Since its birth in the 1950s, AI has shifted from a research topic to a pervasive technology shaping everyday life: it accelerates medical diagnoses, makes mobility safer and more accessible, optimises industrial production, and augments human decision-making across nearly every domain. The same technology, however, carries equally substantial risks such as biased and discriminatory outcomes, loss of privacy and autonomy, opacity of decision-making, and a diffusion of accountability. The way societies and individuals meet AI systems is rarely a matter of technical functionality alone. Beneficial AI is sometimes greeted with deep scepticism and ultimately rejected, whilst harmful systems are at times adopted uncritically before their consequences come to light. Both patterns are shaped less by what the technology can do than by how it is perceived, evaluated, and ultimately trusted. The ability to realise AI’s benefits whilst containing its risks therefore depends on the human side of the equation, where public perception, evaluation, and trust become decisive determinants of whether beneficial technologies reach the people they could serve. Despite these stakes, only a small fraction of AI research engages with a social-science perspective, and the work typically treats AI as a single, monolithic technology. Yet AI is not encountered as a single technology but as a specific system performing a specific task in a specific domain—leaving context, arguably the most decisive variable for how AI is perceived and evaluated, rarely placed at the centre of enquiry.
In response to this gap, the thesis offers a context-sensitive account of public perception and evaluation of AI, anchored in the question of under which conditions people would trust an AI system to perform a specific task. To answer this question, this work is guided by five interlinked goals: establishing a baseline of perception, deepening acceptance and trust models, identifying psychological drivers beyond socio-demographics, exploring context-sensitivity itself, and clarifying the relationship between trust and acceptance. To address them, four empirical studies follow a progressive and mixed-method research approach—from general to specific, exploratory to confirmatory, observation to implications. More specifically, this thesis draws on a series of data collection and analysis methods that are layered across two foundational studies and two subsequent context-sensitive deep dives into AI in public transportation and AI in health care. Core psychometric and attitudinal constructs, including general attitude towards artificial intelligence (GAAIS), affinity for technology (ATI), belief in a dangerous world (BDW), safety perception, as well as knowledge of and experience with AI, are examined across the studies as potential predictors of AI acceptance. Response variables are kept consistent across all studies and operationalised as use intention and trust. All four samples were collected prior to the public release of generative AI and the launch of ChatGPT in November 2022, positioning this thesis as a historical baseline taken immediately before the technology was introduced to the general public. The foundational studies combine a convenience sample of early adopters (N = 137) and eight semi-structured interviews with a panel representative of the German population on age, gender, and education (N = 918) accompanied by five expert interviews. The two context-sensitive studies consisted of a convenience sample drawn from rural areas for public transportation (N = 124) and an early-adopter sample for health care (N = 348) recruited in the midst of the COVID-19 pandemic.
The findings of the empirical studies are multifaceted and interlinked, revealing the complex landscape of AI perception. The public meets AI not with dystopian fears, but with cautious optimism, framing the technology as a functional tool rather than a social agent; well-defined concerns about misuse, opacity, and missing moral reasoning coexist with a clear appreciation of utility and perceived benefits. This duality is not universal across all contexts. Acceptance patterns shift substantially between domains and even within them: safety concerns dominate autonomous public transport, whilst health anxiety and belief in a dangerous world take over in the health care setting. Personality traits that lie dormant in general perception become active when the stakes shift. Across all four contexts, however, dispositional attitudes carry over: general attitude towards artificial intelligence and affinity for technology remain the most stable predictors, with general trust in AI emerging as the single most influential factor for context-specific trust in the health care context. Demographic variables, by contrast, prove unreliable. Overall, a consistent finding surfaces across every study: use intention exceeds trust, a pragmatic acceptance in which adoption outruns confidence in the technology. Trust nevertheless anchors the picture of acceptance patterns, mediating between concerns and adoption across contexts and emerging as the central pillar around which human, situational, and system-level factors revolve.
A series of contributions arises from the empirical research journey. Theoretically, trust emerges as a central pillar of AI acceptance, mediating the effects of risk perception, safety perception, and worldviews on use intention. The persistent gap between use intention and trust further marks the two as distinct constructs that warrant separate measurement and theorising. Methodologically, the progressive multi-method design offers a transferable blueprint for moving from exploratory to confirmatory analysis across diverse AI application contexts. Practically, the work translates into actionable implications for AI developers and designers, policymakers and regulators, educators, ethics committees, and researchers. The implications converge on the principles of trust by design, domain-specific regulation, promotion of AI literacy, and a communication culture respecting the duality of AI perception, addressing benefits and risks equally. Taken together, this thesis establishes context as a core determinant of AI acceptance and offers a snapshot at a pivot point—the moment immediately before generative AI redrew the public’s mental model of the technology—a baseline against which future research can measure how perception, evaluation, and trust shift in the decade ahead.