When an AI Doctor Gets Personal: The Effects of Social and Medical Individuation in Encounters With Human and AI Doctors

Interpersonal Knowledge and Expressed Individuation

Interpersonal knowledge is “the impression one has about the way a specific target individual responds in a unique fashion to (someone) as distinct from the way that the target individual responds to anyone else” (Walther, 2019, p. 377). It often manifests in dyadic conversations, such as those between patients and their doctors, through message individuation: a communication process in which the patient feels that a doctor responds uniquely by recalling their personal information. As an outcome, individuation can make patients feel special as they are differentiated by their doctors from other people (Maslach, 1974).

Previous studies have argued that individuation can be produced by making references to one’s physical and social information, such as name and location (Maslach, 1974). However, obtaining information about others without actually interacting with them is insufficient to accumulate interpersonal knowledge, thus limiting uncertainty reduction and potential relationship benefits within any particular dyad. To individuate, communicationally, is to introduce or refer to something distinctive that a conversation partner expressed in a previous utterance.

There are three communication-related components to make individuation work in interpersonal communications (Walther, 2021). First, individuation involves information memory and recall. An individual should be able to retrieve the information about a conversational partner from memory and tailor messages that includes that information. Second, individuation is often achieved via reciprocal and interactive communication over time; it takes time and the exchange of many messages to accumulate interpersonal knowledge, particularly in text-based digital interactions (see Walther, 1992 for a review). Third, individuation requires interactivity and message contingency across conversations. Responding to someone in a subsequent conversation by referencing information from a prior conversation imbues a sense of interconnected interaction, demonstrates interpersonal knowledge, and enhances relationships.

Individuation and Patient Satisfaction

Medical consultation is one context that needs individuation (Longoni et al., 2019). Experiencing individuation in a doctor’s visit is likely to generate various positive outcomes, such as positive evaluation of the doctor, liking of the doctor, as well as patient satisfaction and compliance. Among these, we focus on patient satisfaction, which has been found to be essential for patients’ compliance and positive doctor-patient relationships (Carr-Hill, 1992). Satisfaction with a physician is based on whether a medical encounter meets the patient’s needs and expectations (Jackson et al., 2001; Korsch et al., 1968). Individuation, in this respect, is likely to increase patients’ satisfaction.

Considering that individuation is primarily about recalling and mentioning information about another person obtained in previous conversations, one mechanism that may influence the evaluation of individuation is perceived effort exerted by the conversationalist. We define perceived effort as a patient’s perception of the cognitive resources expended by a doctor to memorize, retrieve, and overtly reference information unique to a patient. We predict that differentiating the patient from others by recalling their unique characteristics obtained from earlier interactions will result in higher perceived effort by the doctor. This perceived effort will, in turn, enhance patient satisfaction.

Another possible positive outcome resulting from individuation is perceived relational closeness, which is defined as patients’ perceptions that their interaction in a medical encounter has cultivated a beneficial and lasting relationship with the physician (Hennig-Thurau et al., 2002). Given that good quality care is characterized as individualized (Attree, 2001), we predict that recalling the patient’s personal information will foster the feeling of relational closeness from the patient’s perspective, which in turn would enhance patient satisfaction.

Not all outcomes of individuation are positive, however. Previous research has identified perceived privacy intrusiveness as a primary concern associated with individuation (Chen et al., 2021). Drawing on prior conceptualizations of privacy (Warren & Brandeis, 1890), we define perceived intrusiveness to privacy as the extent to which a patient believes that their personal information space is intruded upon by others, violating their right to be left alone. In line with previous studies (Chen et al., 2021), we predict that individuation, which involves using personal information for personalized and contingent conversations, will result in greater perception of privacy intrusiveness, thereby lowering patient satisfaction. Together, we propose the following three mediation hypotheses to explore the effect of individuation on patient satisfaction:

Doctor Identity as a Moderator

The identity of the doctor providing individuation matters. The effect of individuation has been demonstrated to be psychologically distinct for human doctors versus AI doctors (Chen et al., 2021; Yun et al., 2021). For example, a recent study found that patients feel special when a human doctor remembers their personal information, but they feel intruded upon when an AI doctor remembers it (Chen et al., 2021). This difference is further explained by neural evidence showing that the brain area associated with implicit apathy is activated when interacting with an individuating AI medical system, whereas the brain area pertaining to prosociality is triggered when talking with an individuating human doctor (Yun et al., 2021).

We define the AI doctor in our study as a conversational agent linked to an expert system offering diagnoses and medical advice to patients with minimal human intervention. This type of AI doctor often manifests as a symptom checker chatbot where patients interact by clicking buttons and typing responses to the system’s questions. The system then evaluates the input and provides a comprehensive evaluation and recommendation at the conclusion of the conversation (Kolata, 2023).

Practically, patients are more likely to encounter AI-assisted human doctors in current medical consultations, that is, a human doctor assisted by an AI medical system (Topol, 2019). But few studies have examined how users perceive individuation from an AI-assisted human doctor. One experiment by Chen et al. (2021) suggests that patients do not perceive AI-assisted human doctors to be any different from unaided human doctors. Participants who interacted with human doctors, with or without AI assistance, showed similar patterns of low perceived intrusiveness and high patient compliance, compared to those who interacted with an AI doctor. It may be that patients may not notice or care about AI assistance as long as there is a human doctor, even if they are told that the human doctor is collaborating with a machine.

Other research on AI-mediated communication renders different attributions about AI-assisted human performance. Being aware of the presence of AI in a human’s work can undermine evaluations of the person, probably due to perceived laziness when attempting to personalize messages (Jakesch et al., 2019). Research on collaboration between humans and AI generates inconsistent evaluations, alternatively showing no effect (Chen et al., 2021), or ill effect (Jakesch et al., 2019), on perceptions of humans individuating messages for others.

