Keywords

1 Introduction

E-tourism researchers often “chase down” research deemed “sexy” and innovative, with no meaningful charting of ethical ontology within the subject or topic areas. There are benefits to this. For one thing, published scholarship and knowledge dissemination flourishes in real time, as innovative technologies are deployed in industry. Another benefit of the current publication model of e-tourism research is that knowledge organically develops and grows without constraints as scholarly search and scientific discovery exploit new and emerging ontologies. The explosive rate of “smartness” in tourism practice and the rapid pace of published research justifies deeper and more focused inquiry into the identification and treatment of ethical, human, and social bias issues in artificial intelligence (AI) in e-tourism research. It is important to note here, that this study does not focus on the nature of artificially intelligent systems, the technology on which it is built or generated, nor the nature of its use in the production and consumption of tourism. Rather, the paper focuses on the collection of AI research review papers published between 2015–2021 to better understand how ethical issues related to AI in tourism are addressed.

AI research in tourism shows an increasing publication trend in recent years, as evidenced by publication indices (e.g., Scopus, SSCI). Much of this work has adopted a post-positivist stance in seeking to engage and make sense of the disruption AI is having on historically theorized models of tourism. There are specially published spaces within technology-focused (e.g., Journal of Tourism Technology) and non-technology-focused (e.g., Journal of Hospitality and Tourism Technology) titles as well as dedicated published collections (e.g., Annals of Tourism Curated Collection, Robonomics) which have focused primarily on the impact of AI.

At the same time, the global community has grown increasingly wary of “better, smarter, faster” innovations. Customers are keenly interested in understanding how personal data is collected, stored, and used in business processes. Tourism managers and policy influencers, having observed the historical awakening and present-day ideals for ethics, environmental sustainability, and corporate social responsibility, should therefore seek to understand the critical importance of issues of data privacy, transparency, and anthropomorphic developments. E-tourism researchers, and specifically those in AI, should also determine our collective responsibility to not merely chart the rapid deployment of AI systems in tourism, but lead scholarly discourse on the importance of maintaining ethics as part of the development of innovative research.

This paper takes a transformative approach [1, 2] to investigate the following research question: “to what extent has e-tourism research identified and treated ethics and bias issues within the topic area of artificial intelligence (AI)?” We employ a scoping study [3, 4], defined as “a form of knowledge synthesis that addresses an exploratory research question aimed at mapping key concepts, types of evidence, and gaps in research related to a defined area or field by systematically searching, selecting, and synthesizing existing knowledge” [5, p. 373] to systematically identify, chart, analyze and summarize identification and treatment of these issues in e-tourism research. In the following sections, we review literature and explain the methodological approach of conducting the scoping review. We then discuss results and close with implications and considerations for future work.

2 Literature

2.1 Artificial Intelligence

AI comprises training and input data, the algorithmic ‘rules’ by which data is processed and analyzed and output results [6]. AI is classified as limited or general. Much of the research in and around tourism has focused primarily on limited AI applications designed to complete a discrete and defined task [7]. Limited AI applications in tourism include big data analytics; smart devices in the Internet of Things (IoT); biometrics like speech and facial recognition; service robots; and blockchain technologies that enable services like personalized recommendations and smart chat bots; self-driving luggage carts in airports; smart check-in, venue access, and security; contactless food delivery or housekeeping services; and identity verification at border crossings [8, 9]. Li et al. [8] classify limited AI in tourism into four categories of AI service encounters: 1) AI-supplemented which includes use of real-time and historical data from searching, purchasing, and social media activity to make personalized recommendations, streamline and deliver contactless services; 2) AI-generated which includes use of biometric facial, speech, and movement recognition systems to facilitate self-check-in, smart tourism, and health monitoring; 3) AI-mediated which comprises service robots, virtual and augmented reality (VR and AR) to enhance virtual booking; and 4) AI-facilitated which includes customer experience (CX) and customer relationship management (CRM) [8].

