Unlike previous generations of artificial intelligence (AI), AI assistants today can autonomously perform actions without human input or intervention. Several studies have proposed but not tested the notion that increased levels of AI autonomy may ultimately conflict with consumers’ fundamental need for autonomy themselves. Across five experiments (N = 1981), including representative samples and pre-registered hypotheses, we investigate consumer responses to high (vs. low) AI autonomy in the context of online shopping. The results show a pronounced negative effect of high AI autonomy on consumers’ adoption intentions – an effect mediated by consumers’ relative state of powerlessness in the presence of high AI autonomy. However, when consumers face situations characterized by scarcity, such as when preferred options are being sold out rapidly (e.g. Black Friday), the aversive aspects of high (vs. low) AI autonomy are attenuated. Together, these findings offer novel insights regarding whether, when and why consumers are willing to adopt high (vs. low)-autonomy AI assistants in online shopping settings.

Less is more… how can that be? It's impossible… more is more.

Yngwie Malmsteen – multi-instrumentalist

Introduction

Artificial intelligence (AI) is emerging as a powerful assistive technology that challenges what managers and scholars know about consumer responses, considering the many novel ways AI assistants can facilitate consumer engagement and decision-making (Frank et al., 2021; Hajli et al., 2022; Nguyen and Malik, 2022; Söderlund and Oikarinen, 2021; Tiberius, Gojowy and Dabić, 2022). While still years away from superintelligence (Bostrom, 2014), modern AI assistants are moving from human-in-the-loop operations, in which the AI relies on user input to perform tasks, such as instructions or permissions (de Bellis and Johar, 2020; Evanschitzky et al., 2015; Sestino et al., 2022), to human-out-of-the-loop operations, characterized by actions performed by the AI autonomously, without the need of human intervention (Fink et al., 2023; Sobolev, 2022). For example, Amazon's Alexa, the voice-controlled smart home AI assistant, has received the skill to act on its own ‘hunches,’ performing actions autonomously, such as switching off lights and radiators at home in the absence of the owners, calling emergency services when it detects unusual activities that the owners may not be capable of tackling themselves and reordering utilities when they are estimated to be running low (Campbell, 2021). While these actions may seem comforting, how would consumers respond to an autonomous AI assistant proactively ordering wine, flowers and condoms when it estimates their new relationship to have approached a more intimate stage – would they be pleasantly surprised or shocked to receive the delivery?

To examine people's general perceptions of high (vs. low) AI autonomy, we asked 190 Americans (age: M = 37.5, SD = 12.2; 50.5% female) how likely they would be to use an AI assistant for purchasing a new TV (1 = extremely unlikely to 7 = extremely likely), given that the AI would either suggest they purchase the TV by notifying them when potential savings were highest or automatically purchase the TV when the potential savings were highest without notification. In both cases, participants were reminded that the purchase would be made with their own money. Despite representing a considerably less provocative scenario than that of the condoms, we found a remarkably strong and negative response towards AI autonomy, with a significantly lower preference for the AI high in autonomy (M_High = 2.56, SD = 1.60) than for the AI low in autonomy (M_Low = 4.75, SD = 1.58); t(189) = −15.71, p < 0.001, d = −1.14.1 This result sparks an intriguing question: Why are consumers less interested in high (vs. low)-autonomy AI, despite its potential to better assist them?

The AI adoption literature provides mixed answers to this question, with some studies mirroring the above results (Casidy et al., 2021; Huang and Qian, 2021) and others suggesting AI autonomy to drive consumer AI acceptance (Dietvorst, Simmons and Massey, 2018; Wen et al., 2022). This status quo leaves managers and institutional decision-makers at risk, as the uncertainty about the potential negative consequences of high-autonomy AI on consumer responses is yet to be systematically addressed (Botti, Iyengar and McGill, 2023; Frank et al., 2023).

In what follows, we investigate the influence of AI autonomy on consumers’ intentions to adopt AI assistant services in their daily lives. We propose that the shift in autonomy from consumers to AI should be closely linked to consumers’ personal sense of power. Across five experiments, with a total sample of almost 2000 consumers, we study consumer adoption of high (vs. low)-autonomy AI assistants across different online shopping settings. Our results document a strong negative effect of high (vs. low) AI autonomy on consumers’ AI adoption intentions. Consumers’ personal sense of power acts as a psychological mechanism driving this effect, such that consumers who encounter a shopping situation involving high (vs. low) AI autonomy experience a state of powerlessness, which ultimately lowers their AI adoption intentions. Importantly, the influence of AI autonomy on consumers’ personal sense of power and AI adoption intentions is mitigated in shopping situations where popular options are quickly sold out, such as during Black Friday, highlighting scarcity as an important boundary condition. These results are robust across shopping settings, experimental designs and study paradigms, and apply even after controlling for a series of potential confounds. Additionally, the findings generalize across countries, representative samples and pre-registered study designs.

Our research contributes to the literature in three crucial ways. First, we expand the existing body of AI adoption research (Leone et al., 2021; Sestino et al., 2022) by empirically validating consumers’ fundamental need for autonomy (Botti, Iyengar and McGill, 2023; de Bellis and Johar, 2020; Wertenbroch et al., 2020) in modern shopping settings facilitated by AI agents.

Second, we document a novel psychological mechanism that helps explain consumers’ reluctance towards AI high in autonomy, supporting the notion that a transfer of autonomy from consumers to AI poses a challenge (Frank et al., 2023; Leung, Paolacci and Puntoni, 2018), as this shift induces powerlessness in consumers, thereby lowering their AI adoption intentions.

Third, we demonstrate a meaningful boundary condition, with scarcity moderating our focal effect. More precisely, we find that consumers’ aversion towards high (vs. low) AI autonomy is reduced when the limited availability of products is made salient. In other words, scarcity reduces consumers’ reluctance towards AI assistants high in autonomy, irrespective of whether behavioural intentions or consumer choice constitute the primary outcome.

From a practical perspective, our findings are relevant to managers and AI decision-makers in the service and retail sectors (Evanschitzky and Goergen, 2018; Ko et al., 2022), who are challenged to deliver optimal consumption experiences and utilize increasingly autonomous AI. In particular, the results offer valuable insights regarding strategies that may be successful in the future development, implementation and utilization of high-autonomy AI technologies. Proposed initiatives include establishing a common labelling for the autonomous capabilities of AI, as is done for vehicle driving automation systems (SAE International, 2021), to actively address the potential state of powerlessness consumers may face when transferring agency to highly autonomous AI assistants.

Theoretical background

Even before the advent of AI, researchers were interested in understanding the factors that influence consumer adoption and use of intelligent and even superintelligent systems (Bostrom, 2014). Since the dawn of AI-related research, scholars have studied a wide range of aspects related to consumers’ AI adoption (e.g. Dietvorst, Simmons and Massey, 2015; Longoni and Cian, 2020; Schmitt, 2020). Although such scholarly work has yielded profound insights, the rise of AI autonomy and the lack of research that has manipulated and quantified the consequences of said autonomy on consumers’ AI adoption have prompted scholars to call for more research on this topic (Frank et al., 2023). In the following sections, we integrate several streams of literature to provide plausible predictions of whether, why and under what circumstances AI autonomy should influence consumers’ AI adoption decisions. Rather than relying on a single theory, we combine frameworks centred on autonomy and consumer choice (Wertenbroch et al., 2020), with conceptualizations of personal sense of power (Magee and Galinsky, 2008) and scarcity (Hamilton et al., 2019), given the value of using such an integrative approach (Graham, 2013; Takeuchi et al., 2005). From a meta-theoretic perspective, we use the ‘explaining theory’ type (Sandberg and Alvesson, 2021), in which we explain a phenomenon through a set of logically related variables using relevance criteria focused on accuracy and testability, while specifying the causal impact, generality and boundary conditions of our proposed chain of events (Bacharach, 1989; Hambrick, 2007).

AI, autonomy and consumer choice

Scholars in the field of AI research differentiate between two primary directions: narrow AI, also known as weak AI, and strong AI, commonly referred to as general AI (Linde and Schweizer, 2019). Some also propose the existence of a third direction, known as superintelligent AI (Bostrom, 2014; Scott et al., 2022). Narrow AI, as defined by Linde and Schweizer (2019), refers to the ability of an AI agent to achieve specific goals within a limited set of environments. It is expected that narrow AI will outperform humans in these specific tasks. Examples of narrow AI include self-driving cars and popular virtual assistants like Siri, Alexa and Google's automated translation feature (Scott et al., 2022). Beyond this, AGI, or strong (general) AI, aims to encompass a unified system capable of comprehensive human-like intelligence across various cognitive domains such as language, perception, reasoning, creativity and planning (Scott et al., 2022).

AI systems also differ in their level of autonomy. In this research, AI autonomy is defined as ‘the ability of the AI technology to perform tasks derived from humans without specific human interventions’ (Hu et al., 2021, p. 2). As such, AI autonomy refers to the extent to which AI can perform actions and execute tasks without requiring human intervention, feedback or approval (Frank et al., 2023). A systematic review of the AI adoption literature, with a focus on autonomy aspects (Table 1), shows that prior studies fall short in capturing the unique shift in power dynamics that results from modern, highly autonomous AI systems. In fact, only a few studies have examined AI autonomy as an independent variable for explaining consumer choices to adopt (vs. not adopt) AI, including its advice and dependencies. Even fewer studies have experimentally manipulated AI autonomy. This makes managerial decision-making regarding AI autonomy a risky business, considering the lack of cause–effect relationships established.