Yet a third, positive effect on perceptions of AI-assisted humans is noted in the literature (Molina & Sundar, 2022; Waddell, 2019). The construct of mutual augmentation suggests that an AI medical system can augment a human’s capability, and humans can assist AI to generate better collaborative outcomes (Sundar, 2020). If individuating is effortless for machines but cognitively effortful for humans, then an AI-assisted human doctor may be able to individuate patients with relative ease, reaping the positive relational outcomes without promoting the perception that individuating information is merely data points.

To explore the individuation effect further as a function of a doctor’s identity, this study examined three types of doctors, AI doctors, human doctors, and AI-assisted human doctors, as they produced individuated or non-individuated conversations, to examine how individuation interacts with a doctor’s identity in influencing patient satisfaction. Although our questions primarily address the comparison of human to AI doctors, it has become increasingly common for human doctors to consult AI medical systems for suggestions and insights (Topol, 2019; Wang et al., 2021). Given that the difference is mostly observed between human and AI doctors (Chen et al., 2021), we incorporate the AI-assisted doctor condition to explore how individuation effects vary when participants are cognizant of the presence of both entities: a human doctor and the AI medical system. The newly added condition is meaningful as it can help test whether the theory of interpersonal knowledge holds true when the recall of interpersonal knowledge can be facilitated by a machine.

Next, we delineate the psychological mechanisms to explain how doctor identity moderates the effect of individuation on patient satisfaction via three mediators—perceived effort, perceived relational closeness, and perceived intrusiveness—described in the sections that follow.

The Mediating Role of Perceived Effort

Given that individuation requires the recall of information from previous conversations, the perceived effort exerted by the doctor in providing individuating conversations is likely to affect patient satisfaction. Prior research supports this argument, showing that seeing others putting effort into decision-making leads to greater satisfaction with the decision process (Bechwati & Xia, 2003). However, it depends on who exerts the effort. Given that machines are inherently good at data storage and retrieval, patients may attribute less effort to machines compared to humans, especially considering that patients’ idiosyncrasies are simply data points to a machine. From this perspective, individuation by an AI doctor may be perceived as involving less effort than when individuation is attempted by a human doctor or an AI-assisted human doctor. This contention is supported by research showing that individuals perceive electronic decision aids as exerting less effort than human decision aids in career service (Bechwati & Xia, 2003). Unlike AI, humans have limited cognitive resources and are known to be “cognitive misers” (Miller, 1994). It may require greater effort for a human doctor to recall and reference individuating information about a particular patient than it does for an AI doctor. Based on this rationale, we propose the following hypothesis:

The Mediating Role of Perceived Relational Closeness

Similar to perceived effort, the effect of individuation on perceived relational closeness also depends on the doctor’s identity. Considering that humans are generally social, and that relationship-building seems intuitively exclusive to living beings, individuation from a human doctor is more likely to result in higher perceived relational closeness than that from an AI doctor. Moreover, AI doctors’ efforts to build relationships with their patients may even be experienced as a threat to human identity and uniqueness, leading to negative evaluations of the machine. This possibility reflects the Uncanny Valley of Mind concept, which posits that individuals respond negatively to machines when they become too humanlike (Gray & Wegner, 2012).

It is worth noting that there are increasingly more chatbots developed for relational interaction, such as Replika and Woebot. It is not rare to hear that individuals develop intimate relationships with non-human agents (Leo-Liu & Wu-Ouyang, 2022; Wilkinson & Frost, 2022), and benefit from caring conversations with an AI chatbot (Bram, 2022; Meng & Dai, 2021). From this perspective, interacting with an AI doctor may increase perceived relational closeness. Given the relationship building opportunities with both AI and human doctors, we ask the following research question:

The Mediating Role of Perceived Privacy Intrusiveness

Aside from the positive outcomes, perception of privacy intrusiveness is one potential negative consequence of individuation. Previous research has pointed out that evaluation of privacy intrusiveness can be formed through heuristics (Sundar et al., 2013). One cognitive heuristic that is likely to be activated when interacting with a machine is the machine heuristic, which is a cognitive appraisal that a computer (or other machine) is more objective, and therefore more reliable, than humans (Sundar & Kim, 2019). If individuals use characteristics of machines to understand their interaction with an AI doctor, there may be less perceived privacy intrusiveness because individuals may think that machines, unlike humans, do not inadvertently discuss, disclose, or gossip. Based on this rationale, an individuating AI doctor may stimulate lesser perceptions of privacy intrusiveness than a human doctor.

However, empirical studies do not seem to agree with this speculation. For example, Chen et al. (2021) found that individuation from an AI doctor was perceived as more intrusive to one’s privacy compared to the same individuation from a human doctor and an AI-assisted human doctor. Likewise, Aktan et al. (2022) pointed out that individuals tend to have less trust in data security from AI-based therapy than human therapist. Considering that individuation involves the use of one’s personal information, and AI medical systems are seldom explicit about data storage and usage, the lack of transparency may result in greater perceived intrusiveness to privacy. Based on existing findings, we propose the following hypothesis:

Information Type as a Moderator

The effects of individuation may differ depending on the type of information used for individuation. There are two types of interpersonal knowledge that doctors can obtain from their previous communications with patients: medical and social. Specifically, medical information refers to information pertaining to one’s medical history, health-related behaviors, and health status, such as smoking history and family history of sleep disorders. In contrast, social information refers to details about one’s social life, such as occupation, personal preferences, and relationships.