General AI describes (semi-)autonomous computation that solves complex problems. General AI approaches in tourism include the use of machine learning (ML), neural networks, or deep learning to discern trends in tourist behaviors, perceptions, and preferences to forecast demand, streamline service delivery, and upsell and cross-sell to optimize revenue [10, 11]. Further developments in biometrics, emotion detection, and sentiment analysis signal a shift from smart tourism to “neurotourism,” experiences that are automatically responsive and customized to the unique preferences and even subconscious desires of individual travelers [10]. In their report on post-pandemic travel and tourism, McKinsey & Company declare that sentiment analysis and predictive analytics are necessary for prescriptive business models that maximize return-on-investment [12].

2.2 Ethics, Bias, and Artificial Intelligence (AI)

While there is significant debate in the computer science and technology fields on the need for ethical AI, ethical considerations, including bias, has seen comparatively less dialogue in smart tourism scholarship. In their systematic review, Xivuri and Twinomurinzi [6] determined that most research on AI fairness is not sector-specific (62% of studies examined), while 21% focused on public services sector (criminal justice, immigration, and government), 11% on the health sector, 2% on the financial sector, and 2% from the communications sector [6, p. 276]. In their 2021 systematic review of AI in tourism operations, Li et al. [8] acknowledge “the social and ethical issues of AI, such as ubiquitous surveillance, privacy, and equality, are important but not considered in the present study” [8, p. 8].

Ethical AI encompasses fairness, accountability, privacy, and autonomy. Fairness considers both individual attitudes toward AI outcomes, and the sociocultural context in which the system is deployed. “A fair AI system is one that aligns with shared human values and supports human flourishing” [14, p. 5]. Specific considerations include reciprocity and the ethics of care, beneficence and non-maleficence, fidelity and responsibility, integrity, equitable treatment, justice, and explicability [13, 14].

The introduction of automation poses a risk of economic dislocation and consideration for social responsibility within the tourism sector [8]. Saul and Etemad-Sajadi [11] observe that “seasonal, casual and some operational staff in the hospitality industry could be most impacted over time by the rise of artificial intelligence.” The Organization for Economic Cooperation and Development (OECD) prioritizes “inclusive growth, sustainable development, and wellbeing” along with “human-centered values and fairness” in its “Recommendation of the Council on Artificial Intelligence” [15]. The European Commission’s Ethics Guidelines for Trustworthy AI articulates a framework based on four ethical principles: respect for human autonomy, prevention of harm, fairness, and explicability. Prevention of harm includes awareness of “where AI systems can cause or exacerbate adverse impacts due to asymmetries of power or information,” while fairness describes conditions “free from unfair bias, discrimination, and stigmatization” [16, p. 12].As a component of ethical AI, algorithmic bias describes “systematically unfair outcomes that can arbitrarily put a particular individual or group at an advantage or disadvantage over another” [7, p. 2). AI bias is significant because it is felt on a wider scale than human bias, which tends to be more localized in its impact [17]; thus, “seemingly small error rates can still have a negative impact on a substantial number of individuals” [18]. Examples include the failure of facial recognition software to correctly identify Black and East Asian individuals as well as women and gender minorities; the disproportionate assignment of negative emotions to Black men in biometric sentiment analysis; discriminatory and exclusionary ad placements in search engines and social media platforms; bias in professional recruitment and candidate ranking in human resources; and racial bias in medical automation [7, 18, 19]. AI bias can arise from the training or input data, in which members of minoritized groups may be underrepresented (un-visible) or overrepresented (hypervisible) in the data to their disadvantage [14, 17], dynamics which Dancy & Saucier [20] characterize as “predatory inclusion” and “unwanted exposure” (2022), as well as from the design, development, features, processes, or outputs of the algorithmic model [7, 14]. Bias can also be introduced indirectly when input data is sufficiently correlated with a protected class or characteristic to act as a surrogate for that attribute, such as the Federal Trade Commission’s warning that use of postal codes to determine financial creditworthiness can result in illegal racial and ethnic discrimination [21].

Additional AI ethical considerations include data protection, privacy, autonomy, trust, safety, and artificial intimacy. Akter et al. [7] found that the exploitation of search, browse, and purchase history to shape consumer behavior and decision-making can contribute to trust declines and reputational damage. Additionally, dynamic pricing and price personalization can evolve into actual or perceived price discrimination [7]. The Public Voice [45] name data quality, public safety, cybersecurity, and prohibitions on secret profiling and unitary scoring (so-called “social credit scores”) among its “Universal Guidelines for Artificial Intelligence”, while OECD includes robustness, security, and safety in its “Recommendation of the Council on Artificial Intelligence” [15]. Artificial intimacy describes AI applications designed to mimic social interactions, which can leverage emotional states and perceived closeness to influence customer behavior [22]. For a human-centered enterprise like travel and tourism, with all the complexity and idiosyncrasy that implies, Strauß [23] warns that “AI has a transformative capacity where ‘natural’ aspects of society are at a risk of becoming reduced to machine-readable, datafied models that fit the logics of the artificial” (p. 4).