Table 1. Review of the AI literature

Study	Conceptualization of AI autonomy and related constructs	Sample*	Experimental†	AI autonomy as independent variable/manipulated	Representative‡	Preregistration/open data	Moderation	Mediation	Main findings
Tavitiyaman, Zhang and Tsang (2022)	Service automation: The delivery of predetermined tasks by automated machinery	N = 301					✓		- The effect of the perceived usefulness of hotel technologies on the behavioural intention of tourists is stronger for experienced (vs. inexperienced) tourists.
Voorveld and Araujo (2020)	Virtual assistants: Voice- or text-activated software agents that interpret requests by users and execute commands	N = 180	✓				✓	✓	- Participants who are exposed to a virtual assistant with (vs. without) a human name are less concerned about their autonomy and more strongly persuaded.
Balakrishnan and Dwivedi (2021)	Automated chatbot: A computer program that conducts a conversation in natural language and sends a response based on business rules and data tuned by the organization	N = 410	✓					✓	- Human-to-machine interaction influences cognitive absorption more positively than human-to-human interactions. - User trust and experience mediate the effect of cognitive absorption on consumers’ technology continuation intention.
Casidy et al. (2021)	Autonomous vehicles: Higher-level automated driving systems that control all dynamic aspects of the driving task under all roadway and environmental conditions	N = 294 N = 288 N = 575	✓				✓	✓	- Strong (vs. weak) self-brand connection reduces risk barriers, resulting in higher radical innovation adoption. - Self-brand connection is less (vs. more) effective in reducing risk and usage barriers for brands with conservation (vs. openness) concepts. - Conservation brands can communicate their innovation capability to optimize and strengthen the mediated relationship between self-brand connections and adoption intentions.
Fernandes and Oliveira (2021)	AI-based digital voice assistants: Virtual conversational agents that recognize and understand voice-based user requests and communicate using natural language to accomplish a wide variety of tasks	N = 238					✓	✓	- Functional elements (ease of use, usefulness, subjective norms), social elements (humanness, social interactivity, social presence) and relational elements (trust, rapport) explain most (88%) of user acceptance variance. - Anthropomorphism is not universally positive and high-tech customers who prefer technology-infused frontlines may be the ones more sensitive to the high-touch, relational features of digital voice assistants.
Flavián et al. (2022)	Robo-advisor agents: Agents that automate or assist in managing investments by replacing human advisory services and/or the customer's own management	N = 404			(✓)		✓		- Technological discomfort experienced by customers (e.g. the lack of control and overwhelming technological complexity) has a positive and significant effect on intention to use robo-advisors.
Frank et al. (2021)	AI product autonomy: Products that autonomously perceive their environment and take autonomous actions to achieve their objectives without requiring consumer intervention or feedback	N = 1388	(✓) Within-subjects design	✓	(✓) Urban sample		✓		- Utilitarian, hedonic and symbolic value contribute equally to AI product demand. - AI product autonomy moderates the effects of value on AI product demand: controlling for AI product value, behavioural intent is higher for high (vs. low) autonomy of the AI product. - AI product autonomy weakens the effect of utilitarian value and strengthens the effect of hedonic value slightly, while also strengthening the effect of symbolic value on behavioural intent.
Huang and Qian (2021)	Autonomous vehicles: Vehicles powered by AI technology that are highly or fully automated and require limited or no human intervention	N = 849					✓	✓	- Face consciousness acts as an antecedent to the reasoning process and has positive effects on both reasons for and reasons against adopting autonomous vehicles, and also exerts an indirect influence on individuals’ attitude towards autonomous vehicles via the mediators of the reasons for and against adopting autonomous vehicles. - Consumers’ need for uniqueness strengthens the association between reasons for autonomous vehicles and autonomous vehicle adoption intention, while risk aversion reinforces the negative effect of reasons against autonomous vehicles on both intention to adopt autonomous vehicles and attitude towards autonomous vehicles.
Ribeiro, Gursoy and Chi (2022)	Autonomous vehicles: Vehicles that can drive without human intervention	N = 362						✓	- Trust is the most powerful determinant of performance expectancy and crucial to attenuate the risk of using autonomous vehicles during a tourism experience. - Performance expectancy and hedonic motivation are key determinants of travellers’ positive emotions, which in turn determines the acceptance of autonomous vehicles.
Shareef et al. (2021)	Autonomous system: Any integrated and interconnected process or product that can act independently, making appropriate contextual decisions within a changing environment and without any human intervention	N = 112						✓	- Personal ability and control and empathetic cooperation and social interaction explain most (55%) of the variance in trust in the autonomous system, which in turn predicts adoption intention. - For elderly people, the main issues for developing trust in an autonomous homecare system as an alternative to human support are the ability to use the system and feelings of empathetic interaction with the system.
Alfalhi (2022)	Intelligent automation: Bringing together automation and AI to facilitate value-adding change through new technology, which permits autonomous interaction with people and tasks	N = 330					✓	✓	- Customer engagement with AI leads to mutual trust in an indirect relationship of customer engagement and mutual trust through corporate communication
Ali, Swiety and Mansour (2022)	Robotic process automation: Software that imitates rules based on digital tasks performed by humans	N = 179							- The use of AI-based banking services creates convenience and enhances trust.
Meyer-Waarden and Cloarec (2022)	AI-powered autonomous vehicles: Vehicles in which human drivers are never required to take control to safely operate the vehicle, and where the execution is performed by an AI-based machine agent of a function that was previously carried out by humans	N = 207					✓	✓	- Positive relationships exist between the behavioural intention to use AI-powered autonomous vehicles and people's performance/effort expectancy, social recognition, well-being, hedonism and technology trust, as well as security related to the product.
Osburg et al. (2022)	Autonomous vehicle service: Modes of autonomous transport, which make substantial use of artificial intelligence, operating services for public, private or goods transportation, but excluding privately owned AVs for personal use	N = 27,565 N = 300		✓	✓				- There are four segments of consumers that can be classified across two dimensions: consumers who are open (vs. reserved) towards fully autonomous vehicles and consumers who are strongly averse to risk (vs. comfortable with risk) in their emotional responses. - As three of the four segments identified require some form of human supervision, findings support the service robot deployment model of Wirtz et al. (2018), which calls for human–robot joint service configurations in complex services. - Exposure to positive information about autonomous vehicle services negatively correlates with the likelihood of membership in the two most resistant segments.
Piehlmaier (2022)	Robo-advisors: Algorithmic wealth management tools that rely on AI to efficiently allocate investments using exchange-traded funds and index funds due to their simple cost structure and passive approach to portfolio management	N = 2000			(✓) Investor sample	✓			- Overconfident investors have a significantly higher propensity to adopt robo-advice. - While higher financial literacy seems to decrease robo-advice uptake, unjustified confidence in one's knowledge causally increases it. - Willingness to take financial risk cannot account for the significantly increased adoption of robo-advice among overconfident investors.
Sestino et al. (2022)	Product intelligence: The extent to which consumers perceive a product as an intelligent entity that is able to behave autonomously, learn and interact as a human being	N = 343		✓			✓	✓	- The perceived intelligence of a self-driving car increases consumers’ behavioural control (the tendency to feel capable of controlling an object), which in turn positively influences their intention to use self-driving cars. - The effect of perceived intelligence of a self-driving car on behavioural control is significant only among change seekers (the extent to which individuals stably look for new and stimulating experiences).
Sharma et al. (2022)	Autonomous decision-making process (in the context of retailing): A system that can perform shopping processes delegated by customers, such as decisions and purchases	N = 454					✓	✓	- All four Unified Theory of Acceptance and Use of Technology (UTAUT) variables (e.g., effort expectancy, performance expectancy, facilitating conditions and social influence) are positively associated with customers’ autonomous decision-making process adoption intention. - The positive association between facilitating conditions and attitudes towards the autonomous decision-making process is not stronger for customers who are high (vs. low) in uncertainty avoidance.
Song and Kim (2022)	Retail service robot: Emerging robotic technology that employs AI to provide automated in-store customer service	N = 1424					✓	✓	- Humanoid retail service robots’ usefulness, social capability and appearance positively influence consumers’ attitudes towards human–robot interaction. - Consumers’ favourable attitudes towards human–robot interaction predict anticipation of greater service quality, which positively influences their acceptance of humanoid retail service robots. - Consumers’ pre-existing anxiety towards robots moderates the relationship between the interaction facilitators of the humanoid robots and customers’ attitudes towards human–robot interaction in that it weakens the effect of social capability and strengthens that of appearance.
von Walter, Kremmel and Jäger (2022)	Robo-advisors: Investment strategies without human intervention	N = 454 N = 73 N = 227	✓				✓	✓	- Individuals who believe that AI is intellectually superior (vs. inferior) to humans are more likely to adopt algorithmic advice in a real market setting. - Task complexity acts as an important boundary condition and shows that lay beliefs about AI only have a significant impact on adoption of algorithmic advice when the decision task is perceived to be high (vs. low) in complexity.
Wen et al. (2022)	AI autonomy: The extent to which AI can make autonomous decisions and execute tasks correctly without requiring human feedback	N = 528						✓	- Perceived AI autonomy is the strongest (positive) predictor of customer participation behaviour in an equation that also includes perceived personalization, community identity, trust in AI and self-efficacy.
Aw et al. (2023)	Perceived autonomy: The extent to which a robot requires human intervention and support to function	N = 375						✓	- Perceived anthropomorphism (human-like attributes) and perceived autonomy positively influence consumers’ perception of justice regarding robo-advisory services. - Perceived justice has a negative impact on consumers’ privacy concerns and perceived intrusiveness associated with robo-advisors. - Privacy concerns and perceived intrusiveness positively influence consumers’ resistance to adopting robo-advisory services
Chua, Pal and Banerjee (2023)	Automation bias: When users readily buy into computer recommendations instead of relying on their own judgement	N = 368	✓				✓	✓	- Attitude towards AI is positively associated with trust in AI and perceived accuracy of AI, and behavioural intention to accept AI-based recommendations. - Trust in AI and perceived AI accuracy, in turn, are not significantly associated with behavioural intention to accept AI-based recommendations. - Risk level moderates how attitude, trust and perceived accuracy vary with behavioural intention to accept AI-based recommendations, such that trust and perceived accuracy are better predictors of AI acceptance intention in high (vs. low)-risk situations.
Dorotic, Stagno and Warlop (2023)	AI autonomy: The extent to which technology operates without human control or intervention	N = 4523 N = 790 N = 367	✓	✓		✓		✓	- The factors influencing AI adoption tensions are tied to people's perceptions of the individual costs and the degree of freedom in decision-making in relation to the perceived benefits, both on a personal and societal scale. - Perceptions of costs associated with AI are influenced by a reduced sense of control and autonomy
Fink et al. (2023)	Fully autonomous vehicles (FAVs): Vehicles where humans are riding with an autonomous AI driver	N = 187							- Fully autonomous vehicle rideshares are less inclined to prioritize emotional support, casual conversation and trust establishment compared to rideshares operated by humans. - Participants with visual impairments express a desire to interact with fully autonomous vehicles in a collaborative manner just as they do with human drivers, highlighting the importance of human-in-the-loop operations.
Qian et al. (2023)	Autonomous vehicles (AVs): Equipped with sensors and connecting devices, AVs can automatically detect road infrastructure, pedestrians and nearby vehicles	N = 880			✓			✓	- Consumers’ intention to use AV hailing services is influenced by their ethical attitudes and the reasons they have for or against AVs. - Consumers’ ethical attitudes are influenced by personal values and their reasons for and against AVs.
The present research	AI autonomy: The extent to which AI can make decisions and execute tasks without requiring human intervention, feedback or approval	N = 710 N = 392 N = 241 N = 417 N = 221	✓	✓ (High vs. Low)	✓ Consumer sample	✓	✓	✓	- There is a strong negative effect of high (vs. low) AI autonomy on consumers’ AI adoption intentions in online shopping settings. - Consumers’ perception of powerlessness (vs. powerfulness) under conditions of high (vs. low) AI autonomy mediates this effect. - Scarcity moderates this interplay, with the negative effect of high (vs. low) AI autonomy being attenuated in situations characterized by good options being sold out rapidly.

Note: *Denotes sample sizes in main studies, excluding pretests and pilot studies. †Parentheses denote studies that were not experimental between-subjects. ‡Parentheses denote studies that were close to but not fully representative of the target population. The literature presented originates from a Scopus title/abstract/keyword search on peer-reviewed articles published in the domains of ‘Psychology’ and ‘Business, Management and Accounting’ in English (as of 23 October 2023). We restricted the articles to those that matched the following text-search criteria: (‘artificial intelligence’ OR ‘AI’) AND (service OR retail OR consumer OR customer OR management OR market* OR shop*) AND (adopt*) AND (auto*). This search resulted in a total of 178 articles that were screened for suitability. Articles based on quantitative empirical evidence with a focus on AI autonomy in consumer choice and shopping settings were selected for inclusion, resulting in a final sample of 25 empirical articles (plus the current article) as summarized above.

Narrow AI systems, such as most existing AI assistants, can attain unsettling levels of autonomy, despite the fact that they do not reach the same level of capability as strong AI and superintelligent AI. It is essential to recognize that consumers’ decision to adopt highly autonomous AI constitutes a fundamental contrast to the adoption of low-autonomy AI (de Bellies and Johar, 2020; Sestino et al., 2022). That is, by adopting high-autonomy AI, which performs tasks autonomously (with humans out of the loop), consumers are sacrificing their autonomy of choice, defined as ‘consumers’ ability to make and enact decisions on their own, free from external influences imposed by other agents’ (Wertenbroch et al., 2020, p. 430). Similar to how AI assistants undermine consumers’ autonomy and subsequent enjoyment in video games (Kim et al., 2019), we posit that consumers’ AI adoption intentions by means of their willingness to use, try and subscribe to AI are lower for AI high (vs. low) in autonomy when it comes to online retail shopping, which is the focus of the current research. Accordingly, we hypothesize:

H1: Online retail shopping involving high (vs. low) AI autonomy will have a negative impact on consumers’ AI adoption intentions.

Personal sense of power

The act of adopting high-autonomy AI entails a shift in decision power from consumers to AI agents. This transfer of decision power to AI systems represents a fundamental alteration in the power dynamics between consumers and technology (Faraji-Rad, Melumad and Johar, 2017). Hence, we argue that understanding the ramifications of this shift in power is key for grasping the negative impact of high levels of AI autonomy on consumers’ AI adoption. In this context, the aspect of empowerment emerges as a critical component within consumer agency, encompassing the capacity to make autonomous choices and exert influence over one's own decision-making processes (Denegri-Knott, Zwick and Schroeder, 2006). By acknowledging the impact of AI autonomy on consumers’ sense of agency, researchers can gain deeper insights into the factors influencing consumer adoption and the potential barriers arising from the perceived loss of power.