Patients’ information might be data points to a machine, making the retrieval of medical and social information equally effortful. However, the recall of medical versus social information may entail different amounts of cognitive effort for a human doctor. Human doctors may primarily focus on medical information, which is mirrored in lab reports and other recent analyses that doctors often have at hand. It may appear easier for a human doctor to retrieve and to integrate medical individuation into a conversation than to store and retrieve a patient’s social information. Cognitive science research suggests that people tend to have lower rates of information recall when the particular information is stored at their fingertips (Sparrow et al., 2011). Applying this notion to the current study, when a human doctor recalls patients’ medical information, it may be perceived as requiring little effort, since a patient’s medical information is stored externally and readily accessible. Based on this rationale, we proposed the following hypothesis:

Similarly, we predict that information type can affect perceived relational closeness. While medical information is commonly used for diagnosis, social information is often used for rapport and relationship-building (Meng & Dai, 2021). Therefore, if an AI doctor recalls and refers to a patient’s socially individuating information during their conversation, it is expected to produce greater relational closeness than mentioning medically individuating information. Given that there are relationship building opportunities with both human and AI doctors, and it is unclear how information type may moderate the interaction effect between individuation and doctor identity on perceived relational closeness, we propose the following research question:

In addition, patients’ perception of privacy intrusiveness may depend on the type of information used in the individuated conversation. While both medical and social information are valuable in helping doctors know their patients better, medical information is typically more frequently requested and used for diagnosis and medical treatment compared to social information. Despite the presence of various risk factors or potentially stigmatizing data in medical information, professionals are bound by ethical guidelines and legal obligations to maintain personal medical data confidential (see Health Information Privacy, n.d).

In contrast, rules for protecting and sharing social information are less clear in the healthcare context. As patients may not expect sharing certain social information, such as their cooking habits at home or smartphone use before bedtime, to be recorded by their doctors, we hypothesize that patients may perceive the recall of social information as more intrusive to their privacy than the retrieval of medical information. This speculation finds support in at least one early study, which discovered that employees perceive a greater invasion of privacy when personality information was disclosed without their permission compared to job performance-related information in a workplace context (Tolchinsky et al., 1981). Thus, we predict that social individuation by an AI doctor is likely to result in higher perceived privacy intrusiveness than other conditions, as stated below.

Privacy Control as a Moderator

Given that individuation indispensably involves the discussion of one’s personal information, the effects of individuation may be further subject to differences in patients’ sense of control over their identifiable information. Will a doctor share it with others? Will it become part of a networked database? These natural concerns may be mitigated by privacy control, often defined as the extent to which patients are provided a choice over the release and access of their private data (Agaku et al., 2014; Cho et al., 2020). Given that this study focuses on both privacy protection initiated by the system and choices made by the users, we take the technological affordances perspective by defining the provision of privacy control as the presence of interface features that allow users to indicate their preference for the release and access of their personal medical records.

Whether to engage with privacy control depends on the calculus of behaviors. According to the privacy calculus approach, individuals tend to evaluate the potential risks and benefits of certain behaviors when it comes to information disclosure (Laufer & Wolfe, 1977). In the healthcare context, when patients weigh perceived risks more than perceived benefits, they are more likely to restrict access to their personal medical records. By contrast, when patients perceive the benefits of disclosing one’s information as exceeding the risks to privacy, they are less likely to take action to protect their personal information.

The calculus of privacy-related behaviors may be further affected by the agent who provides the services and is limited by it. If AI systems work best when they can access historical and networked data, limitations on their access to those data due to privacy control may be disadvantageous. In other words, AI doctors rely on patients’ social and medical data for individualized conversation. When users decide to delete their data, it undermines the AI’s ability to provide personalized conversation, akin to erasing memory for a machine. If there is no memory about the patient, individuation is not possible, denying patients the potential benefit of having a long-term and more meaningful relationship with the AI doctor. If the patient sees the opportunity to build a long-term relationship with an AI doctor, which uses social information rather than medical information for individuation, this effect may be attenuated by the provision of privacy control.

One benefit of privacy control is that it may mitigate perceived privacy intrusiveness when receiving social individuation by an AI doctor. This speculation aligns with a recent study on virtual try-on apps, which found that giving users control over the privacy settings significantly reduces the app’s perceived intrusiveness (Feng & Xie, 2019).

Different from AI doctors, human doctors’ memory of their patients will remain unaffected by users’ privacy control choice. Whether patients choose to retain or delete their data from the platform, human doctors may still recall the patient’s information. It is primarily because humans’ memories reside in their heads and are not erased by the push of a button according to the patients’ preferences. The inability to delete one’s memory may lead to less belief that privacy control can be affected by a human doctor compared to an AI doctor. From this perspective, the provision of privacy control is less likely to reduce perceived privacy intrusiveness or increase perceived relational closeness from an individuating human doctor. Based on the preceding rationale, we propose the following research question and hypothesis:

We present the study model in Figure 1.

OPEN IN VIEWER

Figure 1.

Study model.

Method

To answer the research questions and test our hypothesis, we conducted a 3 (doctor identity: AI doctor vs. human doctor vs. AI-assisted human doctor) × 2 (individuation: presence vs. absence) × 2 (information type: social vs. medical) × 2 (privacy control: presence vs. absence) between-subjects online experiment.

Procedure and Stimuli

The study was presented to prospective participants as an online conversation with a doctor to check one’s wellness after the first major U.S. wave of the COVID-19 pandemic. Because many people experienced detrimental lifestyle and mental health issues during the pandemic (World Health Organization, 2022), and numerous telemedicine practices were quickly adopted during that time (Davenport & Kalakota, 2019; Yang et al., 2019), a virtual wellness check-up was a plausible scenario for prospective participants.

