The research focus on AI anthropomorphism has gradually shifted from the anthropomorphic design of AI social cues to anthropomorphic descriptions, which is the use of anthropomorphic rhetoric to describe AI features in text. Players read written instructions about a robot and the game rules before collaborating with the robot to complete a math game. The purpose of the human-like framed robot was perceived more positively after experiencing a comprehensible failure than the machine-like framed robot (Roesler, 2023). However, another study that conducted the same experiment found opposite results, indicating that participants were more willing to donate to a robot described using machine-like framing (Onnasch & Roesler, 2019). Moreover, a collaborative robot described with human-like framing significantly increased company employees’ trust in their robot colleagues, but this effect only existed when the human-robot relationship was perceived as collaborative. When employees perceived the relationship with their robot colleagues as competitive, the human-like framing actually reduced their trust in the robot colleagues (Kopp et al., 2022). These mixed findings suggest a need to explore the consequences of anthropomorphic news framing of healthcare AI.

Older adults are generally considered to be technologically inadequate. They tend to adhere more strongly to old habits and resist changes brought about by new technologies (Broady et al., 2010). Evidence for an effect of AI anthropomorphic design on the elderly’s attitudes toward AI is also scarce and somewhat inconsistent. For instance, when robots provide care services to the elderly, older adults tend to exhibit more positive attitudes towards robots with a lower degree of anthropomorphic appearance (Lehmann et al., 2021). Conversely, in the context of robots depicted in movies, the higher the degree of anthropomorphic appearance, the more positive the attitudes of older adult viewers towards the robots (Sundar et al., 2016). Therefore, it is hard to determine how AI anthropomorphic framing influences the attitudes of older adults towards AI.

Psychological distance refers to the subjective experience of something or someone being close to or far from the self, here, and now (Trope & Liberman, 2010). Previous research has also provided empirical evidence for the mediating role of psychological distance between AI anthropomorphism and user responses. Anthropomorphic AI shopping assistants can strengthen the relationship between consumers and brands by reducing the psychological distance between consumers and AI (Jham et al., 2023). Users exhibit more positive attitudes towards highly anthropomorphic AI assistants, and this effect is mediated by psychological distance (Li & Sung, 2021). In chatbot health counseling, users experience a closer psychological distance with highly anthropomorphic chatbots compared to low-anthropomorphic ones, making them more willing to follow the mental health advice provided by the highly anthropomorphic chatbots (Park et al., 2023). Therefore, it can be inferred that, compared to machine-like framing, human-like framing can reduce the psychological distance between older adults and AI, thereby leading to more positive attitudes towards AI among older adults.

Participants for this experiment were recruited from several residential communities in China. Eligibility criteria included: (1) 55 years of age or older; (2) normal or corrected-to-normal visual acuity; (3) no hearing impairments; (4) good cognitive status; and (5) capable of unobstructed communication. A total of 37 individuals met all the requirements and expressed interest in participating. Since 5 individuals were excluded for failing to complete certain steps of the experiment, the final sample comprised of 32 participants, including 17 males and 15 females aged between 55 and 89 (M_age = 59.38). The participants were distributed as follows in terms of their educational background: 15.63% (N = 5) had a lower than a high school education, 37.5% (N = 12) had completed high school, and 46.88% (N = 5) held a bachelor’s degree or higher.

This study collaborated with a professional video production team to create short videos featuring health AI. Information was conveyed through voiceovers and keywords displayed in the video. The difference between the human-like framing and machine-like framing videos was only in the script, while keeping all other aspects of the videos identical. This experiment used two real news reports about surgical robots (People.cn, 2023) and guided robots (People.cn, 2018) as prototypes, with appropriate content reductions and adaptations. Specific information such as surgery names, disease names, and hospital names were omitted to avoid interference with the experimental results.

Manipulation of anthropomorphic framing. The anthropomorphic description of AI primarily takes two forms (Bunz & Braghieri, 2022): (1) Name anthropomorphism, which involves giving the AI a human-like name; (2) Behavior anthropomorphism, which involves describing the AI’s actions as if it were a human. Therefore, this study manipulated the anthropomorphic framing by altering the robot’s name (e.g., mechanical device vs. robot doctor), body parts (e.g., storage system vs. brain), and the description of behaviors (e.g., the control console of the surgical robot can present many minute details vs. the surgeon’s eyes can see many minute details). A total of four health AI short videos were produced, featuring the following combinations: human-like framing with a surgical robot, machine-like framing with a surgical robot, human-like framing with a guided robot, and machine-like framing with a guided robot. All videos were kept to a duration of approximately one minute.