3 Methods

3.1 Data Scoping Steps

By undertaking a scoping study of published review papers, we methodologically associate this study with what Arksey & O’Malley [3] and Pham et al. [5] refer to as “a scoping review of scoping reviews”. While similar to systematic reviews, the aim of this study was narrower in nature. Specifically, we aimed to map the literature in tourism to better understand identification and treatment of ethics in AI research.

In response to the guiding research question, peer-reviewed articles which met the following criteria were selected for the study: published 1) in English, 2) in tourism-focused research journals, 3) between 2015 and 2021. Electronic database searches include CAB, ProQuest, and Business Source Premier (BSP). Search terms across all databases included “tourism” OR “hospitality” AND “artificial intelligence.” Initial and updated searches were conducted in tourism and business databases, journals, and conference proceedings resulting in 33,848 and 15,414 articles, respectively. Further filtering for “type = peer reviewed”, “language = English” and “time = 2015–2021” resulted in 2,030 articles.

Following Pham et al. [5], we conducted an initial title and abstract review of these articles. Next, we decided to refine the search strategy by applying the search criteria to the abstract field to increase the relevance of the corpus of literature that would be retained for the study. Two investigators independently reviewed abstracts and full titles to determine final inclusion. Discussions were held to address challenges and uncertainties related to articles to be included in the study. Review of these articles for relevance and accessibility led to the retention of 170 articles for manual screening. Two investigators manually screened these abstracts for relevance (i.e., review-type articles), duplicity, and accessibility. This was done until 100% agreement on the final pool of articles was achieved between the investigators. A pool of n = 27 articles which met all study criteria for relevance, publication period, language, accessibility, and review method was retained for charting analysis. Scoping identification, search, selection, and screening strategy steps are depicted in the first column and halfway down the middle column in Fig. 1 (see Supplementary Material).

3.2 Data Charting

Data charting was an iterative process which involved cross-charting between investigators to address discrepancies and divergencies in charting activities. Two investigators were responsible for charting the data to determine deficiencies in AI-focused tourism literature as it relates to empirically reflecting and treating underlying issues of ethics and bias. To determine this, investigators used thematic analysis to identify and document ethical and social bias implications of the AI technologies, techniques, or applications discussed in each article. This was followed by documenting the explicit discussion of ethics, and bias reduction specifically, in each article. Thus, the ethical and social bias implications documented in each article may be either explicit or inferred by us during charting analysis; while the treatment of ethics and bias was sought to be explicit in the article. The presence of an ethical or social bias implication that is not met with considered discussion elucidates the gap in ethical consideration which we seek to discover and document in this scoping review. Discussion and debriefs were held before and after the first round of charting to determine variable parameters which would constitute relevance to the research question. To ensure validity of the final sample, charting entries were cross-checked by two investigators if there was uncertainty or challenge during charting (e.g., a viewpoint paper did not review literature; identical but differing chronology of authors revealed duplication; and use of primary data related to customer experience with service robots). This process of cross-charting resulted in the deletion of five articles from the sample pool resulting in a final sample of n = 23 articles. Summary charting notes are provided in the Appendix (see Supplementary Material).

3.3 Data Overview

Descriptive characteristics for the pool of (n = 23) review articles are shown in Table 1 (see Supplementary Material). One (1) review article was published in 2017 and one (1) in 2018, three (3) in 2019, eight (8) in 2020, and seven (7) in 2021. Based on author descriptor, most papers employed a literature review methodology or review derivative (e.g., comprehensive research review, systematic review), bibliometric analysis or critical analysis. Other articles reviewed trade literature and/or industry trends, while others adopted industry use-case analyses. Also included in the pool of review articles were viewpoint or perspective papers. The papers reviewed research on service automation, big data and artificial intelligence (BDAI), service robots (SR) and service automation, and artificial intelligence-enabled internet of things (AI-enabled IoT). The articles spread across eleven tourism and hospitality-focused journals with first author institution affiliation spanning North America and The Caribbean, The United Kingdom, South and Southeast Asia, Western and Southeastern Europe, and the South Pacific.