The study involved two separate interactions between participants and the doctor. The doctor was presented consistently across both interactions as either a human, an AI, or an AI-assisted human. We developed a chatbot prototype using FlowXO, an online chatbot software, to enable text-based conversations between the patient and doctor. A virtual doctor was enabled by a pre-generated script, mostly comprised of questions that sought participants’ individual responses, which were recorded for potential use in personalizing the second conversation. The scripting system was not static, but rather, could carry forward individual participants’ responses from the first visit into the second visit, depending on a class of question/answer responses by the researchers. By this technique, in the second visit, the doctor could be made to appear to remember and mention the particular responses from the first visit, according to requirements of the experimental condition. Whatever answers the participants provided about their respective social and medical characteristics in the first visit could be “recalled” through the scripting in the second visit, for example, or the doctor would not recall them and ask for the information a second time. These scripting variations reflected the dynamics of interpersonal knowledge theory and its precepts regarding how conversations reflect memory of individuals from one conversation to another. The script was based on a recent study by Chen et al. (2021), and we modified it to fit the context of a wellness check-up. We visualize the flow of interactions in Figure 2.

OPEN IN VIEWER

Figure 2.

Flow of interactions in a two-phase doctor visit.

First Doctor Visit

Manipulation of Doctor Identity

After checking in, participants were randomly assigned to interact with one of the three doctor identities: a human doctor, an AI doctor, or an AI-assisted human doctor. We manipulated doctor identity by providing different text-based descriptions of the doctor’s background in the chat (see Table 1). The system prompted participants to read the description while the receptionist connected them to the doctor.

OPEN IN VIEWER

Table 1.

Manipulation of Doctor Identity.

Doctor identity	Introduction
AI doctor	Thanks for completing the form. Please bear with us until we connect you to AI Dr. Alex. Meanwhile, you can read a brief introduction of AI Dr. Alex.
	Dr. Alex is an AI medical system developed at the University of Pittsburgh School of Medicine. It has been trained based on deep learning algorithms for providing preventive care and medicine. AI Dr. Alex’s particular expertise includes nutrition, fitness, and lifestyle.
Human doctor	Thanks for completing the form. Please bear with us until we connect you to Dr. Alex. Meanwhile, you can read a brief introduction of Dr. Alex.
	Dr. Alex received a medical degree from the University of Pittsburgh School of Medicine in 2005 and has been board certified in preventive care and medicine. Dr. Alex’s particular expertise includes nutrition, fitness, and lifestyle.
AI-assisted human doctor	Thanks for completing the form. Please bear with us until we connect you to Dr. Alex that will be assisted by an AI medical system. Meanwhile, you can read a brief introduction of Dr. Alex and the AI medical system.
	Dr. Alex is a board-certified preventive medicine specialist who received a medical degree from the University of Pittsburgh School of Medicine in 2005. Dr. Alex’s particular expertise includes nutrition, fitness, and lifestyle.
	The AI medical system assisting Dr. Alex is based on deep learning algorithms for providing preventive care and medicine.

During the first doctor visit, the doctor asked questions about the patients’ dietary habits, sleep patterns, mental health, and physical exercise during the COVID-19 pandemic. The doctor also asked or noted certain unobtrusive social information to facilitate interaction (such as the participant’s preferred name) or of tangential relevance to preventive healthcare. After the chat, the doctor provided health advice adapted from US Centers for Disease Control and Prevention (CDC; www.cdc.gov), on diet, exercise, and mental health.

Second Doctor Visit

After about 2 weeks, participants were reminded through Cloud Research to participate in the second doctor visit. In this visit, participants interacted with the same doctor they encountered in the first visit; the doctor identity was presented again at the beginning of the interaction, with the description of the doctor’s background being the same as the first visit. The doctor chatted with the participants about recent changes to their health and wellbeing.

Manipulation of Individuation and Information Type

During the interaction, the doctor either recalled the patient’s information from the first visit and employed it in their conversation or did not appear to remember it. Within the condition of information recall, the doctor personalized the conversation by either mentioning the patients’ social or medical information. Social information that the doctor worked into the conversation included items such as their occupation and how much screen time before bed they had reported in the prior visit. Medical information included information pertaining to diet, exercise, sleep, mental health, and lifestyle. The individuation manipulations appear in Table 2. After the conversation with the patient, the doctor provided health advice similar to the first visit.

OPEN IN VIEWER

Table 2.

Manipulation of Individuation and Information Type.

Individuation (medical information)	Non-individuation (medical information)	Individuation (social information)	Non-individuation (social information)
It seems like you were [not/slightly/moderately/severely] depressed about 2 weeks ago. How about now?	Do you feel down, depressed, or hopeless in the past 2 weeks?	I remember you are [patient’s occupation], right?	Could you please remind me what you do for a living?
I remember you [have/do not have] a family history of sleep disorder, correct?	Refresh my memory. Do any of your relatives have a sleep disorder?	Last time, you mentioned that you [have/do not have] a good relationship with your family. Is it still the case?	I do not recall whether we discussed your family relationship previously. Is it ok if I ask a few questions?
It seems that you did not consume much [the least consumed food group] last time. Have you followed the recommendation to eat from all five food groups?	I do not remember whether we discussed your diet previously. Is it okay if I ask a few questions about it?	I remember you [did not/did] cook at home. Is it correct?	I do not have this information on top of my head. How often do you cook at home?
Last time, you mentioned that you [do not have/have] food allergies. Is this correct?	I do not recall our discussion of food allergies last time. Are you allergic to any food?	Earlier, you said that you [have/do not have] screen time before bed. Is it still the case?	Refresh my memory. Do you use your smartphone right before sleep?
I remember you [are/are not] in a healthy weight range. Any changes in your weight recently?	Could you please remind me of your weight?	Have you been able to [do your favorite activity for relaxation] lately?	I may have forgotten. Could you please tell me again what your favorite activity is when under stress?

