Modelling Human Psychological Responses to Robots: The Positive, the Negative, and the Competent
Post by Rachel Sharp
The takeaway
As robots become increasingly present in our everyday lives, human reactions to them become more complex. Human responses to robots can be categorized into three dimensions – positive, negative, and competence-related, and predictors can be identified for each response category, thereby establishing the positive-negative-competence (PNC) model.
What's the science?
Robots are becoming increasingly more integrated with human society, resulting in more common and diverse human-robot interactions. Understanding how we think, feel, and react to robots in different spaces informs how we incorporate robots into pre-existing social structures. Previous studies on this topic have not included a comprehensive framework for the range of psychological responses we have, nor the diverse types of robots that humans interact with today. This week in Nature Human Behavior, Krpan, Booth and Damien sought to develop a model that could classify the broad spectrum of psychological responses to robots, organize these responses into grouped patterns (dimensions), and identify thought patterns that most strongly predicted these dimensions.
How did they do it?
The authors conducted 7 studies over 3 phases to develop a framework of robots (Phase 1; studies 1 & 2), develop a categorization for how we respond to robots (Phase 2, studies 3-5), and determine what best predicts these responses and why (Phase 3, studies 6 & 7). Across studies, participants responded to stimuli of robots as images and descriptions across 28 areas of human activity where robots are present. First, a group of participants were asked to generate characteristics they associated with robots, resulting in 277 unique characteristics. Then, a new set of participants grouped these characteristics into categories, from which 5 clusters of robot characteristics were found using hierarchical clustering. The authors then linked the themes of these clusters, forming a general definition of robots. Participants were then given this definition and asked to list all human domains they could think of that robots operate within. A new set of participants were given this comprehensive list of domains and asked to list all thoughts, feelings, and behaviors they could think of regarding robots operating in these domains. Based on their responses, the authors generated prompts to probe the reported thoughts and feelings, and asked new participants to answer them in response to an example of a robot from one of the original 28 domains (i.e. “This robot is like a human”). Responses to these prompts were used to generate statistical models to measure underlying factors representing the diversity of responses.
They found that responses could be most appropriately represented by 3 categories: positive, negative, and competence-related. They then confirmed that the model remains accurate across changes in participant demographic information or robot type. Lastly, the authors trained machine learning models for the positive, negative, and competence dimensions separately and identified the key predictors for each response type. They validated the identified predictors and used parallel mediation analysis to analyze potential mediators of the relationship between a predictor and the associated PNC dimension. For example, general risk propensity was a predictor for the positive dimension because people who scored highly for this trait valued the risks associated with using robots in society and were curious about how they would influence the world.
What did they find?
The authors established the positive-negative-competence (PNC) model to represent the spread of human psychological responses to robot representation. They also identified unique predictors for each of these dimensions and were able to identify mediators for 3/3 of the positive dimension predictors, 2/4 of the negative dimension predictors, and 2/2 of the competence dimension predictors. The key positive predictors of each dimension with an example of an identified mediator of the predictor in parenthesis, where applicable, were: general risk propensity (valued risks of robot use), anthropomorphism (felt positively towards inanimate objects with human features), and parental expectations (felt robots could help humans fulfill their high expectations). The key predictors for the negative dimension were trait negative affect (were more likely to be in a state of displeasure), psychopathy (felt inferior towards technologies they were not proficient in), anthropomorphism (no mediator found), and expressive suppression (no mediator found). For the competence dimension, key predictors were approach temperament (valued exceptional skills and competencies) and security-societal (linked advanced technologies with the degree of societal advancement).
What's the impact?
While previous research on human-robot relations has examined psychological responses to robots, this is the first study to investigate the spectrum of human psychological reactions under a comprehensive construct. Importantly, the authors illustrated that while the spectrum of human responses to robots is diverse, they can be explained by three dimensions of psychological processing: positive, negative, and competence-related. The proposed model allows future research to measure responses to robots more easily and accurately, helping us to construct a framework to understand how humans think, feel, and react to robots encountered in society.