4 Results and Discussion

From preliminary analysis of results of this scoping study on identification and treatment of ethics and bias issues in AI research in tourism and hospitality, we identify five categories across articles reviewed: 1) privacy and bias, 2) protection and transparency, 3) (de)humanization and sustainability, 4) inclusion and safety, and 5) policy and legal. These issues, their treatment in reviewed literature, and implications for research and practice are summarized below.

4.1 Privacy and Bias

Ethical issues related to guest privacy were mentioned in several papers, and include behavioral tracking of guests [24, 25], use of sensitive guest data, privacy literacy, intimate privacy [26], employee privacy, voice-recognition algorithms, and other “creepy” human surveillance which could be deemed to violate the sense of trust between guests and hosts. Bowen and Whalen [24] suggest an article, “Disclosing personal information via hotel apps: A privacy calculus perspective” for “further reading”. Hajal and Rowson [25] go further. They acknowledge that ethical concerns around the implementation of AI-driven technology “could be considered a threat to personal privacy and data rights” [25, p. 55], given the inaccuracy shown in recent studies, when it comes to identifying African American and Asian faces in comparison to Caucasian faces. Other examples include the ability of IoT to keep companies connected to driverless vehicles post-sale and hotels and restaurants’ ability to keep connected to guests beyond the initial stay or visit, for target marketing purposes. Research has found that behavioral tracking data may be used to enhance guest experience [24] and authors generally agree on AI as a source of social progress [25]. However, the use of sensitive guest data for targeted marketing raises issues of which data guests agree to disclose, and their control over when and how that data will be used. In keeping with guidelines for the respect for privacy and data governance [16], AI researchers should engage in applied research collaborative partnerships to address data privacy, trust of AI and implications for guest experiences. As well, we should support the establishment of normative practices for industry which promote sustainable target marketing, data privacy, protection and transparency.

4.2 Protection and Transparency

Protection and transparency involve aspects of artificial intimacy, data protection, ownership, consent [28], trustworthiness [29], and guest and employee protections. There is an urgent need for accountability and transparency in the secure use of sensitive biometric and psychographic data which feeds autonomous services and the IoT. Still, fulsome treatment of data protection and transparency issues have been relegated to futuristic possibilities in published research, leaving data ownership and safekeeping of sensitive data under-research. While Samara et al. [30] acknowledge the EU’s recently introduced General Data Protection Regulation and the impact it is expected to have on the way BDAI is conducted, McCartney and McCartney [28] highlight several deficiencies. They acknowledge neglect of the topic of data protection and protocols; the need for the hospitality and tourism industry to consider legislative oversight for data protection; and weak protections for humans in the face of continued SR integration. They call for an SR hospitality research agenda to address emerging risks and security concerns to include data protection and protocols. Cobanoglu and Demicco [32] found that hotels bypass critical cybersecurity protections for computers and software, thereby exposing guest and employee data to cyber-related risk.

4.3 (De)humanization and Sustainability

(De)humanization and sustainability issues relate to employee protections and include socio-economic dislocation [33, 34] and environmental displacement [10]. The underlying value of AI rests in its anthropomorphic capabilities to perform humanlike tasks based on algorithmic programming. Several researchers highlight risks facing human workers whose talents are being supplemented by the very AI systems replicated from human design (e.g., Chipotle’s Chippy). While AI researchers agree on the benefits to guest experiences, these models can lead to dehumanization [25, 36] of the workforce arising from emotional exhaustion among service employees [37] stemming from use of SR employees with whom absenteeism, workplace conflict, and performance-based compensation issues do not arise; and who are able to perform at higher levels of efficiency. The result is social and economic dislocation due to replacement of low-skill and low-wage positions [35]. Environmental displacement concerns have also been raised based on the use of AR and VR technologies to facilitate immersive experiences for guests. On the one hand, environmental and socio-cultural sustainability can be enhanced [10]. On the other hand, cultural and heritage erosion, human-nature (dis)engagement, and unsustainable tourism communities emerge and should be researched.

