When we started this project in late 2018, we had a very ambitious goal: to encapsulate the entire human psychological functioning in response to robots. In other words, we aimed to:
a) capture all thoughts, feelings, and behaviors that humans can experience toward robots from all domains where robots can be encountered (for example, education, hospitality, industry, art, sex) and understand whether these thoughts, feelings, and behaviors can be reduced to some more basic dimensions.
b) measure a wide range of individual differences (that is, personality traits, values, and beliefs) and understand which of them are the most important predictors of people’s thoughts, feelings, and behaviors regarding robots.
c) get a deeper understanding of why the most important individual differences identified in relation to aim b) predict the thoughts, feelings, and behaviors in question.
To accomplish this aim, we decided to use a data-driven, inductive approach - in other words, rather than starting our research with a priori hypotheses, we allowed the data to guide us because the vastness of our research topic meant that we could not easily use previous theorizing to make valid hypotheses. We also used a range of different methods, from qualitative analyses in Studies 1-3 (e.g., iterative categorization) to quantitative analyses in Studies 4-7 (e.g., exploratory factor analyses, exploratory structural equation modelling, various machine learning models). In fact, each of our 7 studies used a different methodology and statistical technique, which allowed us to get diverse and nuanced insights about the questions we studied.
Importantly, because robots are typically defined by experts, and yet we were interested in how humans more generally who are typically non-experts perceive robots, the starting point of our investigation was to develop a definition of robots that is based on their views, and to understand different areas of human activity where robots can be encountered. The definition we developed is too comprehensive so we will not copy paste it here, but it can be found in Table 2 in the article (the article is freely available to everyone). We also found that robots, as described using this definition, can be encountered in 28 different domains (for example, health; social and companionship; security and surveillance; leisure, recreation, and travel; dangerous and/or risky work; art; sex, etc.).
Therefore, to achieve aims a), b), and c) described above, we used examples of robots from each of these 28 domains.
Our main findings for each of the three aims are as follows:
1. Capturing all thoughts, feelings, and behaviors (i.e., psychological processes) that humans can experience toward robots.
We identified 149 unique psychological processes - all of them can be seen in Table 3 in the article. Examples are Attachment (for example, “I would feel attached to this robot”), Robot Rights (for example, “I think this robot should have rights”), Threat (for example, “I feel threatened by this robot”), Dehumanization (for example, “I would feel dehumanized when interacting with this robot”), Avoidance (for example, “I would avoid this robot”), Performance (for example, “This robot can effectively achieve a certain result or a specified outcome”), and Speed (for example, “This robot is fast at what it does”). Importantly, we found that these thoughts, feelings, and behaviors can be categorized into 3 dimensions - positive, negative, and competence-related psychological processes.
Positive psychological processes are all thoughts, feelings, and behaviors that are to some degree positive or have positive connotations (for example, attachment, robot rights), whereas the negative ones are all those that are to some degree negative and have negative connotations (e.g., threat, dehumanization, avoidance). Finally, competence-related psychological processes are all thoughts, feelings, and behaviors linked to how effective robots are in accomplishing any tasks they are specialized in (e.g., performance, speed). Overall, we summarized the three dimensions that emerged in our research under the label of the positive-negative-competence (PNC) model.
2. Identifying key individual differences (that is, personality traits, values, and beliefs) that predict the psychological processes.
From the 79 different individual differences we measured, we found that general risk propensity (that is, people’s susceptibility to risk taking), anthropomorphism (that is, seeing human characteristics in non-living entities), and parental expectations (that is, perfectionistic expectations from one’s parents) were the strongest predictors of the positive psychological processes, with higher levels of these individual differences being associated with more positive thoughts, feelings, and behaviors toward robots.
Moreover, we found that psychopathy (that is, one’s tendency to exploit others for personal gains combined with a lack of empathy), negative affect (that is, how negative people generally feel), anthropomorphism (that is, seeing human characteristics in non-living entities), and expressive suppression (that is, one’s tendency to avoid expressing emotions) were the strongest predictors of the negative psychological processes, with higher levels of these individual differences being associated with more negative thoughts, feelings and behaviors toward robots (note: higher anthropomorphism was associated with both more positive and more negative psychological processes).
Finally, we found that approach temperament (that is, the extent to which people are motivated by rewards and positive outcomes) and security-societal (that is, how important it is to people to live in a society that can defend its citizens and protect them from threats) were the strongest predictors of the competence-related psychological processes, with higher levels of these individual differences being associated with seeing robots as more competent.
3. Explaining why the individual differences outlined under b) above predict the relevant psychological processes.
For the positive psychological processes, general risk propensity was a significant predictor because people higher on this trait valued the risks associated with robot adoption and were curious to see how robots would change the world. Moreover, anthropomorphism was a significant predictor because people higher on this trait generally felt positive toward inanimate entities with human features, and because interacting with such entities helped them fulfil the need to experience strong emotions regularly. Parental expectations was also a significant predictor due to being associated with valuing robots because they were closer to perfection than humans, because they could help humans fulfil their own high expectations, and because they could help humans cope with their own high expectations of themselves.
For the negative psychological processes, psychopathy was a significant predictor because people high on this trait were also more likely to be in a state of activated displeasure (that is, generally feeling scared and upset), had negative feelings toward other people’s inventions, and felt inferior toward technologies they were not proficient in. Furthermore, negative affect was a significant predictor because people high on this trait were more likely to be in the state of activated displeasure (see above). For expressive suppression and anthropomorphism, we did not manage to explain the mechanism behind their relationship with the negative dimension.
Finally, for the competence-related psychological processes, approach temperament was a significant predictor because people high on this trait were more likely to value exceptional skills and competencies. In addition, security societal was a significant predictor because it was associated with people linking advanced technology (for example, robots, machines) with how powerful society is.
Our primary wish is that these findings can enable researchers to systematically study psychological processes involving robots in three ways. First, the positive-negative-competence (PNC) model we developed could help them navigate these processes and identify the ones that may be linked. Second, the model could help them position their research within the appropriate PNC dimension and recognize other processes as well as individual differences potentially worth considering in their studies. Third, the Psychological Responses to Robots (PRR) scale (see Table 4 in the article) we developed could be used to investigate which causal influences shape psychological processes in response to robots because it can measure these processes and thus capture how various experimental manipulations affect them. Overall, we hope these three advancements will lead to theoretical progress by making the research field more organized and easier to navigate, and ultimately prepare us for the imminent transformation of our society in which robots live together with humans.
Our final hope is that our article inspires more scholars in the social sciences to conduct data-driven rather than hypothesis-based research. In these sciences, theorizing is still not sufficiently advanced to tackle some topics, especially the cutting-edge ones such as robots. Therefore, to maximize the effectiveness of the social sciences in helping us to understand and solve many real-world issues and cutting-edge problems, it will be essential to advance data-driven research practices.