<\/span><\/h3>\nHow artificial intelligence (AI) is framed in news articles is significant as framing influences society\u2019s perception and reception of this emerging technology. Journalists may depict AI as a tool that merely assists individuals to perform a variety of tasks or as a (humanoid) agent that is self-aware, capable of both independent thought and creativity. The latter type of representation may be harmful, since anthropomorphism of AI not only generates unrealistic expectations of this technology, but also instils fears about technological singularity, a hypothetical future in which technological growth becomes unmanageable. To determine how and to what extent the media in South Africa anthropomorphise AI, we employed framing theory to conduct a qualitative content analysis of articles on AI published in four South African online newspapers. We distinguished between social anthropomorphism, which frames AI in terms of exhibiting a human form and\/or human-like qualities and cognitive anthropomorphism, which refers to the tendency to conflate human and machine intelligence. Most articles reflected social anthropomorphism of AI, while a few framed it only in terms of cognitive anthropomorphism. Several reflected both types of anthropomorphism. Based on the findings, we concluded that anthropomorphism of AI may hinder the conceptualisation of the epistemological and ethical consequences inherent in this technology.<\/p>\n
<\/p>\n
Keywords:<\/strong> Artificial intelligence; Framing theory; News articles; Cognitive anthropomorphism; Social anthropomorphism<\/p>\n<\/span>1. Introduction<\/span><\/h3>\n<\/span>1.1 Social and cognitive anthropomorphism of artificial intelligence<\/span><\/h4>\nAnthropomorphism describes our inclination to attribute human-like shapes, emotions, mental states and behaviours to inanimate objects\/animals, and depends neither on the physical features nor on the ontological status of those objects\/animals (Giger, Pi\u00e7arra, Alves\u2010Oliveira, Oliveira and Arriaga, 2019:89). Artificial intelligence (AI), including computers, drones and chatbots, can be anthropomorphised regardless of material dissimilarities between these technologies and humans, and despite the absence of an evolutionary relationship (Giger et al.<\/em>, 2019:89). In this study, we examine both social and cognitive anthropomorphism of AI, where the former process ascribes human traits and forms to AI (cf. Giger et al.<\/em>, 2019:112), and the latter designates the expectation that AI mimics human intelligence (Mueller, 2020:12). We base this distinction on our datasets, which reflect a focus either on the anthropomorphic form and\/or human-like qualities of AI, especially when framing human-robot interaction, or on cognitive processes when describing the intelligence of machine learning and deep learning, for instance. Several articles reflect both cognitive and social anthropomorphism (see Section 4).<\/p>\nCognitive anthropomorphism saturates news coverage of both weak and strong AI; that is, when AI is framed as merely simulating human thinking or as matching human intelligence (Bartneck, 2013; Damiano and Dumouchel, 2018:5; Salles, Evers and Farisco, 2020). The penchant to conflate artificial and human intelligence is unsurprising, historically speaking. Watson (2019:417) traces the practice especially to Alan Turing\u2019s (1950) eponymous Turing Test for determining whether a machine can \u2018think\u2019. Since then, technology experts and laypeople have framed AI in epistemological terms, constructing it as capable of thinking, learning, and discerning. Human intelligence is notoriously resistant to easy definition, and AI might be more challenging still (Kaplan and Haenlein, 2020). Consequently, when humans envision AI, human intelligence offers a ready touchstone (cf. Cave, Craig, Dihal, Dillon, Montgomery, Singler and Taylor, 2018:8; Kaplan and Haenlein, 2019:17). In part, the propensity to anthropomorphise AI in cognitive and\/or social terms derives from speculative fiction (Salles et al<\/em>., 2020:91). However, many AI researchers also employ anthropomorphic descriptions (Salles et al<\/em>., 2020:91). Salles et al.<\/em> (2020:91) suggest that the practice is driven by \u201ca veritable inflation of anthropocentric mental terms that are applied even to non-living, artificial entities\u201d or to \u201can intrinsic epistemic limitation\/bias\u201d on the part of AI scholars. They also speculate that anthropomorphism could stem from the human need to both understand and control AI in order to experience competence (Salles et al<\/em>., 2020:91). We propose that journalists too are motivated to anthropomorphise AI to understand and control it, particularly because it is an emerging and therefore uncertain science: \u201cpeople are more likely to anthropomorphize when they want to [\u2026] understand their somewhat unpredictable environment\u201d (Salles et al<\/em>., 2020:89-90).<\/p>\nWhen news media anthropomorphise AI, one epistemological consequence is the risk of exposing the public to exaggerated or erroneous claims (cf. Proudfoot, 2011; Samuel, 2019; Watson, 2019). To appraise this risk, we used framing theory to conduct a content analysis of anthropomorphism in articles published in four South Africa newspapers, namely, the Citizen<\/em>, the Daily Maverick<\/em>, the Mail & Guardian Online<\/em>, and the Sowetan LIVE<\/em>. We addressed the following research questions:<\/p>\nResearch question 1:\u00a0\u00a0 <\/em>What were the most salient topics in the coverage of AI?<\/p>\nResearch question 2:<\/em>\u00a0 How was AI anthropomorphised?<\/p>\nOur analysis does not intend to, in Salles et al.<\/em>\u2019s (2020: 93) words, defend \u201cmoral human exceptionalism\u201d. Instead, we are interested, from an ontological vantage, in AI-human differences. Moreover, we wish to interrogate the potential epistemological and ethical impacts of anthropomorphic framing of AI for public consumption. Therefore, we question how news media constitute and interrelate \u2018humans\u2019 and \u2018machines\u2019. This undertaking is important. How we anthropomorphise AI compels us to re-evaluate how we conceptualise human and artificial intelligences (cf. Curran, Sun and Hong, 2019). For the second research question, we honed our interest on the nature of anthropomorphic framing in South African news articles, instead of unpacking why<\/em> coverage might differ across the outlets. Providing a methodical and nuanced account of inter-outlet variability exceeds the purview of this study; however, in future, we plan to question this variability by attending, among other things, to the agenda of each outlet, its target audience, and gatekeeping by editorial boards.<\/p>\n