In the AI adoption process, consumers transfer agency to the AI agent at the expense of their own power, with power broadly being defined as the asymmetric control (of an agent) over valued resources (Anderson, Hildreth and Howland, 2015; Keltner, Gruenfeld and Anderson, 2003; Magee and Frasier, 2014). While consumers’ personal sense of power has been widely discussed in consumer research (Choi and Mattila, 2016; Fuchs, Prandelli and Schreier, 2010; Mourali and Nagpal, 2013; Rucker and Galinsky, 2008; Wang, Raghunathan and Gauri, 2022), its role in shaping AI responses is yet to be examined. Personal sense of power has been discussed both as a stable trait and a situation-specific state (Anderson and Galinsky, 2006; Anderson, John and Keltner, 2012; Rucker and Galinsky, 2008). We argue that consumption contexts wherein consumers face situations of high (vs. low) AI autonomy will momentarily decrease their personal sense of power, with this decrease in power leading to lowered intentions to use and adopt AI. Therefore, we hypothesize:

H2: The negative impact of high (vs. low) AI autonomy on consumers’ AI adoption intentions will be mediated by consumers’ personal sense of power, with consumers who encounter online retail shopping involving high (vs. low)-autonomy AI being more (vs. less) prone to experience a state of powerlessness, thereby decreasing their AI adoption intentions.

Scarcity

Consumers’ sense of power interacts with various situational factors (Fan, Rucker and Jiang, 2022; Magee and Galinsky, 2008). This implies that certain contexts, such as a consumer recalling an instance when someone else had power over him- or herself, can induce feelings of powerlessness. Conversely, reflecting upon a situation where the consumer exerted power over another individual can trigger the reverse (Galinsky, Gruenfeld and Magee, 2003). One situational factor that can influence this dynamic is scarcity.

Scarcity is broadly conceptualized as a state of shortage compelling individuals to engage in specific behaviours (Elbæk et al., 2023; Hamilton et al., 2019; Suri et al., 2007). In consumers’ everyday lives, such a shortage is often related to time and product scarcity. Consumers also frequently face scarcity-based messages and events (limited edition, last chance to buy, Black Friday; Abendroth and Diehl, 2006; Folwarczny et al., 2022; Tang, Li and Su, 2022).

Transferring autonomy to an AI under conditions of scarcity may not necessarily be perceived as sacrificing one's own freedom of choice to the same extent as when the autonomy transfer is done under more ordinary circumstances. Rather, in situations characterized by scarcity, consumers may perceive that high-autonomy AI grants them a competitive edge in making certain shopping decisions that they would otherwise have to perform themselves (on dimensions such as being responsible for performing all actions to acquire a desired product before it is sold out). In other words, when shopping under conditions of scarcity, the general aversion towards high (vs. low)-autonomy AI, as applicable under regular circumstances, should no longer emerge. Consumers should realize the added value of high-autonomy AI in such settings when it comes, for example, to predicting the best possible time and price for a product purchase based on a powerful algorithm that excels at minimizing costs and maximizing savings, in conjunction with the ability to make the purchase faster than any human could possibly do. Accordingly, we anticipate that the negative impact of high (vs. low) AI autonomy on consumers’ intentions to adopt AI will be attenuated when scarcity is emphasized as part of the shopping situation. Formally, we hypothesize:

H3: Consumers’ intentions to adopt high (vs. low)-autonomy AI will be moderated by scarcity, such that the negative effect of high (vs. low)-autonomy AI on AI adoption is attenuated when the limited availability of products is made salient.

Combining the psychological mechanism assumed to drive our focal effect (H2) with the boundary condition outlined above (H3), and synthesizing both these elements into one testable model, we propose a fourth and final hypothesis. Specifically, we suggest that the adverse consequences associated with high (vs. low) AI autonomy on consumers’ intentions to adopt AI will be indirect through their heightened feelings of powerlessness in regular online retail shopping that lacks explicit cues to scarcity (Keltner, Gruenfeld and Anderson, 2003; Wang, Raghunathan and Gauri, 2022). However, this chain of events will be ‘turned off’ in the presence of scarcity. Stated differently, relative to regular online retail shopping without cues to scarcity, consumers should exhibit a greater relative willingness to embrace highly autonomous AI in shopping settings that are characterized by scarcity (Kurtz, 2008; Lee, Ryu and Chun, 2018; Liang, Ye and Liu, 2021). In such situations, the shift in autonomy from themselves to the AI system should no longer be perceived as a loss of power. Consequently, we hypothesize:

H4: The impact of high (vs. low) AI autonomy on consumers’ adoption intentions will be mediated by their personal sense of power and moderated by scarcity. In regular shopping settings, consumers who encounter a high (vs. low)-autonomy AI will be more (vs. less) prone to experience a state of powerlessness, thereby decreasing their AI adoption intentions. In shopping settings where the limited availability of products is made salient, by contrast, consumers’ personal sense of power will no longer mediate the link between AI autonomy and adoption intentions.

We test these hypotheses in a series of experiments across diverse samples, shopping settings and situational circumstances. Figure 1 depicts our conceptual model.

Details are in the caption following the image — **Figure 1**
Open in figure viewer PowerPoint

Conceptual model.

Overview of studies

Table 2 provides an overview of the empirical work of the current research. Studies 1A and B serve to establish the main effect of AI autonomy on consumers’ intentions to adopt AI. Study 2 provides evidence for the psychological mechanism driving this effect through consumers’ personal sense of power. Studies 3 and 4 document scarcity as an important boundary condition, with consumers being more inclined to choose (Study 3) and adopt (Study 4) high (vs. low)-autonomy AI under conditions of scarcity, with the negative effect of AI autonomy on consumers’ personal sense of power attenuated in such settings (Study 4). All reported studies were experiments and involved real consumers from the United States and the United Kingdom. The manipulations were effective and deemed realistic, as confirmed by our manipulation and realism checks available in the Online Appendix.

Table 2. Empirical overview

Study	Method	Sample	Design	Focus	Findings
1A	Pre-registered experiment	N = 710; representative	2 AI autonomy; between-subjects	Main effect	AI autonomy exerts a robust and negative effect on consumers’ AI adoption intentions.
1B	Experiment	N = 392	2 AI autonomy; between-subjects	Main effect	The negative effect of AI autonomy on consumers’ AI adoption intentions occurs even after controlling for consumers’ personality traits, technology readiness, purchase decision involvement, AI and product category knowledge, and AI experience.
2	Experiment	N = 241	2 AI autonomy; between-subjects	Mediation	Consumers’ personal sense of power mediates the negative effect of AI autonomy on their AI adoption intentions.
3	Choice experiment	N = 417	2 AI autonomy × Scarcity; within-subjects	Moderation	Scarcity attenuates the negative influence of high AI autonomy in a consumer choice context.
4	Experiment	N = 221	2 AI autonomy × 2 scarcity; between-subjects	Moderated mediation	Scarcity attenuates the negative effect of high (vs. low) AI autonomy on consumers’ AI adoption intentions, with personal sense of power no longer mediating the autonomy–adoption intentions link under such circumstances.

Study 1A: Pre-registered test of the main effect

Study 1A aims to establish evidence of the hypothesized negative effect of high (vs. low) AI autonomy on consumers’ AI adoption intentions using a pre-registered study based on representative samples of US and UK consumers (for details, see https://osf.io/54hkj/).

Participants, design and procedure

Responses from representative samples (N = 710) in terms of age, gender and ethnicity were collected from the online panel Prolific. Participants were from the United States (n = 354; age: M = 44.4, SD = 15.9; 50.8% female) and the United Kingdom (n = 356; age: M = 44.8, SD = 15.0; 52.5% female) and took part in the study voluntarily after providing written consent in exchange for monetary compensation.

Participants were asked to consider allowing an AI shopping assistant to help them purchase a new digital camera. The AI high in autonomy would automatically purchase the camera when it determined that the potential savings were the highest, whereas the AI low in autonomy would merely notify them to make the purchase (between-subjects).

In response to the respective shopping scenario, participants rated their intention to use the AI shopping assistant (‘How likely are you to use the artificial intelligence assistant for purchasing the digital camera?’; 1 = extremely unlikely to 7 = extremely likely), followed by a manipulation check, as described in our pre-registration.

Results

As shown in Figure 2, the results provide support for H1, irrespective of whether the US and UK samples were analysed separately or together. Three Bonferroni-corrected independent sample t-tests revealed that participants were significantly less inclined to use the high (vs. low)-autonomy AI in each of these cases: (a) US sample (M_High = 3.94, SD = 1.93 vs. M_Low = 5.04, SD = 1.72; t(352) = −5.65, p < 0.001; d = −0.60); (b) UK sample (M_High = 3.48, SD = 2.04 vs. M_Low = 4.65, SD = 1.78; t(354) = −5.74, p < 0.001; d = −0.61); and (c) total sample (M_High = 3.70, SD = 2.00 vs. M_Low = 4.85, SD = 1.76; t(708) = −8.14, p < 0.001; d = −0.61).2

Discussion

Study 1A provides pre-registered evidence for the notion that an AI assistant with high autonomy leads to significantly lower usage intentions than an AI assistant with low autonomy in the context of online shopping. This effect holds in a high-powered study with representative samples from the United States and the United Kingdom, and represents a large effect size by conventional standards (Funder and Ozer, 2019).

Study 1B: Replication of main effect

To further test the generalizability of the negative main effect of AI autonomy on consumers’ AI adoption intentions (H1), we conducted a follow-up experiment based on the same AI assistant shopping scenario in Study 1A. This time, we controlled for a number of factors that could potentially have influenced our initial results, including consumers’ personality traits, technology readiness, purchase-decision involvement, subjective and objective knowledge about AI and self-reported experience with AI. We included these additional factors to rule out the possibility that our results could have emerged due to any of these confounds rather than, as we expected, AI autonomy.

Participants, design and procedure

Responses from a US consumer sample (N = 392; age: M = 46.6, SD = 14.4; 49.5% female) were collected from the online panel Prolific. Participants were exposed to the same shopping scenario introduced in Study 1A, and responded to the same intention-to-use item. Additionally, this study included a series of validated scales on consumers’ purchase-decision involvement (e.g. ‘How important would it be to you to make a right choice of an artificial intelligence shopping assistant?’; 1 = not at all important to 7 = extremely important; Cronbach's α = 0.78; Mittal, 1995), technology readiness (e.g. ‘I keep up with the latest technological developments in my areas of interest’; 1 = strongly disagree to 7 = strongly agree; α = 0.83; Parasuraman, 2000) and Big Five personality dimensions (extraversion, agreeableness, conscientiousness, negative emotionality and open-mindedness) using the 15-item BFI-2-XS inventory (α_E = 0.60, α_A = 0.62, α_C = 0.76, α_N = 0.78, α_O-M = 0.65; Soto and John, 2017). Moreover, participants answered questions about their subjective knowledge about the product category (‘How knowledgeable are you about digital cameras?’; 1 = not at all to 7 = a great deal), subjective knowledge about AI (‘How knowledgeable are you about artificial intelligence in general?’; 1 = not at all to 7 = a great deal), personal experience with AI assistants (‘How often do you use artificial intelligence assistants (e.g. Siri, Alexa, Google Assistant, ChatGPT, etc.)?’; 1 = never to 7 = all the time) and responded to six true–false statements as a measure of their objective knowledge about AI (e.g. ‘Artificial intelligence (AI) refers to the simulation of human intelligence in machines’, ‘AI systems always improve their performance over time without human intervention’). The study ended with a manipulation check (‘Who made the final decision on purchasing the digital camera?’; 1 = myself to 7 = the artificial intelligence assistant) and a realism check (‘To me, the described artificial intelligence assistant seems…’; 1 = unrealistic to 7 = realistic).