Manipulation of Privacy Control

The experimental manipulation for privacy control appeared near the end of the conversation. For half of the participants, the doctor asked participants about their preference for medical information archiving, using two questions: “Do you want me to put today’s visit in your record?” and “Do you want to keep your personal medical information on our platform? The patient was provided with yes/no response options to each question. For the other half of participants, who were not offered privacy control opportunities, the doctor thanked them for their visit and directed them to chat with the receptionist. Participants then received a code to complete the rest of the questionnaires.

A debriefing followed the questions in the post-test questionnaire. It explained that there was no actual human or AI doctor involved in the wellness checkups. It explained that all participants interacted with pre-programed bots that differed only in whether they were labeled as a human doctor, AI doctor, or AI-assisted human, in order to keep the interactions constant within and across the three conditions. Following this, we provided resources adapted from the CDC with additional recommendations about diet, exercise, and mental health practices.

Results

Testing the Moderating Role of Doctor Identity

We used Model 7 to test the interaction effect between individuation and doctor identity on patient satisfaction through perceived effort (H4). Results from Model 7 supported this hypothesis: Index = −0.80, SE = 0.34, 95% CI: [−1.50, −0.16]. Specifically, individuation by an AI doctor (M = 5.59, SD = 0.87) was perceived as more effortful compared to individuation by a human doctor (M = 4.85, SD = 1.28). The mean difference was significant at .05 level. We present the interaction effect in Figure 3.

OPEN IN VIEWER

Figure 3.

Interaction effect between individuation and doctor identity on perceived effort.

RQ1 asked how individuation interacts with doctor identity in influencing perceived relational closeness and further affects patient satisfaction. Results from Model 7 did not support this hypothesis when comparing the AI and AI-assisted human doctors, Index = −0.47, SE = 0.31, 95% CI: [−1.09, 0.13], and when comparing the AI and human doctors, Index = −0.06, SE = 0.39, 95% CI: [−0.83, 0.68].

H5 stated that individuation by an AI doctor would be perceived as more privacy intrusive compared to individuation by a human doctor. Perceived privacy intrusiveness, in turn, would lower patient satisfaction. However, we did not find supporting evidence from the data when comparing AI and AI-assisted human doctors, Index = 0.04, SE = 0.06, 95% CI: [−0.06, 0.18], and when comparing AI and human doctors, Index = 0.05, SE = 0.07, 95% CI: [−0.06, 0.21]. Thus, H5 was not supported.

Testing the Moderating Role of Information Type

H6 predicted that information type moderates the interaction between individuation and doctor identity on perceived effort, which would be positively associated with patient satisfaction. Results from Model 11 revealed a significant conditional indirect effect when the information type was social, Index = −0.97, SE = 0.35, 95% CI: [−1.64, −0.30], rather than medical information: Index = −0.37, SE = 0.47, 95% CI: [−1.33, 0.48]. A close examination of the relative conditional indirect effects revealed that social individuation by an AI doctor was perceived to be more effortful than by a human doctor, leading to higher patient satisfaction. Thus, H6 was supported. Table 3 shows the details of the conditional indirect effects by information type.

OPEN IN VIEWER

Table 3.

Conditional Indirect Effects on Patient Satisfaction through Perceived Effort by Information Type.

Doctor identity	Individuation	Information type	Effect (B)	SE	95% CI
Doctor identity → Perceived effort → Patient satisfaction
AI-assisted human (vs. AI doctor)	No	Medical	−0.05	0.26	[−0.58, 0.44]
AI-assisted human (vs. AI doctor)	No	Social	0.27	0.27	[−0.24, 0.82]
AI-assisted human (vs. AI doctor)	Yes	Medical	−0.15	0.20	[−0.53, 0.25]
AI-assisted human (vs. AI doctor)	Yes	Social	−0.31	0.16	[−0.63, −0.01]
Indices of conditional moderated mediation
Medical information: Index = −0.10, SE = 0.33, 95% CI [−0.74, 0.54]
Social information: Index = −0.58, SE = 0.32, 95% CI [−1.22, 0.01]
Human (vs. AI doctor)	No	Medical	−0.07	0.32	[−0.69, 0.57]
Human (vs. AI doctor)	No	Social	0.49	0.28	[−0.05, 1.06]
Human (vs. AI doctor)	Yes	Medical	−0.44	0.34	[−1.18, 0.13]
Human (vs. AI doctor)	Yes	Social	−0.47	0.20	[−0.88, −0.08]
Indices of conditional moderated mediation
Medical information: Index = −0.37, SE = 0.47, 95% CI [−1.34, 0.51]
Social information: Index = −0.97, SE = 0.35, 95% CI [−1.68, −0.29]

RQ2 asked whether information type moderates the interaction between individuation and doctor identity on perceived relational closeness and thereby patient satisfaction. The index of the conditional moderated mediation was significant when the information type was social, Index = −0.97, SE = 0.46, 95% CI [−1.88, −0.04]. As shown in Table 4, when the AI-assisted human doctor forgot to recall the patient’s social information, it resulted in higher perceived relational closeness, which further increased patient satisfaction.

OPEN IN VIEWER

Table 4.

Conditional Indirect Effects on Patient Satisfaction through Perceived Relational Closeness by Information Type.

