Narratives about artificial intelligence (AI) entangle autonomy, the capacity to self-govern, with sentience, the capacity to sense and feel. AI agents that perform tasks autonomously and companions that recognize and express emotions may activate mental models of autonomy and sentience, respectively, provoking distinct reactions. To examine this possibility, we conducted three pilot studies (N = 374) and four preregistered vignette experiments describing an AI as autonomous, sentient, both, or neither (N = 2,702). Activating a mental model of sentience increased general mind perception (cognition and emotion) and moral consideration more than autonomy, but autonomy increased perceived threat more than sentience. Sentience also increased perceived autonomy more than vice versa. Based on a within-paper meta-analysis, sentience changed reactions more than autonomy on average. By disentangling different mental models of AI, we can study human-AI interaction with more precision to better navigate the detailed design of anthropomorphized AI and prompting interfaces.

How do robot designers anthropomorphize their own creations? Because robot designers have the ability to alter the robot, identify as its creator, and understand their robot’s internal makeup, their process of anthropomorphism and its outcomes may be different from that of the typical robot user. We investigate this research question in the domain of combat robots, where anthropomorphism is critical to decision-making, communication, and trust in high-stakes, high-emotion combat situations faced by robot-soldier teams. We conducted an in-depth case study of a university’s student-led Combat Robotics Design Team over the design, construction, testing, and competition phases for their competitive combat robot. Based on inductive computational and human coding of extensive field notes, supplemented with interviews and surveys, we found that these robot designers anthropomorphize for three purposes. First, they anthropomorphize the bot to manage impressions of it within their team and to outsiders like competitors, spectators, and sponsors, specifically presenting it as a warrior. Second, they anthropomorphize it like a child, a pet, or simply treat it as a non-anthropomorphic mechanical set of parts as a way to calibrate their relationship and attach with or detach from their own creation. Third, they anthropomorphize the bots to assign blame either blaming it, its parts, or others based on their expectations of whether it is performing based on how they designed it. We conclude with implications for anthropomorphism by robot designers and application to military robot design.

Artificial intelligence (AI) agents are increasingly being used as teammates, not just as tools, across many domains, and teammates’ moral behaviour can alter impressions of themselves and the team. How good, powerful, and active is an AI versus human team member engaging in an ethical or unethical behaviour? How good, powerful, and active is their team? To address these questions, we conduct four studies across three domains (chess, esports, and poetry composition) where participants rate their impressions of team members and teams presented in a scenario. In the scenario, a member of a hybrid team of 2 humans and 2 AIs is presented with an opportunity to cheat, and either does or does not. We manipulate which team member (AI vs human) is acting and the morality of that action (non-cheating vs cheating). Across the studies, results show that ethical behaviour significantly increases the goodness of the human more than the AI, and unethical behaviour significantly increases the power of the AI more than the human. However, there were no systematic human versus AI differences on team impressions.

The rise of complex AI systems in healthcare and other sectors has led to a growing area of research called Explainable AI (XAI) designed to increase transparency. In this area, quantitative and qualitative studies focus on improving user trust and task performance by providing system- and prediction-level XAI features. We analyze stakeholder engagement events (interviews and workshops) on the use of AI for kidney transplantation. From this we identify themes which we use to frame a scoping literature review on current XAI features. The stakeholder engagement process lasted over nine months covering three stakeholder group's workflows, determining where AI could intervene and assessing a mock XAI decision support system. Based on the stakeholder engagement, we identify four major themes relevant to designing XAI systems - 1) use of AI predictions, 2) information included in AI predictions, 3) personalization of AI predictions for individual differences, and 4) customizing AI predictions for specific cases. Using these themes, our scoping literature review finds that providing AI predictions before, during, or after decision-making could be beneficial depending on the complexity of the stakeholder's task. Additionally, expert stakeholders like surgeons prefer minimal to no XAI features, AI prediction, and uncertainty estimates for easy use cases. However, almost all stakeholders prefer to have optional XAI features to review when needed, especially in hard-to-predict cases. The literature also suggests that providing both system- and prediction-level information is necessary to build the user's mental model of the system appropriately. Although XAI features improve users' trust in the system, human-AI team performance is not always enhanced. Overall, stakeholders prefer to have agency over the XAI interface to control the level of information based on their needs and task complexity. We conclude with suggestions for future research, especially on customizing XAI features based on preferences and tasks.

Interest in and ownership of smart home voice assistants like Amazon Alexa and Google Home devices have exponentially increased in recent years. Many people may purchase or be gifted such devices without knowing their potential for connecting with other home technology, listening to private conversations, sharing information with companies, and creating problems due to misunderstanding vocal commands or technological capabilities. Concerns and worries about these devices may be exacerbated over time or by a specific incident. To understand reactions to such situations, we conducted semi-structured in-depth interviews with 10 people who reported different types of worrying incidents with a range of smart home devices and their reactions to reduce that worry. Conducting a thematic coding analysis, we detail how each case study shows a person’s worries about their smart home technology developed vis-à-vis the incident or over time, and their strategies to alleviate their worry. The two dominant reactions were restricted acceptance or discontinuance of the smart home technology, while three other interviews revealed nuanced reactions on the acceptance-rejection continuum. For each interviewee, we highlight their technology use, any major incidents, and their psychological processes leading up to their actions to reduce worry. This provides an in-depth look at worry around smart home technology products themselves, not their ability to perform, and how discontinuance, restricted acceptance, and other reactions reduce those worries.

What are the relationships between knowledge of, perceived benefits, adoption of, and use of smart home products? To explore this question, in our first two studies we focus on the general population’s perceptions of benefits across many types of smart home products by creating a corpus of smart home product descriptions. Study 1 (n = 399) shows that previous product knowledge influences a range of perceived benefits. Study 2 (n = 242) demonstrates which benefits increase non-owners’ likelihood of adopting these products. In study 3, we longitudinally survey eight residents in living laboratory houses equipped with 10 integrated smart home products. We find over a year access to the products increases perceptions of their benefits, but does not increase their actual use. Collectively, these studies contribute to an increased understanding of the relationship among benefits, use, and adoption of this emerging technology.

Do people personally witness artificial intelligence (AI) committing moral wrongs? If so, what kinds of moral wrong and what situations produce these? To address these questions, respondents selected one of six prompt questions, each based on a moral foundation violation, asking about a personally-witnessed interaction with an AI resulting in a moral victim (victim prompts) or where the AI seemed to engage in immoral actions (action prompt). Respondents then answered their selected question in an open-ended response. In conjunction with liberty/privacy and purity moral foundations and across both victim and action prompts, respondents most frequently reported moral violations as two types of exposure by AIs: their personal information being exposed (31%) and people’s exposure to undesirable content (20%). AIs expose people’s personal information to their colleagues, close relations, and online due to information sharing across devices, people in proximity of audio devices, and simple accidents. AIs expose people, often children, to undesirable content such as nudity, pornography, violence, and profanity due to their proximity to audio devices and to seemly purposeful action. We argue that the prominence in reporting these types of exposure may be due to their frequent occurrence on personal and home devices. This suggests that research on AI ethics should not only focus on the prototypically harmful moral dilemmas (e.g., autonomous vehicle deciding whom to sacrifice) but everyday interactions with personal technology.