Results

To examine potential confounding effects related to the impact of AI autonomy on consumers' AI adoption intentions, we performed a linear regression analysis (PROCESS Model 0; Hayes, 2018). AI autonomy (high = 1, low = 0) was the predictor, AI adoption intention was the outcome variable and all control variables (purchase-decision involvement, technology readiness, Big Five personality traits, subjective and objective knowledge about AI, subjective knowledge about the product category and personal experience with AI assistants) acted as covariates. The continuous variables were z-scored before analysis to allow for standardized coefficients (Gasiorowska, Folwarczny and Otterbring, 2022).

The fitted model successfully accounted for 27.4% of the variation in participants’ AI adoption intentions, as indicated by the R² value. Supporting H1, this model showed a significant and strong negative effect of high (vs. low) AI autonomy on participants’ AI adoption intentions (b = −1.32, SE = 0.17, p < 0.001). Besides AI autonomy, only two of the other variables yielded significant effects: participants’ personal experience with AI assistants (b = 0.62, SE = 0.11, p < 0.001) and their agreeableness levels (b = 0.24, SE = 0.10, p = 0.016), implying that higher values on these variables were linked to greater AI adoption intentions. Follow-up moderation analyses revealed that neither personal experience with AI assistants nor agreeableness moderated the main effect of high (vs. low) AI autonomy on participants’ AI adoption intentions (ps > 0.40 in both cases), underscoring the robustness and replicability of our main finding.

Discussion

Study 1B confirms a strong and negative main effect of high (vs. low) AI autonomy on consumers’ AI adoption intentions even after accounting for several theoretically relevant control variables, thus attesting to the generality of our focal effect and ruling out several alternative explanations for our results.3

Study 2: Mediation through personal sense of power

Study 2 investigates the extent to which consumers’ personal sense of power explains the negative relationship between AI autonomy and consumers’ AI adoption intentions (H2). We predict that consumers are less inclined to adopt AI high (vs. low) in autonomy to do their shopping because the shift in autonomy away from consumers and towards the AI induces feelings of powerlessness. Personal sense of power has been linked to self-esteem and negative affect (Talaifar et al., 2021; Wang, 2020; Wojciszke and Struzynska-Kujalowicz, 2007), with low self-esteem and negative affect also yielding effects on consumer responses similar to those occurring for the powerless (Rucker and Galinsky, 2008; Sivanathan and Pettit, 2010). Therefore, to provide more substantive evidence for the uniqueness of our proposed mechanism and rule out these two alternative accounts, we also collected data on participants’ self-esteem and affect.

Participants, design and procedure

Respondents (N = 241; age: M = 37.6, SD = 11.8; 56.0% female), recruited from Amazon's MTurk panel, were randomly assigned to an AI autonomy shopping scenario using the same single-factor between-subjects design and shopping scenarios introduced in Study 1A. In response to their assigned scenario, participants indicated their usage intention (‘How likely are you to use the artificial intelligence assistant for purchasing…?’), trial intention (‘How likely are you to try the artificial intelligence assistant…’) and subscription intention (‘How likely are you to subscribe to the artificial intelligence assistant…’) on seven-point scales (1 = extremely unlikely to 7 = extremely likely) regarding the respective AI shopping assistant that was framed as either being high or low in autonomy. As these measures of intention were highly correlated (α = 0.87), they were subsequently averaged into a composite index of AI adoption intentions.

Next, participants completed an adapted version of the eight-item Personal Sense of Power Scale (Anderson, John and Keltner, 2012) by indicating their agreement on eight statements (e.g. ‘I can get the AI to listen to what I say’, ‘I think I have a great deal of power’) using a seven-point scale (1 = disagree strongly to 7 = agree strongly; α = 0.90). To increase the internal validity of the study and hence control for the potentially confounding factors of self-esteem and affective influences, participants also replied to the well-validated single-item self-esteem scale (Robins, Hendin and Trzesniewski, 2001) by indicating their agreement with the statement ‘I have high self-esteem’ (1 = disagree strongly to 5 = agree strongly), after which they replied to the ten-item International Positive and Negative Affect Schedule (PANAS) Short Form scale (1 = never to 5 = always; α_PA = 0.78, α_NA = 0.86; Thompson, 2007).4 Finally, they replied to the same manipulation and realism checks as those used in Study 1B.

Results

To examine whether participants’ personal sense of power, shown in Figure 3A, mediates the effect of AI autonomy on participants’ AI adoption intentions, shown in Figure 3B, we performed a mediation analysis (PROCESS Model 4; Hayes, 2018) with AI autonomy (high = 1, low = 0) as the predictor, personal sense of power as the mediator and the composite index for adoption intentions as the outcome variable. 95% confidence intervals (CIs) were computed using 5000 bootstrap samples.

The model explained roughly 33% of the variance in consumers’ AI adoption intentions (R² = 0.33), with the results revealing a significant and negative effect of high (vs. low) AI autonomy (b = −1.48, SE = 0.20, p < 0.001). Personal sense of power, in contrast, had a significant and positive effect on AI adoption intentions (b = 0.67, SE = 0.08, p < 0.001). However, when AI adoption intentions were regressed on both AI autonomy and personal sense of power, the effect of AI autonomy on adoption intentions was clearly reduced (b = −0.92, SE = 0.20, p < 0.001). The bootstrapped indirect effect of AI autonomy through personal sense of power was significantly different from zero (95% CI: [−0.81, −0.35]). Thus, in line with H2, high (vs. low) AI autonomy resulted in significantly lower adoption intentions as the more (vs. less) autonomous AI shopping assistant decreased participants’ personal sense of power.5 An additional meditation analysis that controlled for participants’ self-esteem, positive and negative affect, age and gender showed no consistent influence of any of these variables on AI adoption intentions, nor did the inclusion of these variables change the nature or significance of our results.

Discussion

Study 2 provides evidence that AI high in autonomy induces a state of powerlessness in consumers, whereas AI low in autonomy induces a state of powerfulness. Importantly, we find support for the mechanism that explains the negative effect of high (vs. low) AI autonomy on consumers’ adoption intentions. More precisely, our results show that the shift in autonomy associated with a more autonomous AI induces a state of powerlessness in consumers, driving their AI adoption intentions in an aversive way. Further, we rule out a series of alternative accounts, such as affective influences or effects emerging due to self-esteem. However, we have yet to demonstrate a relevant boundary condition for our focal effect; a topic we address in Study 3.

Study 3: Moderation through scarcity

In Study 3, we test the degree to which the presence (vs. absence) of scarcity cues can attenuate the negative effect of high (vs. low) AI autonomy on consumers’ AI adoption intentions (H3). According to our theorizing, consumers who face a situation characterized by scarcity, such as shopping during a narrow time window when products are sold out rapidly, should be less reluctant to adopt an AI high (vs. low) in autonomy, despite a shift in autonomy towards the AI. Therefore, we posit that the general negative effect of high-autonomy AI should be reduced when consumers use such an AI in settings characterized by scarcity, because the higher autonomy of the AI should not be perceived as a loss of autonomy to the same extent under such conditions (Lynn, 1992). To enhance realism and boost the ecological validity of our work (Loebnitz et al., 2022; Morales, Amir and Lee, 2017; Otterbring, 2021), we test this thesis in a setting wherein participants make an active choice between a high- and low-autonomy AI assistant.

Participants, design and procedure

Responses from US consumers (N = 417; age: M = 43.0, SD = 13.4; 51.3% female) were collected from the online panel Prolific. Participants were exposed to two scenarios (within-subjects) in which they were asked to imagine shopping for a new vacuum cleaner and, for each scenario, were offered a high- or a low-autonomy AI assistant to help them.

In the scarcity condition, participants were told that the availability of the target product (i.e. the vacuum cleaner) would be severely limited in the near future. The control condition did not emphasize that product availability constituted a concern, with desirable options instead described as having excellent availability in the near future.6

After having made their AI shopping assistant choice for each scenario (scarcity vs. control), participants indicated their responses on one manipulation check per scenario, in which they rated the availability of desirable options of vacuum cleaners over the next couple of days on a scale from 1 = severely limited to 7 = completely unrestricted. Finally, they replied to the same realism check used in the former studies.

Results

Consistent with H3, a 2 (condition: scarcity vs. control) × 2 (consumer choice: high vs. low AI autonomy) Pearson's chi-square analysis revealed that the proportion of participants choosing the shopping assistant characterized by high (vs. low) AI autonomy was significantly greater in the scarcity condition (21.1%) than in the control condition (15.6%; χ²(1) = 4.23, p = 0.040).

Discussion

The results of Study 3 demonstrate that scarcity shifts consumer choice more towards high-autonomy AI relative to the shopping settings used across all former studies that lacked explicit cues to scarcity. Thus, although consumers are generally reluctant to prefer and choose high-autonomy AI, even under conditions of scarcity, this aversion is reduced when products are described as only having limited availability, irrespective of whether such product availability aspects are made salient in isolation or are combined with objective manipulations of time constraints.

Study 4: Testing the entire moderated mediation model

In Study 4, we test whether the general negative effect of AI autonomy on consumers’ AI adoption intentions is mediated by their personal sense of power and moderated by scarcity, with the mediation through personal power only occurring under regular shopping settings but not those characterized by scarcity (H4). According to our theorizing, and as observed in Study 3, consumers who face a shopping situation where the scarcity concept is made salient should be less reluctant to adopt an AI high in autonomy, despite a shift in autonomy from themselves to the AI. We posit that this occurs because the higher autonomy of the AI, unlike regular shopping settings, should no longer be perceived as a personal loss of power under conditions of scarcity. To rule out alternative accounts, we capture participants’ desire for unique consumer goods (Lynn, 1992). Prior research has discussed powerlessness in terms of consumers being more prone to prefer exclusive and unique products (Mandel et al., 2017; Rucker and Galinsky, 2008; Zou et al., 2014). Although the power construct is distinct from various facets of uniqueness (Folwarczny et al., 2021), a critic might argue that our mechanism is not necessarily restricted to a sense of powerlessness. Study 4 sought to exclude this possibility and thus add further support to our precise psychological mechanism (personal sense of power), while simultaneously demonstrating the robustness and replicability of our moderator (scarcity).

Participants, design and procedure

US consumers (N = 221; age: M = 38.8, SD = 11.3; 54.8% female), recruited from Amazon's Mechanical Turk panel, were randomly assigned to one of four experimental conditions in a 2 (AI autonomy: high, low) × 2 (scarcity: yes, no) between-subjects design. They were instructed to imagine being in the process of purchasing a digital camera, either during Black Friday ‘with good options being sold out very quickly’ or, alternatively, in the next couple of months ‘with good options available at all times’ (without any reference to Black Friday). As before, they were offered either an AI shopping assistant high in autonomy, or alternatively, an AI assistant low in autonomy.

After reading their assigned scenario, participants replied to the same primary measures formerly used to capture our mediator and dependent variable, with the index variables of personal sense of power (α = 0.87) and adoption intentions (α = 0.83) found to be reliable. As a manipulation check of scarcity, participants rated the statement ‘In the described scenario, the availability of the digital camera was…’ (1 = very low to 7 = very high). Further, we included the eight-item Desire for Unique Consumer Products scale (adapted from Lynn and Harris, 1997) with items such as ‘I am very attracted to rare objects’ and ‘I am more likely to buy a product if it is scarce’ (1 = strongly disagree to 5 = strongly agree), and combined the responses into a composite index (α = 0.86) to control for the potential confound of this construct in the analyses.

Results

To examine our focal hypothesis (H4), whereby the impact of AI autonomy on consumers’ adoption intentions should be mediated through their personal sense of power under regular shopping conditions, but not in settings characterized by scarcity, we performed a moderated mediation analysis with participants’ personal sense of power (continuous) as mediator and scarcity (yes = 1, no = 0) as moderator (PROCESS Model 8; Hayes, 20182018). AI autonomy (high = 1, low = 0) was the predictor, and adoption intentions (continuous) acted as the outcome variable. We used 5000 bootstrap samples to compute the 95% CIs.