Doctor identity	Individuation	Information type	Effect (B)	SE	95% CI
Doctor identity → Perceived relational closeness → Patient satisfaction
AI-assisted human (vs. AI doctor)	No	Medical	0.04	0.33	[−0.62, 0.66]
AI-assisted human (vs. AI doctor)	No	Social	0.85	0.36	[0.14, 1.58]
AI-assisted human (vs. AI doctor)	Yes	Medical	−0.04	0.39	[−0.81, 0.74]
AI-assisted human (vs. AI doctor)	Yes	Social	−0.12	0.28	[−0.67, 0.46]
Indices of conditional moderated mediation
Medical information: Index = −0.08, SE = 0.51, 95% CI [−1.11, 0.94]
Social information: Index = −0.97, SE = 0.46, 95% CI [−1.88, −0.04]
Human (vs. AI doctor)	No	Medical	−0.09	0.37	[−0.80, 0.66]
Human (vs. AI doctor)	No	Social	1.11	0.42	[0.30, 1.92]
Human (vs. AI doctor)	Yes	Medical	0.13	0.55	[-1.00, 1.18]
Human (vs. AI doctor)	Yes	Social	0.69	0.29	[0.15, 1.31]
Indices of conditional moderated mediation
Medical information: Index = 0.23, SE = 0.66, 95% CI [−1.09, 1.49]
Social information: Index = −0.42, SE = 0.50, 95% CI [−1.38, 0.59]

H7 predicted that AI doctors that used social information for individuation accrue higher perceived privacy intrusiveness compared to human doctors who used social information for individuation. However, none of the conditional moderated mediation indices were significant, as shown in Table 5. Thus, H7 was not supported.

OPEN IN VIEWER

Table 5.

Conditional Indirect Effects on Patient Satisfaction through Perceived Privacy Intrusiveness by Information Type.

Doctor identity	Individuation	Information type	Effect (B)	SE	95% CI
Doctor identity → Perceived privacy intrusiveness → Patient satisfaction
AI-assisted human (vs. AI doctor)	No	Medical	−0.07	0.08	[−0.28, 0.05]
AI-assisted human (vs. AI doctor)	No	Social	−0.03	0.06	[−0.15, 0.11]
AI-assisted human (vs. AI doctor)	Yes	Medical	−0.03	0.07	[-.20, .10]
AI-assisted human (vs. AI doctor)	Yes	Social	0.02	0.06	[−0.08, 0.15]
Indices of conditional moderated mediation
Medical information: Index = 0.04, SE = 0.10, 95% CI [−0.14, 0.28]
Social information: Index = 0.05, SE = 0.08, 95% CI [−0.12, 0.23]
Human (vs. AI doctor)	No	Medical	−0.03	0.07	[-.20, .10]
Human (vs. AI doctor)	No	Social	−0.04	0.07	[−0.19, 0.09]
Human (vs. AI doctor)	Yes	Medical	−0.03	0.08	[−0.23, 0.09]
Human (vs. AI doctor)	Yes	Social	0.10	0.08	[−0.03, 0.28]
Indices of conditional moderated mediation
Medical information: Index = −0.00, SE = 0.10, 95% CI [−0.23, 0.19]
Social information: Index = 0.13, SE = 0.12, 95% CI [−0.05, 0.40]

Testing the Moderating Role of Privacy Control

RQ3 asked whether privacy control reduces the perceived relational closeness from an individuating AI doctor. Given that PROCESS Macro only allows two moderators, and we did not observe an interaction effect between individuation and information type, we created a four-level variable called individuation types, including social individuation, medical individuation, non-social individuation, and non-medical individuation, and we entered it as a moderator (W), along with privacy control (Z), in Model 11. Results revealed three significant conditional moderated mediation effects on patient satisfaction through perceived relational closeness (see Table 6). Compared to AI doctors, the presence of human doctors and AI-assisted human doctors alone led to higher perceived relational closeness when both individuation and privacy control were absent. Moreover, when privacy control was present, social individuation by a human doctor resulted in greater relational closeness compared to social individuation by an AI doctor.

OPEN IN VIEWER

Table 6.

Conditional Indirect Effects on Patient Satisfaction through Perceived Relational Closeness by Privacy Control.

Doctor identity	Individuation type	Privacy control	Effect (B)	SE	95% CI
Doctor identity → Perceived relational closeness → Patient satisfaction
AI-assisted human (vs. AI doctor)	Social individuation	No	−0.41	0.41	[−1.20, 0.39]
AI-assisted human (vs. AI doctor)	Social individuation	Yes	0.31	0.39	[−0.45, 1.11]
AI-assisted human (vs. AI doctor)	Medical individuation	No	0.17	0.45	[−0.70, 1.06]
AI-assisted human (vs. AI doctor)	Medical individuation	Yes	−0.26	0.65	[−1.54, 1.04]
AI-assisted human (vs. AI doctor)	Non-social individuation	No	1.24	0.46	[0.33, 2.13]
AI-assisted human (vs. AI doctor)	Non-social individuation	Yes	0.25	0.59	[−0.81, 1.51]
AI-assisted human (vs. AI doctor)	Non-medical individuation	No	−0.17	0.49	[−1.14, 0.75]
AI-assisted human (vs. AI doctor)	Non-medical individuation	Yes	0.22	0.45	[−0.71, 1.08]
Human (vs. AI doctor)	Social individuation	No	0.72	0.53	[−0.35, 1.79]
Human (vs. AI doctor)	Social individuation	Yes	0.67	0.30	[0.11, 1.27]
Human (vs. AI doctor)	Medical individuation	No	0.03	0.67	[−1.47, 1.31]
Human (vs. AI doctor)	Medical individuation	Yes	0.21	0.91	[−1.72, 1.94]
Human (vs. AI doctor)	Non-social individuation	No	1.13	0.48	[0.15, 2.07]
Human (vs. AI doctor)	Non-social individuation	Yes	0.95	1.31	[−1.04, 3.07]
Human (vs. AI doctor)	Non-medical individuation	No	−0.26	0.47	[−1.24, 0.63]
Human (vs. AI doctor)	Non-medical individuation	Yes	0.22	0.66	[−1.08, 1.60]

Regarding H8, we did not observe any significant conditional moderated mediation effects on perceived privacy intrusiveness and the sequential effect on patient satisfaction, as shown in Table 7. Thus, H8 was rejected.

