within the study.
Subsequent to these preliminary findings, researchers are developing complementary studies that involve face-to-face interactions with brokers in diverse settings, such as laboratory environments and more casual scenarios. Will they scrutinize the extent to which their discoveries stretch?