An additional 18 participants (10 males, 8 females, M_age = 61.28) who met the experimental requirements were recruited to test whether the manipulation of the anthropomorphic framing was successful. Each participant watched the four videos in a random order. After watching each video, they were asked to rate the degree of AI anthropomorphism on a 7-point semantic differential scale (Li et al., 2023). The results of the paired samples t-test show that participants perceived the human-like framed surgical robot as significantly more anthropomorphic than the machine-like framed surgical robot (M_human-like = 12.06, M_machine-like= 9.17, p = 0.039 < 0.05). Similarly, participants perceived the human-like framed guided robot as significantly more anthropomorphic than the machine-like framed guided robot (M_human-like = 12.61, M_machine-like = 9.11, p = 0.018 < 0.05). There is no significant difference in the perceived anthropomorphism between the human-like framed surgical robot and the human-like framed guided robot (p = 0.593 > 0.05). Similarly, there is no significant difference in the perceived anthropomorphism between the machine-like framed surgical robot and the machine-like framed guided robot (p = 0.963 > 0.05). These results indicate that the manipulation of the anthropomorphic framing was successful.

This experiment was conducted offline using a tablet to play the videos. Given that many older adults had difficulty accurately comprehend the questionnaire items, researcher verbally presented the questions in a more comprehensible manner and recorded their responses accordingly. To eliminate any influence of the questioning way on participants’ answers, data collection was performed by a single researcher.

The specific procedures are as follows: first, demographics of the participants, including gender and age, was collected. Then, each participant was required to randomly watch one human-like framing video and one machine-like framing video, with different types of AI in each video (e.g., human-like framing and surgical robot, machine-like framing, and guided robot). After watching each video, participants filled out a questionnaire measuring their perception of AI anthropomorphism (Li et al., 2023), psychological distance (Aron et al., 1992), attitudes towards the AI (Li & Sung, 2021), and usage intention (Liu et al., 2022).

All participants signed a consent form and a confidentiality agreement prior to the experiment and were compensated with 20 CNY (approximately 3 USD) upon completion. This experiment was granted approval from the University Research Ethics Committee and adhered strictly to the general ethical guidelines.

Cronbach’s alpha was used to access the reliability of the scales used in this study. The Cronbach’s alpha value of anthropomorphism (Cronbach’s α = 0.868), attitude toward the AI (Cronbach’s α = 0.864), and usage intention (Cronbach’s α = 0.942) all exceed the recommended threshold value of 0.7, which suggests high construct reliability. Psychological distance is measured using a single item, thus there is no need for a reliability analysis.

The results of the paired samples t-test showed that participants perceived the human-like framed health AI as significantly more anthropomorphic than the machine-like framed health AI (M_human-like = 21.94, M_machine-like = 20.16, p < 0.05). Therefore, the anthropomorphic framing was successfully manipulated in this experiment.

The repeated-measures ANOVA revealed a significant effect of anthropomorphic framing on senior participants’ attitudes towards health AI (F (1, 31) = 5.244, p = 0.029 < 0.05). Specifically, compared to machine-like framed health AI (M = 17.16, SD = 3.79), participants had more positive attitudes towards human-like framed health AI (M = 17.97, SD = 2.99).

To explore whether the effect of anthropomorphic framing on AI attitude is mediated by psychological distance, this study used PROCESS Model 4 (parallel mediation model) with a bootstrapped sample of 5000 (Hayes, 2015). As in Figure 2, both the effect of anthropomorphic framing on psychological distance (b = -0.531, p < 0.05) and that of psychological distance on AI attitude (b = -0.427, p < 0.05) were significant. Thus, the effect of anthropomorphic framing on AI attitude was mediated by psychological distance (b = 0.774, Boot SE = 0.415, 95% Boot CI [0.083, 1.705]). Furthermore, the direct effect was not significant (b = 0.774, p = 0.962), suggesting complete mediation in which psychological distance fully explained the relationship between anthropomorphic news framing and AI attitude.

This study treated AI attitude as the independent variable in the regression analysis, with usage intention as the dependent variable, while age, gender, and education level were controlled for as covariates. The results showed that the attitudes of senior participants towards health AI significantly and positively influenced their willingness to use it (β = 0.656, p < 0.05). In other words, a positive attitude towards health AI can transform into a willingness to use it among the older adults.