OPEN IN VIEWER

Table 7.

Conditional Indirect Effects on Patient Satisfaction through Perceived Privacy Intrusiveness by Privacy Control.

Doctor identity	Individuation type	Privacy control	Effect (B)	SE	95% CI
Doctor identity → Perceived privacy intrusiveness → Patient satisfaction
AI-assisted human (vs. AI doctor)	Social individuation	No	0.01	0.08	[−0.15, 0.18]
AI-assisted human (vs. AI doctor)	Social individuation	Yes	0.05	0.08	[−0.09, 0.23]
AI-assisted human (vs. AI doctor)	Medical individuation	No	−0.14	0.12	[−0.42, 0.05]
AI-assisted human (vs. AI doctor)	Medical individuation	Yes	0.10	0.13	[−0.11, 0.39]
AI-assisted human (vs. AI doctor)	Non-social individuation	No	−0.03	0.10	[−0.24, 0.17]
AI-assisted human (vs. AI doctor)	Non-social individuation	Yes	−0.01	0.08.	[−0.17, 0.17]
AI-assisted human (vs. AI doctor)	Non-medical individuation	No	−0.03	0.11	[−0.31, 0.18]
AI-assisted human (vs. AI doctor)	Non-medical individuation	Yes	−0.09	0.11	[−0.36, 0.06]
Human (vs. AI doctor)	Social individuation	No	0.11	0.11	[−0.05, 0.37]
Human (vs. AI doctor)	Social individuation	Yes	0.09	0.09	[−0.04, 0.28]
Human (vs. AI doctor)	Medical individuation	No	−0.08	−0.08	[−0.38, 0.11]
Human (vs. AI doctor)	Medical individuation	Yes	0.02	0.02	[−0.22, 0.24]
Human (vs. AI doctor)	Non-social individuation	No	−0.02	−0.02	[−0.19, 0.13]
Human (vs. AI doctor)	Non-social individuation	Yes	−0.10	−0.10	[−0.61, 0.30]
Human (vs. AI doctor)	Non-medical individuation	No	−0.00	−0.00	[−0.25, 0.24]
Human (vs. AI doctor)	Non-medical individuation	Yes	−0.06	−0.06	[−0.33, 0.14]

Exploratory Analysis

Given the absence of sufficient theoretical background and empirical evidence to hypothesize the moderating effect of privacy control on the interaction between individuation, doctor identity, information type on perceived effort, and recognizing the practical importance of understanding the effect of privacy control on perceived effort of individuation, we conducted an exploratory analysis to delve into this aspect. Results from Model 11 in PROCESS Macro showed a significant conditional moderated mediation effect when comparing AI-assisted human and AI doctors, Index = −0.60, SE = 0.24, 95% CI: [−1.09, −0.17] and human and AI doctor, Index = −0.67, SE = 0.27, 95% CI: [−1.23, −0.17]. As shown in Table 8, when privacy control was offered, social individuation by an AI doctor was perceived as expending more effort compared to social individuation by a human doctor or an AI-assisted human doctor. The perceived effort, in turn, increased patient satisfaction.

OPEN IN VIEWER

Table 8.

Conditional Indirect Effects on Patient Satisfaction through Perceived Effort by Privacy Control.

Doctor identity	Individuation type	Privacy control	Effect (B)	SE	95% CI
Doctor identity → Perceived effort → Patient satisfaction
AI-assisted human (vs. AI doctor)	Social individuation	No	−0.08	0.21	[−0.49, 0.33]
AI-assisted human (vs. AI doctor)	Social individuation	Yes	−0.60	0.24	[−1.09, −0.17]
AI-assisted human (vs. AI doctor)	Medical individuation	No	−0.04	0.28	[−0.58, 0.52]
AI-assisted human (vs. AI doctor)	Medical individuation	Yes	−0.27	0.29	[−0.83, 0.33]
AI-assisted human (vs. AI doctor)	Non-social individuation	No	0.41	0.36	[−0.28, 1.13]
AI-assisted human (vs. AI doctor)	Non-social individuation	Yes	0.01	0.42	[−0.66, 0.97]
AI-assisted human (vs. AI doctor)	Non-medical individuation	No	−0.12	0.37	[−0.91, 0.54]
AI-assisted human (vs. AI doctor)	Non-medical individuation	Yes	0.02	0.37	[−0.71, 0.73]
Human (vs. AI doctor)	Social individuation	No	−0.27	0.34	[−0.97, 0.39]
Human (vs. AI doctor)	Social individuation	Yes	−0.67	0.27	[−1.23, −0.17]
Human (vs. AI doctor)	Medical individuation	No	−0.82	0.55	[−1.98, 0.17]
Human (vs. AI doctor)	Medical individuation	Yes	−0.07	0.22	[−0.47, 0.39]
Human (vs. AI doctor)	Non-social individuation	No	0.41	0.33	[−0.24, 1.05]
Human (vs. AI doctor)	Non-social individuation	Yes	0.57	0.95	[−0.88, 2.10]
Human (vs. AI doctor)	Non-medical individuation	No	−0.38	0.37	[−1.17, 0.31]
Human (vs. AI doctor)	Non-medical individuation	Yes	0.48	0.47	[−0.38, 1.49]

In sum, we find different mechanisms by which a doctor’s communicative efforts to reflect knowing a patient distinctly, whether human, AI, or AI-assisted human, can influence patients’ satisfaction. Specifically, individuals thought AI doctors exerted greater effort compared to human doctors when differentiating the patients from others, especially when the doctor recalled social information about them. Moreover, this moderated mediation effect held true when privacy control was provided. By contrast, the presence of human doctors appeared to increase perceived relational closeness, which further enhanced patient satisfaction, with or without social individuation and privacy control. Similar to human doctors, the presence of AI-assisted human doctors tended to increase patient satisfaction through perceived relational closeness. However, this indirect effect was true only in the absence of social individuation and privacy control.