4.4 Inclusion and Safety

Ethical issues of social inclusion and safety relate to linguistic and human bias towards anthropomorphic robots [28], algorithmic bias [30], gender bias [38], and biases related to e-human resource management [24]. For example, algorithmic bias related to AI linguistics [39, 40] can serve to exclude traditionally marginalized groups such as Black, Indigenous, and People of Color (BIPOC). AI recommender and information systems for example, are programmed mostly by non-BIPOC groups who unknowingly create algorithms reflective of social biases and inequities in society. Bhushan [31] encourages diversity in AI development teams as a means of controlling this form of bias. This is especially important given that much AI development has occurred in westernized, “white male” contexts, privileging this group over comparatively underserved gendered and minoritized groups when it comes to use of AI systems for voice- and face-recognition. Methods research dedicated to the measurement and management of diversity, equity, and inclusion in AI-use in tourism should address accessibility across stakeholder groups and accessibility across spatial-temporal dimensions. This would mean that fair and inclusive AI access form parts of key destination performance measures.

4.5 Policy and Legal

Policy and legal issues include capabilities and responsibilities of service robots, issues of risk and compliance [28], and rights of both SRs and humans interacting in the tourism space. Issues of digital and material inequalities in guest interactions with SRs and sexbots [35] suggest need for laws to adapt and protect both humans and robots, as well as to protect human sex workers from their algorithmic competition. McCartney and McCartney [28] identify the need to address ethical issues of trust and emotional attachment between SRs and children in their care. Also relevant are issues of power, discrimination, equality and justice between SRs and employees whose positions and roles become intelligently automated [39]. Li et al. [42] acknowledge social and ethical issues of AI, such as ubiquitous surveillance, privacy, and equality as important considerations for future research. Gaur et al. [40] state that it is essential to study the ethics involved in adopting AI and robotics in the hospitality industry, while Lv et al. [43] call for greater ethical care in the use of big data artificial intelligence (BDAI) in analytics and forecasting methods. These authors acknowledging that while existing research mainly focuses on the bright sides that big data brings into hospitality and tourism, the dark sides remain largely unexplored and could be investigated by future research. In response, Cain et al. [35] encourage the need for regulatory and legal frameworks in AI. These positions invariably reflect the overall neglected treatment of policy and legal issues in AI studies.

5 Conclusions and Future Work

The hospitality and tourism sector cannot afford to ignore ethical issues in AI. As Akter et al. [7] point out, there is convincing consensus among scholars that the future source of competitive advantage of a firm is dependent on the extent to which it can safely and securely deploy bias-free AI solutions to deliver real-time decisions and solve critical business problems. Tourism managers and policy influencers who ignore or fail to effectively incorporate AI ethics in the deployment of innovative technologies risk unfavorable scrutiny from a global community now keenly interested in safe and secure AI. Experienced and emerging AI researchers are encouraged to forge ahead of the ethics curve and lead critical discourse on the ethics of AI-use in tourism, even as we continue to expand cutting-edge AI research.

This study provides support for the shared responsibility we have as tourism researchers and practitioners to ensure that AI research and use are as ethically encompassing as they are novel. By placing attention on the occurrence and treatment of AI ethics issues in tourism scholarship, this study calls on e-tourism researchers to be more deliberate in addressing AI ethics in scholarly work; on conference organizers and journal editors to create meaningful spaces for AI ethics discourse; and on industry stakeholders to actively engage strategies to mitigate impact of unethical AI practices in designing service experiences. Furthermore, scanning the journal titles from our set of twenty identified articles suggests that the conversation about AI, ethics, and tourism appears to be occurring in our niche journals, rather than in the mainstream of our literature.

Limitations of this scoping study create opportunities for future work. For example, use of additional databases may have revealed other relevant articles. The dramatic level of shrinkage in the number of articles, from thousands down to twenty, suggests the need to further refine search terms, since the terms that were used were identifying both those articles relevant to this study’s goals as well as many articles that mentioned AI and AI-related terms in passing. Finally, investigating treatment of ethics within specific application areas in tourism and hospitality (e.g., big data artificial intelligence, service robots) presents another future research direction to determine whether AI scholarship in tourism tracks with other service research fields.