In support of H4, the conditional indirect effect of high (vs. low) AI autonomy through personal sense of power on adoption intentions was not significant under conditions of scarcity (b = −0.06, SE = 0.08, 95% CI: [−0.21, 0.11]) but was significant and negative under regular circumstances (b = −0.31, SE = 0.11, 95% CI: [−0.54, −0.12]), thereby replicating the pattern of our earlier studies. Supporting this finding, the index of moderated mediation was significant (b = 0.26, SE = 0.14, 95% CI: [0.02, 0.56]), confirming that the indirect effect of AI autonomy on adoption intentions through consumers’ personal sense of power was indeed moderated by scarcity. Additional moderated mediation analyses controlling for participants’ desire for unique consumer products, besides age and gender, showed no consistent influence of these factors on our focal variables, and did not change the nature or significance of our results. The ratings for consumers’ personal sense of power and adoption intentions by AI autonomy × scarcity conditions are depicted in Figure 4.

Discussion

The results from Study 4 conceptually replicate the mediating role of consumers’ personal sense of power in the (no-scarcity) control condition, such that consumers’ adoption intentions are consistently lower for high relative to low AI autonomy due to a sense of powerlessness in such regular shopping settings. Importantly, the negative influence of high (vs. low) AI autonomy on consumers’ perceived personal sense of power levels and adoption intentions was neutralized when the shopping setting was characterized by scarcity, thus highlighting an important boundary condition of the previously demonstrated aversive aspects associated with high AI autonomy.

General discussion

The present research investigates consumers’ intentions to adopt AI as a function of the AI's autonomy. The results reveal a robust, negative influence of high (vs. low) AI autonomy on AI adoption in the context of online shopping, as long as the shopping situation is not characterized by scarcity. We also find that this effect is driven by a state of powerlessness linked to consumers’ transfer of agency from themselves towards the more autonomous AI. These findings challenge the notion that more autonomous AI technologies will be superior in generating favourable consumer responses. If anything, our studies reveal the reverse – as long as the consumption context does not counter the aversive aspects associated with AI autonomy, as demonstrated in the presence of scarcity.

Theoretical contribution

From a theoretical perspective, our work makes three key contributions. First, the current studies contribute to the growing stream of research on consumer adoption of AI technologies (Frank et al., 2023; Hajli et al., 2022; Leone et al., 2021; Sestino et al., 2022) by empirically validating consumers’ need for autonomy as a central construct in consumption contexts that feature increasingly autonomous technology. Contrary to prevailing assumptions and anecdotal evidence (Dietvorst, Simmons and Massey, 2018; Wen et al., 2022), we find that AI autonomy does not generally add value for customers, although shopping settings characterized by scarcity seem to attenuate the aversion towards high (vs. low) AI autonomy. This finding underscores the complex interplay of consumer agency and AI autonomy (Dorotic, Stagno and Warlop, 2023; Mourali and Nagpal, 2013; Wertenbroch et al., 2020), and also adds to the ongoing discussion about automation anxiety in other AI adoption contexts (Kelan, 2023).

Second, the results show that the level of autonomy of AI shopping assistants has downstream effects on consumers’ personal sense of power, substantiating earlier speculations that high AI autonomy might be less preferred by consumers (Carmon et al., 2019; see also Schweitzer and Van den Hende, 2016). Our extensive testing of the psychological mechanism driving consumers’ adoption intentions of high (vs. low) AI autonomy through their personal sense of power thus contributes to earlier findings, while ruling out alternative accounts for AI reluctance, such as those linked to different consumption contexts (Longoni and Cian, 2020), consumers’ self-esteem (Wang, 2020) and their need for uniqueness (Longoni, Bonezzi and Morewedge, 2019), as well as a wide array of other individual difference factors, including the Big Five personality traits (Cerekovic, Aran and Gatica-Perez, 2017), consumers’ technology readiness (Flavián et al., 2022) and their purchase-decision involvement (Sihi, 2018), as well as subjective and objective knowledge about AI (Cadario, Longoni and Morewedge, 2021) and prior experiences with AI (Frank et al., 2023).

Third, by documenting the boundary condition of scarcity, the present research offers a more nuanced understanding of whether, why and when consumers are particularly prone to rely on more autonomous AI assistance. Specifically, our results document that scarcity neutralizes the negative effect of high (vs. low) AI autonomy on consumers’ AI adoption intentions, with scarcity motivating a relative shift towards high-autonomy AI alternatives. As such, our studies paint a more complete picture regarding under what circumstances customers might be willing to forego their own autonomy (Botti, Iyengar and McGill, 2023).

Practical implications

Our findings suggest that marketers should implement strategies that help overcome, or at least reduce, consumers’ perceived power loss to counter their strong reluctance towards highly autonomous AI assistants, which will be increasingly prevalent in the future retailing landscape. These strategies could rely on implementing mechanisms that retain some level of control over the AI system (de Bellis and Johar, 2020; Whang et al., 2021) or allow consumers to receive more nuanced explanations for the rationale behind autonomous actions (Zhang et al., 2022). In the example used at the beginning of this paper, Amazon Alexa proactively informs users why and when it has taken action though ‘hunches,’ and gives users the option to object to future actions of that sort. Similarly, to preserve autonomy of one's own choices, our findings suggest that it might be better if the AI only notifies consumers that a given action can be performed. The AI should then wait for consumer approval before proceeding with the action, abandoning the concept of AI making these decisions proactively (Otterbring et al., 2023). Another approach to tackle the powerlessness induced by high AI autonomy might be to implement self-affirmative actions into the routines of AI assistants, given that self-affirmation can offset powerlessness with potential effects on subsequent consumption responses (Moeini-Jazani, Albalooshi and Seljeseth, 2019; Sivanathan and Pettit, 2010). An example of such an approach would be to program AI systems to regularly communicate their goals and offer assistance when they sense that customers likely encounter problems, have inquiries or face insecurities.

Another implication that emerges from this research is that AI adoption may be more successful in shopping settings where autonomous AI assistants are less likely to induce a sense of powerlessness in consumers. Our results show that the negative effect of high-autonomy AI shopping assistants is less prevalent in situations that convey cues to scarcity. This finding implies that consumers are more willing to transfer autonomy to AI assistants in certain contexts as a means to avoid sole responsibility over outcomes resulting from their decisions (Botti, Orfali and Iyengar, 2009). Thus, AI autonomy does not necessarily harm consumer adoption across all consumption contexts, as long as marketers can leverage appropriate situations (e.g. promotional activities during a predefined and narrow time window) or marketing messages (e.g. last chance to buy, limited edition, final offer), such that consumers’ personal sense of power remains, without the experience that autonomy is lost in the purchasing process. However, these tactics should be used with caution, as repeated exposure to scarcity appeals when promoting AI high in autonomy can backfire and lead to reputational costs for companies if claims related to the limited availability of certain products – in time, space or quantity – are based on false information and communicated in an insincere way (cf. Otterbring and Folwarczny, 2024).

Our research complements the endeavours of law- and policymakers in various regions, such as the European Union, United States and Canada, who are working to establish regulations for AI system design. These regulations are driven by concerns about the potential adverse societal implications and ethical dilemmas associated with AI (Tiberius, Gojowy and Dabić, 2022) and other impactful modern technologies adopted by consumers (Nyhus et al., 2023). Our findings indicate that implementing a labelling scheme for the different levels of AI autonomy could be beneficial, as it might help consumers understand if, when and how AI assistants are going to make critical actions and decisions on their behalf. These labels may relate to actions (‘this purchase was performed by an AI’), communications (‘this message was sent by an AI’) and representations (‘this system is automated and operated by an AI’), allowing for traceability and accountability of AI. Importantly, the use of a labelling scheme would empower consumers to deliberately choose in favour of high-autonomy AI, whenever the transfer of agency is deemed superior.

Limitations and future research

While we find a robust negative effect of high (vs. low) AI autonomy on consumers’ AI adoption intentions, with no signs of moderation through the individual difference factors included, we acknowledge that other individual differences not addressed herein could still moderate our obtained results. Similarly, despite the fact that consumers’ prior experiences with AI did not moderate our focal effect, this finding does not necessarily generalize after repeated exposure to, and experiences with, highly autonomous AI. That is, we expect consumer experiences with AI to develop over time, likely leading to greater acceptance of high AI autonomy. Hence, future research may benefit from studying consumer responses towards AI autonomy longitudinally, or over a longer time span, to capture how current and future AI offerings are shaping, changing and developing consumers’ AI adoption intentions.

Scarcity in promotional sales settings almost always entails some aspects of both time and product scarcity, given that the limited availability of products would naturally result in options being sold out rapidly. Nevertheless, in an attempt to separate these sources of scarcity, we tested for potential differences in consumers’ choice between high (vs. low)-autonomy AI assistants, depending on whether our scarcity manipulation did or did not include an objective manipulation of time constraints (Study 3). We found no difference in consumer choice across these scarcity settings, although consumers in the combined scarcity condition were significantly more inclined to choose the high-autonomy AI assistant compared to those in the control condition. This finding tentatively suggests that our focal effect may emerge, even without objective manipulations of time constraints. However, we maintain that it is challenging to isolate the effects of time and product scarcity in experimental settings, meaning that these tentative conclusions need to be validated in future studies.

This research, while novel and managerially relevant, is restricted to weak AI. As indicated in the introduction to this paper, the development of more intelligent AI is around the corner (Linde and Schweizer, 2019). While existing AI systems may not (yet) have reached the stage of superintelligence (Bostrom, 2014), AI development is happening rapidly (Scott et al., 2022), implying that future AI will likely continue to evolve capabilities that challenge managers’ current understanding of consumer responses. Intertwined with autonomy, such capabilities may also be linked to the amount and type of information AI gathers about consumers by tracking their preferences, habits and behaviours to reliably make suggestions or execute certain actions. Tracking of this type may be perceived as an erosion of privacy (Baum et al., 2023). Thus, further research is needed to examine the impeding factors that might slow down not only consumers’ current adoption rates of increasingly autonomous AI, but also their future adoption rates of strong AI.

Conclusions

In this paper, we investigated the influence of AI autonomy on consumers’ intentions to adopt AI assistants in online shopping settings. We found that high (vs. low) AI autonomy can lead to a sense of powerlessness in consumers, which in turn lowers their AI adoption intentions. However, this effect is mitigated in situations where popular options are less available or quickly sold out. Our findings have implications for both theory and practice, as they empirically validate consumers’ fundamental need for autonomy in AI-related consumption contexts. Accordingly, managers should carefully consider the level of AI autonomy and the context in which it will be used when developing and implementing AI-powered technologies. Together, this research highlights the importance for managers, policymakers and researchers to closely focus on the shift in power dynamics between consumers and increasingly autonomous intelligent machines when seeking to understand consumers’ current and future adoption rates of AI assistants.

Transparency statement

In conducting and reporting this research, we have adhered to all APA ethical guidelines as well as the ethical guidelines of Denmark. The results have not been previously published, are reported honestly and the authorship accurately reflects each individual's contributions. The project was approved by the Institutional Review Board at Aarhus University (No. 2021-53). The first author managed the collection of data for all studies using the consumer panels from Prolific (Studies 1A and B, 3 and A1) and Amazon's Mechanical Turk (Studies 2 and 4) in the spring and summer of 2021 and spring of 2023. Participants were unique to each study and did not participate in more than one study. The data were jointly analysed, reported and discussed by both authors. The anonymised data are stored in a public project directory on the Open Science Framework: https://osf.io/undmp/.

Biographies

Darius-Aurel Frank is an Assistant Professor of Marketing in the Department of Management, Aarhus University (Denmark). His research focuses on life with technology in the future, with an emphasis on exploring the dynamic relationship between marketing, innovation and sustainability. His work has appeared in esteemed journals across different fields including Nature Human Behavior, Technological Forecasting & Social Change, Information Technology & People and Psychology & Marketing.
Tobias Otterbring is a Professor of Marketing in the Department of Management, University of Agder (Norway) and a member of the Young Academy of Norway. His research focuses on how the real, imagined or implied presence of others influences consumers’ cognitions, emotions and actual purchase or choice behaviours. His work has appeared in journals such as Nature Human Behaviour, Nature Communications and PNAS, as well as in top-tier publications in marketing, management and psychology, including Journal of Marketing Research, Organizational Behavior and Human Decision Processes, British Journal of Management, Journal of Experimental Social Psychology and Journal of Experimental Psychology: Applied.

