Researchers from Carnegie Mellon University have found that AI has the potential to change the way that humans work, but it will not be effective if humans do not trust it. The team’s study reviewed 20 years of research on human trust in AI. This study helps organizations implement AI in the most effective way. AI's representation and capabilities contribute to a human’s trust in the system.
Source: Pixabay
In the study, researchers addressed the dynamic nature of trust. The elements of trust in AI include the observed role of tangibility, transparency and reliability. They also looked at the role of tangibility, transparency and reliability and they considered task characteristics, immediacy behaviors and anthropomorphism.
The team searched Google Scholar and three databases for articles on human trust in AI from 1999-2019. They found over 300 potential articles on various topics including organizational behavior, human-computer interactions, robot-human interactions, information systems, information technology and engineering. For their research, the team focused on 150 articles.
Results showed that the representation of AI plays an important role in the nature of the cognitive trust that people develop. It was also dependent on the kind of AI being used. For robotic AI, the trajectory of developing trust is similar to how humans create trust in human relationships. Trust starts low and increases with more experience with an AI robot. The opposite was true with virtual or embedded AI, where there is high initial trust that wanes over time.
The level of machine intelligence in AI may moderate the development of cognitive trust with a high level of intelligence can lead to higher trust with experience. Transparency is also important in establishing trust in virtual or embedded AI. The influential factors over emotional trust differ from cognitive trust.
Anthropomorphism was important to the development of trust in AI, but its effect is different depending on the kind of AI. With robotic AI, anthropomorphism had mixed effects. People liked anthropomorphic robots more than mechanical-looking robots. Human-like robots spark feelings of discomfort and a sense of eeriness.
The team proposed a framework that considered the form of AI used, level of machine intelligence, behaviors like responsiveness and reliability. All of these factors have an influence on people’s cognitive and emotional trust.
A paper on this research was published in the Academy of Management Annals.
