Page 163 - ISC PROCEEDINGS 21.4

can lead to negative evaluations, increased perceptions of risk, and resistance behaviors.
Previous studies have applied the Uncanny Valley Theory in the context of
interactions with AI. Wu et al. (2024) found that minor deviations in the movements of AI
news presenters can create a sense of eeriness and weaken emotional connection with
audiences. Abbate et al. (2025) reported that highly human-like AI avatars may reduce
user acceptance when small imperfections trigger the uncanny valley effect. Gong and
Zhang (2025) showed that highly human-like virtual agents can increase users’ discomfort
and resistance. Similarly, Kim et al. (2025) indicated that anthropomorphic AI in
personalized advertising may increase perceptions of privacy risk and lead to user
resistance.
From the perspective of Psychological Reactance Theory, prior research suggests
that high levels of AI autonomy may cause users to feel that their freedom is threatened,
thereby increasing psychological reactance (Oh et al., 2025). Hong et al. (2025) found that
excessive algorithmic influence may lead users to feel manipulated and reduce their
engagement. Feng et al. (2019) also demonstrated that the mandatory adoption of
technology can trigger user resistance. Furthermore, Kim and Wang (2025) showed that
anthropomorphic characteristics and emotional cues of AI can either reduce or increase
psychological reactance depending on how users perceive the interaction process.
2.2. Hypothesis development
In the context of live streaming, perceived anthropomorphism refers to the extent
to which viewers attribute human-like characteristics, emotions, and intentions to a
virtual streamer, treating it as a social entity rather than a mere technological tool. The
positive impact of perceived anthropomorphism on both privacy disclosure risk and
intrusiveness risk is rationalized by a synergistic mechanism between the Uncanny Valley
effect and Psychological Reactance Theory.
According to the Uncanny Valley framework, as virtual streamers achieve high levels
of anthropomorphism without attaining perfect human realism, they engender a state of
ontological ambiguity (Ahn et al., 2022). This ambiguity disrupts the viewer’s ability to
categorize the agent, creating a dissonance where the streamer appears disturbingly
intelligent yet lacks genuine consciousness. Consequently, human-like cues such as a
direct gaze or empathetic responses are not perceived as benevolent social signals, but
rather as forms of deceptive surveillance by an opaque entity (Kim et al., 2024). This
perception of deceptive monitoring directly triggers the defensive mechanisms described
by Psychological Reactance Theory. PRT posits that individuals experience a state of
motivational arousal when they perceive a threat to their behavioral freedom or
autonomy. In this case, the "uncanny" social presence of the streamer imposes a
psychological burden of being watched, which viewers interpret as a threat to their
anonymity and control over the interaction (Aw et al., 2024). Because the observer is an
ambiguous, quasi-human entity, the perceived threat is amplified, compelling users to
react defensively to restore their personal boundaries. This reactance manifests in two
distinct risk perceptions. First, users classify the persistent interaction as a violation of
their private space (Pizzi et al., 2023), leading to intrusiveness risk, defined as a
psychological barrier arising when individuals perceive an unwanted encroachment on
their personal space, time, or cognitive resources (Jiang et al., 2013). Second, fearing that
this unpredictable entity harbors hidden motives for data collection, users perceive a
heightened privacy disclosure risk, which is the threat of personal data being exposed,
misused, or accessed without authorization (Shao & Ho, 2025; Liu et al., 2025).

162