Supporting Information

References

Abendroth, L. J. and K. Diehl (2006). ‘Now or never: effects of limited purchase opportunities on patterns of regret over time’, Journal of Consumer Research, 33, pp. 342–351.
10.1086/508438
Web of Science® Google Scholar
Alfalih, A. A. (2022). ‘Customer engagement design during the COVID 19 pandemic, mutual trust and intelligent automation: a conceptual perspective’, Journal of Innovation and Entrepreneurship, 11, art. 32.
10.1186/s13731-022-00222-7
PubMed Google Scholar
Ali, M. S., I. A. Swiety and M. H. Mansour (2022). ‘Evaluating the role of artificial intelligence in the automation of the banking services industry: evidence from Jordan’, Humanities and Social Sciences Letters, 10, pp. 383–393.
10.18488/73.v10i3.3090
Google Scholar
Anderson, C. and A. D. Galinsky (2006). ‘Power, optimism, and risk-taking’, European Journal of Social Psychology, 36, pp. 511–536.
10.1002/ejsp.324
Web of Science® Google Scholar
Anderson, C., J. A. D. Hildreth and L. Howland (2015). ‘Is the desire for status a fundamental human motive? A review of the empirical literature’, Psychological Bulletin, 141, pp. 574–601.
10.1037/a0038781
PubMed Web of Science® Google Scholar
Anderson, C., O. P. John and D. Keltner (2012). ‘The personal sense of power’, Journal of Personality, 80, pp. 313–344.
10.1111/j.1467-6494.2011.00734.x
PubMed Web of Science® Google Scholar
Aw, E. C.-X., L.-Y. Leong, J.-J. Hew, N. P. Rana, T. M. Tan and T.-W. Jee (2023). ‘Counteracting dark sides of robo-advisors: justice, privacy and intrusion considerations’, International Journal of Bank Marketing, https://doi.org/10.1108/IJBM-10-2022-0439.
10.1108/IJBM?10?2022?0439
Web of Science® Google Scholar
Bacharach, S. B. (1989). ‘Organizational theories: some criteria for evaluation’, The Academy of Management Review, 14, art. 496.
10.2307/258555
Web of Science® Google Scholar
Balakrishnan, J. and Y. K. Dwivedi (2021). ‘Role of cognitive absorption in building user trust and experience’, Psychology & Marketing, 38, pp. 643–668.
10.1002/mar.21462
Web of Science® Google Scholar
Baum, K., J. Bryson, F. Dignum, V. Dignum, M. Grobelnik, H. Hoos, et al. (2023). ‘From fear to action: AI governance and opportunities for all’, Frontiers in Computer Science, 5, art. 1210421.
10.3389/fcomp.2023.1210421
Web of Science® Google Scholar
Bostrom, N. (2014). Superintelligence: Paths, Dangers, Strategies. Oxford: Oxford University Press.
Google Scholar
Botti, S., K. Orfali and S. S. Iyengar (2009). ‘Tragic choices: autonomy and emotional responses to medical decisions’, Journal of Consumer Research, 36, pp. 337–352.
10.1086/598969
Web of Science® Google Scholar
Botti, S., S. S. Iyengar and A. L. McGill (2023). ‘Choice freedom’, Journal of Consumer Psychology, 33, pp. 143–166.
10.1002/jcpy.1325
Web of Science® Google Scholar
Cadario, R., C. Longoni and C. K. Morewedge (2021). ‘Understanding, explaining, and utilizing medical artificial intelligence’, Nature Human Behaviour, 5, pp. 1636–1642.
10.1038/s41562-021-01146-0
PubMed Web of Science® Google Scholar
Campbell, I. C. (2021). ‘ Amazon's Alexa can now act on its own hunches to turn off lights and more’. Available at https://www.theverge.com/2021/1/25/22249044/amazon-alexa-update-proactive-hunches-guard-plus-subscription.
Google Scholar
Carmon, Z., R. Schrift, K. Wertenbroch and H. Yang (2019). ‘ Designing AI systems that customers won't hate’. Available at https://sloanreview.mit.edu/article/designing-ai-systems-that-customers-wont-hate/.
Google Scholar
Casidy, R., M. Claudy, S. Heidenreich and E. Camurdan (2021). ‘The role of brand in overcoming consumer resistance to autonomous vehicles’, Psychology & Marketing, 38, pp. 1101–1121.
10.1002/mar.21496
Web of Science® Google Scholar
Cerekovic, A., O. Aran and D. Gatica-Perez (2017). ‘Rapport with virtual agents: what do human social cues and personality explain?’, IEEE Transactions on Affective Computing, 8, pp. 382–395.
10.1109/TAFFC.2016.2545650
Web of Science® Google Scholar
Choi, C. and A. S. Mattila (2016). ‘The effects of other customers’ dress style on customers’ approach behaviors: the moderating role of sense of power’, Cornell Hospitality Quarterly, 57, pp. 211–218.
10.1177/1938965515619228
Web of Science® Google Scholar
Chua, A. Y. K., A. Pal and S. Banerjee (2023). ‘AI-enabled investment advice: will users buy it?’, Computers in Human Behavior, 138, art. 107481.
10.1016/j.chb.2022.107481
Web of Science® Google Scholar
de Bellis, E. and G. Venkataramani Johar (2020). ‘Autonomous shopping systems: identifying and overcoming barriers to consumer adoption’, Journal of Retailing, 96, pp. 74–87.
10.1016/j.jretai.2019.12.004
Web of Science® Google Scholar
Denegri-Knott, J., D. Zwick and J. E. Schroeder (2006). ‘Mapping consumer power: an integrative framework for marketing and consumer research’, European Journal of Marketing, 40, pp. 950–971.
10.1108/03090560610680952
Web of Science® Google Scholar
Dietvorst, B. J., J. P. Simmons and C. Massey (2015). ‘Algorithm aversion: people erroneously avoid algorithms after seeing them err’, Journal of Experimental Psychology: General, 144, pp. 114–126.
10.1037/xge0000033
PubMed Web of Science® Google Scholar
Dietvorst, B. J., J. P. Simmons and C. Massey (2018). ‘Overcoming algorithm aversion: people will use imperfect algorithms if they can (even slightly) modify them’, Management Science, 64, pp. 1155–1170.
10.1287/mnsc.2016.2643
Web of Science® Google Scholar
Dorotic, M., E. Stagno and L. Warlop (2023). ‘AI on the street: context-dependent responses to artificial intelligence’, International Journal of Research in Marketing, https://doi.org/10.1016/j.ijresmar.2023.08.010.
10.1016/j.ijresmar.2023.08.010
Google Scholar
Elbæk, C. T., P. Mitkidis, L. Aarøe and T. Otterbring (2023). ‘Subjective socioeconomic status and income inequality are associated with self-reported morality across 67 countries’, Nature Communications, 14, art. 5453.
10.1038/s41467-023-41007-0
CAS PubMed Web of Science® Google Scholar
Evanschitzky, H. and M. Goergen (2018). ‘Looking forward, looking back: British Journal of Management 2000–2015’, British Journal of Management, 29, pp. 3–9.
10.1111/1467-8551.12257
Web of Science® Google Scholar
Evanschitzky, H., G. R. Iyer, K. G. Pillai, P. Kenning and R. Schütte (2015). ‘Consumer trial, continuous use, and economic benefits of a retail service innovation: the case of the personal shopping assistant’, Journal of Product Innovation Management, 32, pp. 459–475.
10.1111/jpim.12241
Web of Science® Google Scholar
Fan, L. (Sophie), D. D. Rucker and Y. Jiang (2022). ‘Power and need-for-justification: asymmetrical effects on senders and receivers in marketing communications’, Journal of Consumer Research, 50, pp. 236–254.
10.1093/jcr/ucac052
Web of Science® Google Scholar
Faraji-Rad, A., S. Melumad and G. V. Johar (2017). ‘Consumer desire for control as a barrier to new product adoption’, Journal of Consumer Psychology, 27, pp. 347–354.
10.1016/j.jcps.2016.08.002
Web of Science® Google Scholar
Fernandes, T. and E. Oliveira (2021). ‘Understanding consumers’ acceptance of automated technologies in service encounters: drivers of digital voice assistants adoption’, Journal of Business Research, 122, pp. 180–191.
10.1016/j.jbusres.2020.08.058
Web of Science® Google Scholar
Fink, P. D. S., M. Alsamsam, J. R. Brown, H. D. Kindler and N. A. Giudice (2023). ‘Give us something to chauffeur it: exploring user needs in traditional and fully autonomous ridesharing for people who are blind or visually impaired’, Transportation Research Part F: Traffic Psychology and Behaviour, 98, pp. 91–103.
10.1016/j.trf.2023.09.004
Web of Science® Google Scholar
Flavián, C., A. Pérez-Rueda, D. Belanche and L. V. Casaló (2022). ‘Intention to use analytical artificial intelligence (AI) in services – the effect of technology readiness and awareness’, Journal of Service Management, 33, pp. 293–320.
10.1108/JOSM-10-2020-0378
Web of Science® Google Scholar
Folwarczny, M., N. P. Li, V. Sigurdsson, L. K. L. Tan and T. Otterbring (2021). ‘Development and psychometric evaluation of the Anticipated Food Scarcity Scale (AFSS)’, Appetite, 166, art. 105474.
10.1016/j.appet.2021.105474
PubMed Web of Science® Google Scholar
Folwarczny, M., T. Otterbring, V. Sigurdsson and A. Gasiorowska (2022). ‘Seasonal cues to food scarcity and calorie cravings: winter cues elicit preferences for energy-dense foods’, Food Quality and Preference, 96, art. 104379.
10.1016/j.foodqual.2021.104379
Web of Science® Google Scholar
Frank, D.-A. and T. Otterbring (2023). ‘Being seen… by human or machine? Acknowledgment effects on customer responses differ between human and robotic service workers’, Technological Forecasting and Social Change, 189, art. 122345.
10.1016/j.techfore.2023.122345
Web of Science® Google Scholar
Frank, D.-A., C. T. Elbæk, C. K. Børsting, P. Mitkidis, T. Otterbring and S. Borau (2021). ‘Drivers and social implications of artificial intelligence adoption in healthcare during the COVID-19 pandemic’, PLoS ONE, 16, art. e0259928.
10.1371/journal.pone.0259928
CAS PubMed Web of Science® Google Scholar
Frank, D.-A., L. F. Jacobsen, H. A. Søndergaard and T. Otterbring (2023). ‘In companies we trust: consumer adoption of artificial intelligence services and the role of trust in companies and AI autonomy’, Information Technology & People, 36, pp. 155–173.
10.1108/ITP-09-2022-0721
Web of Science® Google Scholar
Fuchs, C., E. Prandelli and M. Schreier (2010). ‘The psychological effects of empowerment strategies on consumers’ product demand’, Journal of Marketing, 74, pp. 65–79.
10.1509/jmkg.74.1.65
Web of Science® Google Scholar
Funder, D. C. and D. J. Ozer (2019). ‘Evaluating effect size in psychological research: sense and nonsense’, Advances in Methods and Practices in Psychological Science, 2, pp. 156–168.
10.1177/2515245919847202
Web of Science® Google Scholar
Galinsky, A. D., D. H. Gruenfeld and J. C. Magee (2003). ‘From power to action’, Journal of Personality and Social Psychology, 85, pp. 453–466.
10.1037/0022-3514.85.3.453
PubMed Web of Science® Google Scholar
Gasiorowska, A., M. Folwarczny and T. Otterbring (2022). ‘Anxious and status signaling: examining the link between attachment style and status consumption and the mediating role of materialistic values’, Personality and Individual Differences, 190, art. 111503.
10.1016/j.paid.2022.111503
Web of Science® Google Scholar
Graham, J. (2013). ‘Mapping the moral maps: from alternate taxonomies to competing predictions’, Personality and Social Psychology Review, 17, pp. 237–241.
10.1177/1088868313492020
PubMed Web of Science® Google Scholar
Griskevicius, V., N. J. Goldstein, C. R. Mortensen, J. M. Sundie, R. B. Cialdini and D. T. Kenrick (2009). ‘Fear and loving in Las Vegas: evolution, emotion, and persuasion’, Journal of Marketing Research, 46, pp. 384–395.
10.1509/jmkr.46.3.384
PubMed Web of Science® Google Scholar
Hajli, N., U. Saeed, M. Tajvidi and F. Shirazi (2022). ‘Social bots and the spread of disinformation in social media: the challenges of artificial intelligence’, British Journal of Management, 33, pp. 1238–1253.
10.1111/1467-8551.12554
Web of Science® Google Scholar
Hambrick, D. C. (2007). ‘The field of management's devotion to theory: too much of a good thing?’, Academy of Management Journal, 50, pp. 1346–1352.
10.5465/amj.2007.28166119
Web of Science® Google Scholar
Hamilton, R., D. Thompson, S. Bone, L. N. Chaplin, V. Griskevicius, K. Goldsmith, et al. (2019). ‘The effects of scarcity on consumer decision journeys’, Journal of the Academy of Marketing Science, 47, pp. 532–550.
10.1007/s11747-018-0604-7
Web of Science® Google Scholar
Hayes, A. F. (2018). Introduction to Mediation, Moderation, and Conditional Process Analysis: A Regression-Based Approach (Methodology in the Social Sciences) (2nd ed.). New York, NY: The Guilford Press.
Google Scholar
Hu, Q., Y. Lu, Z. Pan, Y. Gong and Z. Yang (2021). ‘Can AI artifacts influence human cognition? The effects of artificial autonomy in intelligent personal assistants’, International Journal of Information Management, 56, art. 102250.
10.1016/j.ijinfomgt.2020.102250
Web of Science® Google Scholar
Huang, Y. and L. Qian (2021). ‘Understanding the potential adoption of autonomous vehicles in China: the perspective of behavioral reasoning theory’, Psychology & Marketing, 38, pp. 669–690.
10.1002/mar.21465
Web of Science® Google Scholar
Kelan, E. K. (2023). ‘Automation anxiety and augmentation aspiration: subtexts of the future of work’, British Journal of Management, 34, pp. 2057–2074.
10.1111/1467-8551.12679
Web of Science® Google Scholar
Keltner, D., D. H. Gruenfeld and C. Anderson (2003). ‘Power, approach, and inhibition’, Psychological Review, 110, pp. 265–284.
10.1037/0033-295X.110.2.265
PubMed Web of Science® Google Scholar
Kim, S. Y., B. H. Schmitt and N. M. Thalmann (2019). ‘Eliza in the uncanny valley: anthropomorphizing consumer robots increases their perceived warmth but decreases liking’, Marketing Letters, 30, pp. 1–12.
10.1007/s11002-019-09485-9
Web of Science® Google Scholar
Ko, G., D. L. Roberts, H. Perks and M. Candi (2022). ‘Effectuation logic and early innovation success: the moderating effect of customer co-creation’, British Journal of Management, 33, pp. 1757–1773.
10.1111/1467-8551.12548
Web of Science® Google Scholar
Kurtz, J. L. (2008). ‘Looking to the future to appreciate the present: the benefits of perceived temporal scarcity’, Psychological Science, 19, pp. 1238–1241.
10.1111/j.1467-9280.2008.02231.x
PubMed Web of Science® Google Scholar
Lee, S. M., G. Ryu and S. Chun (2018). ‘Perceived control and scarcity appeals’, Social Behavior and Personality: An International Journal, 46, pp. 361–374.
10.2224/sbp.6367
Web of Science® Google Scholar
Leone, D., F. Schiavone, F. P. Appio and B. Chiao (2021). ‘How does artificial intelligence enable and enhance value co-creation in industrial markets? An exploratory case study in the healthcare ecosystem’, Journal of Business Research, 129, pp. 849–859.
10.1016/j.jbusres.2020.11.008
Web of Science® Google Scholar
Leung, E., G. Paolacci and S. Puntoni (2018). ‘Man versus machine: resisting automation in identity-based consumer behavior’, Journal of Marketing Research, 55, pp. 818–831.
10.1177/0022243718818423
Web of Science® Google Scholar
Liang, S., D. Ye and Y. Liu (2021). ‘The effect of perceived scarcity: experiencing scarcity increases risk taking’, The Journal of Psychology, 155, pp. 59–89.
10.1080/00223980.2020.1822770
PubMed Web of Science® Google Scholar
Linde, H. and I. Schweizer (2019). ‘ A white paper on the future of artificial intelligence’, https://doi.org/10.13140/RG.2.2.32564.19844.