Discussion

This study began with two basic questions: What does it take to feel as if we are known by somebody, more pointedly, does individuating recall about someone promote this feeling of being known, if the recollections are alluded to in conversation? We complicated the question by asking, can an interactive computerized system appear to know someone as well as a human may appear to, given the relative advantages computers have in their capacity for memory and retrieval? We examined these questions in a particular context where individuation plausibly may or may not occur, and where more than one type of individuating information is potentially valuable: a doctor/patient interaction with both medical and social topics as potential sources of individuation.

One major finding of the study is that patients seem to perceive the social individuation by an AI doctor as exerting more effort, thus resulting in higher patient satisfaction. The perceived effort is heightened when privacy control is provided by the doctor. Just as individuals appreciate having their uniqueness recognized in favorable conditions by other humans (Maslach, 1974; Şimşek & Yalınçetin, 2010), they also appear to value an AI medical system that recalls and incorporates their unique social information and characteristics into their discussions. The importance of social individuation from an AI doctor echoes the study by Lew and Walther (2023), which found that chatbots and humans that are contingent in replying and are relatively fast in response were rated as more credible and attractive.

Why does social individuation by an AI doctor matter so much? Our findings revealed perceived effort as a significant psychological mechanism. Participants seem to perceive that it takes significant effort for an AI doctor to recall and discuss individuating information pertaining to one’s social life, thus contributing to patient satisfaction. It is interesting that participants made this attribution, given that AIs do not expend either greater or less effort. It is unclear at this point whether the attribution is directed to the AI itself, which is somewhat consistent with the Computers are Social Actors paradigm (CASA; Nass & Moon, 2000), which contends that individuals make interpersonal attributions about computers as they do about fellow humans. This attribution, furthermore, was one element that made interacting with an AI doctor experientially different than interacting with a human.

Although an AI doctor’s recollection of a participant’s social information led to a perception of greater expenditure of effort, this effect was significantly different compared to when a human doctor recalled a patient’s social information. Patients perceived that AI doctors exert more effort in recalling and mentioning social information compared to human doctors, and this effort results in greater patient satisfaction. This finding in general echoes the evidence in Bechwati and Xia (2003), who found that individuals enjoy seeing the effort put in by a machine. This finding is also consistent with a recent study which shows that attributes of an AI algorithm, such as its ability to learn about its users, trigger the “helper heuristic” and contribute to greater user trust in the system (Lee et al., 2023).

This interesting interaction effect among individuation, information type, and doctor identity on patient satisfaction through perceived effort becomes more nuanced when considering the influence of privacy control. The finding suggests that patients may have some cognizance that, when the privacy of their data is subject to their own control and potential deletion, an AI medical system possibly works even harder to protect a patient’s personal information. If a machine does a good job of recalling and referencing one’s social information, the provision of privacy control may add to the appreciation of the effort, thus contributing to patients’ satisfaction.

While individuation is beneficial from an AI doctor, is it not also from a human doctor? The answer once again surfaces a difference between thinking one is interacting with a human or with an AI medical system. An AI doctor needs social individuation and privacy control. A human doctor appears to accrue greater relational closeness with or without making these accommodations. Although the provision of social individuation and privacy control may not be necessary for a human doctor, they are very important for an AI doctor, as these factors can promote greater patient satisfaction through the perception of greater effort.

Resembling human doctors to some extent, AI-assisted human doctors seemingly do not require social individuation and privacy control to increase perceived relational closeness. Interestingly, an AI-assisted human doctor who forgot to recall the patient’s social information appears to foster higher perceived relational closeness, which further leads to higher patient satisfaction compared to the condition involving only an AI doctor. This suggests that the human element in AI-assisted medical consultation contributes to positive relational outcomes.

Conclusion

The feeling of being known by someone occurs by different means, and matters differently, when interacting with a human or an AI doctor. Patients can obtain relational closeness from a human doctor who provides neither individuation nor privacy control. AI doctors, however, are more effective when they recall and mention a patient’s social information and enable privacy control. This combination leads conversational participants to perceive the AI doctor as exerting more effort, increasing patients’ satisfaction.

The next decade will embrace more AI systems in healthcare and in numerous other contexts. Leveraging the unlimited memory capability of AI and the relational warmth normally attributed to a human doctor may be the key to providing better doctor-patient interactions. Understanding additional communicative mechanisms that make people feel distinctively known by their communication partners will enhance online communication more generally.

Some in the medical community discount the ability of AI to engender functional interpersonal communication with patients: According to Kulkarni and Singh (2023, E1–E2), “It often requires a compassionate clinician to make an interpersonal human bond with a patient. . .. AI, at least in its current form, is likely unable to do this, and many scientific advances would be required for it to gain this ability in the future.” This position may not only overestimate the notion that medical training imbues all humans with interpersonal communication strategies, it may underestimate what is possible for AI to learn to do. In fact, a recent study conducted a blind comparison by licensed health care professionals of responses to patients’ questions generated by AI chatbots versus those generated by human physicians in a social media forum; “The chatbot responses were preferred over physician responses and rated significantly higher for both quality and empathy” (Ayers et al., 2023). Aside from differences in the average number of words generated by humans or AI, the study did not reveal what message aspects led to these differences. Given that a recent survey found that individuals are very accepting of the idea “to use AI doctors. . .at the primary care level” (Uymaz et al., 2023, p. 13), more research on the generation and reflection of interpersonal knowledge and empathic responses by AI will be incredibly valuable, in medical conversations and elsewhere.

Supplemental Material

Supplemental Material