Commercial AI chatbots and mental health self-care: An ethnography of contemporary AI companionship

这个博士项目探索了公共媒体和用户如何看待商业化的人工智能同伴机器人，这些机器人承诺为精神健康问题提供解决方案。同伴机器人的出现要求对人们与这些技术制品进行交流的关系动力学进行调查(Guzman&Lewis，2019)，特别是在亲密的空间，如心理健康自我护理中的陪伴。这些设备是医学日益商业化和“生活化”的一部分(Saukko等人，2010年)，或作为“社会文化艺术品”，潜在地“对医疗权威、医疗保健和公共卫生工作的提供以及”医患关系“产生重大影响”(Lupton，2014)。通过研究媒体驱动的关于商业营销的人工智能同伴和用户体验的话语，这项研究探索了有争议的商业自我护理领域与人工智能的社交或治疗互动能力以及它们的社会和伦理影响之间的交集(Natale，2019)。该项目的首要研究问题是：RQ，营销为精神健康问题解决方案的商业可获得的人工智能同伴机器人如何或在多大程度上被媒体及其用户视为同伴？这个首要问题转化为三个具体的研究目标：RO1。分析公共媒体关于商业人工智能伴侣机器人对心理健康RO2的关键论述。考察用户如何看待市面上可买到的AI同伴机器人R03。为了分析用户如何使用人工智能同伴机器人并与之互动，以及用户共同创建了什么样的社交角色(朋友、治疗师、助手、游戏)，该项目将专注于2017年推出的大规模营销的人工智能同伴机器人Replika。Replika是迄今为止最成功的人工智能同伴机器人之一，注册人数为250万(Olson，2018)，在Facebook的“Replika Friends”群组上有3.1万活跃成员。这项研究将从定性解释的角度进行。为了实现RO1，我将使用多模态批评话语分析(Machin，2013)来分析媒体对商业伙伴机器人Woebot、Wysa和Replika的讨论，以识别关于这些技术的公共叙事和辩论。我将分析从主要大报和小报上抽样的大约50篇英国和美国在线新闻文章，重点是2017年(机器人推出时)至2021年期间的报道高峰。为了解决RO2问题，我将对大约30名(或直到理论上达到饱和)Replika用户进行深入的半结构化定性采访，询问他们对机器人的看法。为了实现R03，我将把对采访结果的分析与另外两种数据结合起来：用户与他们的“复制品”之间互动的截图分析(应用户要求)和Facebook群“Replika Friends”的六个月数字人种学，在该群组中，用户经常就他们与Replika的关系进行辩论。数字人种志是一种适用于分析在线社区的方法(Pink等人，2013年)。与机器人的访谈和互动以及来自在线社区的数据将首先使用多模式主题分析进行分析(Braun&Clarke，2006)。鉴于政府、医疗服务机构和公众对心理健康应用程序表现出的热情(Hollis，2015)，以及对交流型人工智能(Guzman&Lewis，2019)的新兴趣，该项目特别及时，将确定使用AI同伴和类似AI技术的潜在伦理和社会影响。此外，它还调查了当此类心理健康应用程序仍有争议时，个人如何体验商业人工智能同伴。

This PhD project explores how public media and users perceive commercially available AI companion bots, which promise to provide solutions to mental health problems. The advent of companion bots calls for the investigation of the relational dynamics through which people communicate with these technological artefacts (Guzman & Lewis, 2019), especially within spaces which are intimate such as companionship within mental health self-care. These devices form part of the increasing commercialization and 'lifestylisation' of medicine (Saukko et al., 2010) or as "sociocultural artefacts" with potentially "significant implications for medical authority, the delivery of healthcare and public health work and the "doctor-patient relationship" (Lupton, 2014). By examining the media-driven discourses on commercially marketed AI companions and user experiences, the study explores the intersection between the contentious areas of commercial self-care and AI's ability for social or therapeutic interaction and their social and ethical implications (Natale, 2019).The overarching research question of the project is:RQ How, or to what extent, commercially available AI companion bots marketed as solutions to mental health issues are perceived as companions by the media and their users? This overarching question translates into three specific research objectives:RO1. To analyse key public media discourses on commercial AI companion bots for mental healthRO2. To examine how users perceive commercially available AI companion bots RO3. To analyse how users employ and interact with AI companion bots and what kinds of social roles users co-create (friend, therapist, assistant, game)The project will focus on the mass-marketed AI companion bot 'Replika', launched in 2017. Replika is one of the most successful AI companion bots up to date with 2.5 million sign-ups (Olson, 2018) and an active community of 31,000 members on the Facebook group "Replika Friends". The study will be conducted from a qualitative interpretative perspective. To achieve RO1, I will analyse media discussion on the commercial companion bots Woebot, Wysa and Replika, using multimodal critical discourse analysis (Machin, 2013) to discern public narratives and debates about these technologies. I will analyse approximately 50 UK and US online news articles sampled from major broadsheet and tabloid newspapers, focusing on peaks in coverage between 2017 (when the bots were launched) and 2021. To address RO2 I will conduct in-depth semi-structured qualitative interviews with approximately 30 (or until theoretical saturation is reached) Replika users on how they perceive the bots. To achieve RO3 I will combine the analysis of interview findings with two additional kinds of data: the analysis of screenshots of interaction between users and their 'replikas' (requested from users) and a six-month digital ethnography of the Facebook group "Replika Friends", where users frequently debate their relationship with Replika. Digital ethnography is a method suitable for analyzing online communities (Pink et al, 2013). The interviews and interaction with the bot as a well as the data from the online community will be in the first instance, analysed using multimodal thematic analysis (Braun & Clarke, 2006). Given the enthusiasm for mental health apps shown by governments, health services, and the public (Hollis, 2015) combined with the renewed interest in communicative AI (Guzman & Lewis, 2019) the project is particularly timely and will identify potential ethical and social implications of the use of AI companions and similar AI technologies. Furthermore, it investigates how individuals experience commercial AI companions, when such mental health apps remain controversial.