10.13140/RG.2.2.32564.19844
Google Scholar
Loebnitz, N., P. Frank and T. Otterbring (2022). ‘Stairway to organic heaven: the impact of social and temporal distance in print ads’, Journal of Business Research, 139, pp. 1044–1057.
10.1016/j.jbusres.2021.10.020
Web of Science® Google Scholar
Longoni, C. and L. Cian (2020). ‘Artificial intelligence in utilitarian vs. hedonic contexts: the “word-of-machine” effect’, Journal of Marketing, 86, pp. 91–108.
10.1177/0022242920957347
Web of Science® Google Scholar
Longoni, C., A. Bonezzi and C. K. Morewedge (2019). ‘Resistance to medical artificial intelligence’, Journal of Consumer Research, 46, pp. 629–650.
10.1093/jcr/ucz013
Web of Science® Google Scholar
Lynn, M. (1992). ‘The psychology of unavailability: explaining scarcity and cost effects on value’, Basic and Applied Social Psychology, 13, pp. 3–7.
10.1207/s15324834basp1301_2
Web of Science® Google Scholar
Lynn, M. and J. Harris (1997). ‘The desire for unique consumer products: a new individual differences scale’, Psychology and Marketing, 14, pp. 601–616.
10.1002/(SICI)1520-6793(199709)14:6<601::AID-MAR5>3.0.CO;2-B
Web of Science® Google Scholar
Magee, J. C. and A. D. Galinsky (2008). ‘8 Social hierarchy: the self-reinforcing nature of power and status’, Academy of Management Annals, 2, pp. 351–398.
10.5465/19416520802211628
Web of Science® Google Scholar
Magee, J. C. and C. W. Frasier (2014). ‘Status and Power: The Principal Inputs to Influence for Public Managers’, Public Administration Review, 74, pp. 307–317. Portico. https://doi.org/10.1111/puar.12203
10.1111/puar.12203
Web of Science® Google Scholar
Mandel, N., D. D. Rucker, J. Levav and A. D. Galinsky (2017). ‘The Compensatory Consumer Behavior Model: how self-discrepancies drive consumer behavior’, Journal of Consumer Psychology, 27, pp. 133–146.
10.1016/j.jcps.2016.05.003
Web of Science® Google Scholar
Meyer-Waarden, L. and J. Cloarec (2022). ‘“Baby, you can drive my car”: psychological antecedents that drive consumers’ adoption of AI-powered autonomous vehicles’, Technovation, 109, art. 102348.
10.1016/j.technovation.2021.102348
Web of Science® Google Scholar
Mittal, B. (1995). ‘A comparative analysis of four scales of consumer involvement’, Psychology and Marketing, 12, pp. 663–682.
10.1002/mar.4220120708
Web of Science® Google Scholar
Moeini-Jazani, M., S. Albalooshi and I. M. Seljeseth (2019). ‘Self-affirmation reduces delay discounting of the financially deprived’, Frontiers in Psychology, 10, art. 1729.
10.3389/fpsyg.2019.01729
PubMed Web of Science® Google Scholar
Morales, A. C., O. Amir and L. Lee (2017). ‘Keeping it real in experimental research – understanding when, where, and how to enhance realism and measure consumer behavior’, Journal of Consumer Research, 44, pp. 465–476.
10.1093/jcr/ucx048
Web of Science® Google Scholar
Mourali, M. and A. Nagpal (2013). ‘The powerful select, the powerless reject: power's influence in decision strategies’, Journal of Business Research, 66, pp. 874–880.
10.1016/j.jbusres.2011.12.005
Web of Science® Google Scholar
Nguyen, T. and A. Malik (2022). ‘A two-wave cross-lagged study on AI service quality: the moderating effects of the job level and job role’, British Journal of Management, 33, pp. 1221–1237.
10.1111/1467-8551.12540
Web of Science® Google Scholar
Nyhus, E. K., D. Frank, M. K. Król and T. Otterbring (2023). ‘Crypto cravings: gender differences in crypto investment intentions and the mediating roles of financial overconfidence and personality’, Psychology & Marketing, https://doi.org/10.1002/mar.21921.
10.1002/mar.21921
PubMed Web of Science® Google Scholar
Osburg, V.-S., V. Yoganathan, W. H. Kunz and S. Tarba (2022). ‘Can (A)I Give You a Ride? Development and Validation of the CRUISE Framework for Autonomous Vehicle Services’, Journal of Service Research, 25, pp. 630–648. https://doi.org/10.1177/10946705221118233
10.1177/10946705221118233
Web of Science® Google Scholar
Otterbring, T. (2021). ‘Peer presence promotes popular choices: a “Spicy” field study on social influence and brand choice’, Journal of Retailing and Consumer Services, 61, art. 102594.
10.1016/j.jretconser.2021.102594
Web of Science® Google Scholar
Otterbring, T. and M. Folwarczny (2024). ‘Social validation, reciprocation, and sustainable orientation: cultivating “clean” codes of conduct through social influence’, Journal of Retailing and Consumer Services, 76, art. 103612.
10.1016/j.jretconser.2023.103612
Google Scholar
Otterbring, T., J. Arsenovic, P. Samuelsson, S. Malodia and A. Dhir (2023). ‘Going the extra mile, now or after a while: the impact of employee proactivity in retail service encounters on customers’ shopping responses’, British Journal of Management, https://doi.org/10.1111/1467-8551.12765.
10.1111/1467-8551.12765
Web of Science® Google Scholar
Parasuraman, A. (2000). ‘Technology Readiness Index (TRI): a multiple-item scale to measure readiness to embrace new technologies’, Journal of Service Research, 2, pp. 307–320.
10.1177/109467050024001
Google Scholar
Piehlmaier, D. M. (2022). ‘Overconfidence and the adoption of robo-advice: why overconfident investors drive the expansion of automated financial advice’, Financial Innovation, 8, art. 14.
10.1186/s40854-021-00324-3
Web of Science® Google Scholar
Pieters, R. and L. Warlop (1999). ‘Visual attention during brand choice: the impact of time pressure and task motivation’, International Journal of Research in Marketing, 16, pp. 1–16.
10.1016/S0167-8116(98)00022-6
Web of Science® Google Scholar
Podsakoff, P. M., S. B. MacKenzie, J.-Y. Lee and N. P. Podsakoff (2003). ‘Common method biases in behavioral research: a critical review of the literature and recommended remedies’, Journal of Applied Psychology, 88, pp. 879–903.
10.1037/0021-9010.88.5.879
PubMed Web of Science® Google Scholar
Qian, L., J. Yin, Y. Huang and Y. Liang (2023). ‘The role of values and ethics in influencing consumers’ intention to use autonomous vehicle hailing services’, Technological Forecasting and Social Change, 188, art. 122267.
10.1016/j.techfore.2022.122267
Web of Science® Google Scholar
Ribeiro, M. A., D. Gursoy and O. H. Chi (2022). ‘Customer acceptance of autonomous vehicles in travel and tourism’, Journal of Travel Research, 61, pp. 620–636.
10.1177/0047287521993578
Web of Science® Google Scholar
Robins, R. W., H. M. Hendin and K. H. Trzesniewski (2001). ‘Measuring global self-esteem: construct validation of a single-item measure and the Rosenberg Self-Esteem Scale’, Personality and Social Psychology Bulletin, 27, pp. 151–161.
10.1177/0146167201272002
Web of Science® Google Scholar
Rucker, D. D. and A. D. Galinsky (2008). ‘Desire to acquire: powerlessness and compensatory consumption’, Journal of Consumer Research, 35, pp. 257–267.
10.1086/588569
Web of Science® Google Scholar
SAE International (2021). ‘ Taxonomy and definitions for terms related to driving automation systems for on-road motor vehicles’. Available at https://www.sae.org/standards/content/j3016_202104/.
Google Scholar
Sandberg, J. and M. Alvesson (2021). ‘Meanings of theory: clarifying theory through typification’, Journal of Management Studies, 58, pp. 487–516.
10.1111/joms.12587
Web of Science® Google Scholar
Schmitt, B. (2020). ‘Speciesism: an obstacle to AI and robot adoption’, Marketing Letters, 31, pp. 3–6.
10.1007/s11002-019-09499-3
Web of Science® Google Scholar
Schweitzer, F. and E. A. Van den Hende (2016). ‘To be or not to be in thrall to the march of smart products’, Psychology & Marketing, 33, pp. 830–842.
10.1002/mar.20920
CAS PubMed Web of Science® Google Scholar
Scott, A. C., J. R. Solórzano, J. D. Moyer and B. B. Hughes (2022). ‘The future of artificial intelligence’, International Journal of Artificial Intelligence and Machine Learning, 2, pp. 1–37.
10.51483/IJAIML.2.1.2022.1-37
Google Scholar
Sestino, A., A. M. Peluso, C. Amatulli and G. Guido (2022). ‘Let me drive you! The effect of change seeking and behavioral control in the artificial intelligence-based self-driving cars’, Technology in Society, 70, art. 102017.
10.1016/j.techsoc.2022.102017
Web of Science® Google Scholar
Shareef, M. A., V. Kumar, Y. K. Dwivedi, U. Kumar, M. S. Akram and R. Raman (2021). ‘A new health care system enabled by machine intelligence: elderly people's trust or losing self control’, Technological Forecasting and Social Change, 162, art. 120334.
10.1016/j.techfore.2020.120334
Web of Science® Google Scholar
Sharma, S., N. Islam, G. Singh and A. Dhir (2022). ‘Why do retail customers adopt artificial intelligence (AI) based autonomous decision-making systems?’, IEEE Transactions on Engineering Management, https://doi.org/10.1109/TEM.2022.3157976.
10.1109/TEM.2022.3157976
Web of Science® Google Scholar
Sihi, D. (2018). ‘Home sweet virtual home: the use of virtual and augmented reality technologies in high involvement purchase decisions’, Journal of Research in Interactive Marketing, 12, pp. 398–417.
10.1108/JRIM-01-2018-0019
Web of Science® Google Scholar
Sivanathan, N. and N. C. Pettit (2010). ‘Protecting the self through consumption: status goods as affirmational commodities’, Journal of Experimental Social Psychology, 46, pp. 564–570.
10.1016/j.jesp.2010.01.006
Web of Science® Google Scholar
Sobolev, D. (2022). ‘Rise of the androids: the reflection of developers’ characteristics in computerized systems’, British Journal of Management, 34, pp. 1632–1654.
10.1111/1467-8551.12653
Web of Science® Google Scholar
Song, C. S. and Y.-K. Kim (2022). ‘The role of the human–robot interaction in consumers’ acceptance of humanoid retail service robots’, Journal of Business Research, 146, pp. 489–503.
10.1016/j.jbusres.2022.03.087
Web of Science® Google Scholar
Soto, C. J. and O. P. John (2017). ‘Short and extra-short forms of the Big Five Inventory-2: the BFI-2-S and BFI-2-XS’, Journal of Research in Personality, 68, pp. 69–81.
10.1016/j.jrp.2017.02.004
Web of Science® Google Scholar
Söderlund, M. and E.-L. Oikarinen (2021). ‘Service encounters with virtual agents: an examination of perceived humanness as a source of customer satisfaction’, European Journal of Marketing, 55, pp. 94–121.
10.1108/EJM-09-2019-0748
Web of Science® Google Scholar
Suri, R., C. Kohli and K. B. Monroe (2007). ‘The effects of perceived scarcity on consumers’ processing of price information’, Journal of the Academy of Marketing Science, 35, pp. 89–100. https://doi.org/10.1007/s11747-006-0008-y
10.1007/s11747-006-0008-y
Web of Science® Google Scholar
Takeuchi, R., P. E. Tesluk, S. Yun and D. P. Lepak (2005). ‘An integrative view of international experience’, Academy of Management Journal, 48, pp. 85–100.
10.5465/amj.2005.15993143
Web of Science® Google Scholar
Talaifar, S., M. D. Buhrmester, Ö. Ayduk and W. B. Swann (2021). ‘Asymmetries in mutual understanding: people with low status, power, and self-esteem understand better than they are understood’, Perspectives on Psychological Science, 16, pp. 338–357.
10.1177/1745691620958003
PubMed Web of Science® Google Scholar
Tang, H., L. Li and S. Su (2022). ‘Experiencing less leads to the use of more: the effect of a scarcity mindset on product usage’, Journal of Business Research, 149, pp. 139–148.
10.1016/j.jbusres.2022.05.024
Web of Science® Google Scholar
Tavitiyaman, P., X. Zhang and W. Y. Tsang (2022). ‘How tourists perceive the usefulness of technology adoption in hotels: interaction effect of past experience and education level’, Journal of China Tourism Research, 18, pp. 64–87.
10.1080/19388160.2020.1801546
Web of Science® Google Scholar
Thompson, E. R. (2007). ‘Development and validation of an internationally reliable short-form of the Positive and Negative Affect Schedule (PANAS)’, Journal of Cross-Cultural Psychology, 38, pp. 227–242.
10.1177/0022022106297301
Web of Science® Google Scholar
Tiberius, V., R. Gojowy and M. Dabić (2022). ‘Forecasting the future of robo advisory: a three-stage Delphi study on economic, technological, and societal implications’, Technological Forecasting and Social Change, 182, art. 121824.
10.1016/j.techfore.2022.121824
Web of Science® Google Scholar
von Walter, B., D. Kremmel and B. Jäger (2022). ‘The impact of lay beliefs about AI on adoption of algorithmic advice’, Marketing Letters, 33, pp. 143–155.
10.1007/s11002-021-09589-1
Web of Science® Google Scholar
Voorveld, H. A. M. and T. Araujo (2020). ‘How social cues in virtual assistants influence concerns and persuasion: the role of voice and a human name’, Cyberpsychology, Behavior, and Social Networking, 23, pp. 689–696.
10.1089/cyber.2019.0205
PubMed Web of Science® Google Scholar
Wang, W., R. Raghunathan and D. K. Gauri (2022). ‘Powerlessness, variety-seeking, and the mediating role of need for autonomy’, Journal of Retailing, 98, pp. 706–723.
10.1016/j.jretai.2022.04.004
Web of Science® Google Scholar
Wang, Y. (2020). ‘When power increases perspective-taking: the moderating role of syncretic self-esteem’, Personality and Individual Differences, 166, art. 110207.
10.1016/j.paid.2020.110207
Web of Science® Google Scholar
Wen, H., L. Zhang, A. Sheng, M. Li and B. Guo (2022). ‘From “human-to-human” to “human-to-non-human” – influence factors of artificial intelligence-enabled consumer value co-creation behavior’, Frontiers in Psychology, 13, art. 863313.
10.3389/fpsyg.2022.863313
PubMed Google Scholar
Wertenbroch, K., R. Y. Schrift, J. W. Alba, A. Barasch, A. Bhattacharjee, M. Giesler, et al. (2020). ‘Autonomy in consumer choice’, Marketing Letters, 31, pp. 429–439.
10.1007/s11002-020-09521-z
PubMed Web of Science® Google Scholar
Whang, J. B., J. H. Song, B. Choi and J.-H. Lee (2021). ‘The effect of augmented reality on purchase intention of beauty products: the roles of consumers’ control’, Journal of Business Research, 133, pp. 275–284.
10.1016/j.jbusres.2021.04.057
Web of Science® Google Scholar
Wirtz, J., P. G. Patterson, W. H. Kunz, T. Gruber, V. N. Lu, S. Paluch and A. Martins (2018). ‘Brave new world: service robots in the frontline’, Journal of Service Management, 29, pp. 907–931.
10.1108/JOSM-04-2018-0119
Web of Science® Google Scholar
Wojciszke, B. and A. Struzynska-Kujalowicz (2007). ‘Power influences self-esteem’, Social Cognition, 25, pp. 472–494.
10.1521/soco.2007.25.4.472
Web of Science® Google Scholar
Zhang, J., M. S. Balaji, J. Luo and S. Jha (2022). ‘Effectiveness of product recommendation framing on online retail platforms’, Journal of Business Research, 153, pp. 185–197.
10.1016/j.jbusres.2022.08.006
Web of Science® Google Scholar
Zou, D., L. Jin, Y. He and Q. Xu (2014). ‘The effect of the sense of power on Chinese consumers’ uniqueness-seeking behavior’, Journal of International Consumer Marketing, 26, pp. 14–28.
10.1080/08961530.2014.848078
Google Scholar

1 Because the order of options was randomized, we checked for differences in participants’ usage intentions using only the scenario they were exposed to first (between-subjects). Mirroring the above result, an independent samples t-test found the same significant effect on intention to use with respect to AI that is high (M_High = 2.56, SD = 1.55) vs. low in autonomy (M_Low = 4.77, SD = 1.34); t(188) = −10.53, p < 0.001, d = −1.53.
2 Although not pre-registered, we also conducted a simple moderation analysis (PROCESS Model 1; Hayes, 2018) to test whether the effect of AI autonomy on consumers’ AI adoption intentions was moderated by their nationality (US, UK), which was not found to be the case.
3 In another experiment, we ruled out that our focal effect was moderated by the product, as hedonic (purchase of a one-week vacation) or utilitarian (purchase of a digital camera; see Study A1 in the Online Appendix).
4 To mitigate common method bias, we varied the scale format of our measures (Podsakoff et al., 2003).
5 To verify the uniqueness of our proposed chain of effects, we ran an additional moderation analysis (PROCESS Model 1; Hayes, 2018) with personal sense of power as a moderator rather than a mediator (all else equal). We found no signs of moderation. This supports the notion that our AI autonomy manipulation changed participants’ momentary sense of power, which created spillover effects on their AI adoption intentions.
6 We actually used two distinct scarcity conditions. In one of them, participants could spend just 7 seconds before making their decision, with the time limit stated as part of the instructions. This time limit mirrors previous manipulations of time constraints (Pieters and Warlop, 1999) and was done to enhance the saliency of time scarcity. The other scarcity condition focused more explicitly on product scarcity, meaning that no objective time constraints existed for participants. While this design difference was originally intended to disentangle the effects of product and time scarcity, we found no difference in consumer choice as a function of these conditions (χ² < 1). Therefore, following conventions (Frank and Otterbring, 2023; Griskevicius et al., 2009), we combined these conditions into a joint scarcity condition to facilitate parsimonious analyses.

Citing Literature

Volume35, Issue4

October 2024

Pages 1700-1723

Autonomy, Power and the Special Case of Scarcity: Consumer Adoption of Highly Autonomous Artificial Intelligence

Abstract

Introduction

Theoretical background

AI, autonomy and consumer choice

Personal sense of power

Scarcity

Overview of studies

Study 1A: Pre-registered test of the main effect

Participants, design and procedure

Results

Discussion

Study 1B: Replication of main effect

Participants, design and procedure

Results

Discussion

Study 2: Mediation through personal sense of power

Participants, design and procedure

Results

Discussion

Study 3: Moderation through scarcity

Participants, design and procedure

Results

Discussion

Study 4: Testing the entire moderated mediation model

Participants, design and procedure

Results

Discussion

General discussion

Theoretical contribution

Practical implications

Limitations and future research

Conclusions

Transparency statement

Biographies

Supporting Information

References

Citing Literature

Figures

References

Related

Information