Decomposing Autonomy: Explaining AI Technology Acceptance Through a Liberty-Based Framework

Paper Title

Decomposing Autonomy: Explaining AI Technology Acceptance Through a Liberty-Based Framework

Publication Info

• Topic area: Human autonomy in AI technology acceptance and interaction
• Keywords: Human autonomy, positive liberty, negative liberty, agency, AI acceptance, Human-AI interaction, autonomy framework, liberty duality, system design, user experience

Background and Problem

• Problem / challenge: Autonomy is often vaguely defined in Human-Computer Interaction (HCI) research and conflated with related concepts such as agency, control, and freedom. This lack of clarity hinders the development of actionable design principles and empirical evaluations.
• Significance: Understanding autonomy is critical for designing AI systems that align with human values, support user goals, and are widely accepted. Autonomy influences motivation, well-being, and the willingness to adopt AI technologies.
• Motivation and related work: Prior work has highlighted autonomy as a core psychological need and a guiding principle in AI design. However, autonomy is inconsistently measured, and its subcomponents are rarely disentangled. This paper builds on political philosophy and psychological theories to propose a structured framework for autonomy in Human-AI Interaction (HAI).

Solution

• Proposed approach: The paper introduces a Conceptual Map of Autonomy, which decomposes autonomy into Positive Liberty (freedom to pursue authentic goals), Negative Liberty (freedom from external constraints), and Agency (action ownership).
• Novelty:
  1. Development of a dual-liberty framework (Positive and Negative Liberty) as measurable constructs for autonomy in HAI.
  2. Empirical validation of the autonomy framework through a video vignette study with 194 participants.
  3. Identification of distinct roles of Positive Liberty, Negative Liberty, and Agency in predicting AI system acceptance and user experience.
• Procedure and key techniques:
  • Conceptualization of autonomy based on political philosophy and HCI research.
  • Design of video vignettes illustrating AI-mediated communication scenarios with varying levels of liberty.
  • Collection of participant responses using validated scales for Positive Liberty, Negative Liberty, and Sense of Agency.
  • Statistical analysis, including exploratory and confirmatory factor analyses, regression models, and structural equation modeling (SEM).

Results

• Concrete findings:
  • Positive Liberty (PL) and Negative Liberty (NL) are highly correlated but distinct constructs (PL: α = .92, NL: α = .94).
  • PL strongly predicts intention to use AI systems (β = .668, p < .001), while NL shows a small negative effect (β = −.106, p = .048).
  • NL positively predicts the Sense of Agency (β = .549, p < .001), but Agency does not significantly influence intention to use AI systems.
  • Participants rated AI systems higher for self-use (M = 4.70) than for use by others (M = 4.35, p < .001).
• Advantage over baselines:
  • The framework clarifies autonomy’s subcomponents, enabling more precise measurement and actionable design insights compared to prior monolithic or vague definitions.
  • Empirical evidence supports the theoretical distinction between liberty dimensions and their unique impacts on AI acceptance.
• Experiments / evaluation:
  • A 2x3 video vignette study with 194 participants, featuring two AI scenarios (Summarisation and Notification Management) and three liberty levels (LOW, MIDDLE, HIGH).
  • Measurements included validated scales (e.g., UTAUT2, Sense of Agency Scale) and newly developed constructs for PL and NL.
  • Statistical analyses confirmed the reliability of constructs and tested hypotheses about their relationships and effects.
• Limitations and future work:
  • Vignette design did not fully reflect the intended PL- and NL-dominated scenarios, limiting level contrasts.
  • Results are specific to AI-mediated communication scenarios with productivity goals; generalization to other domains requires further study.
  • Constructs measure subjective perceptions rather than objective conditions, necessitating additional work to translate findings into actionable design guidelines.
  • Future research should explore long-term impacts of autonomy-supportive AI and examine the formation of user values and preferences.

Summary

This paper proposes a novel autonomy framework for Human-AI interaction, decomposing autonomy into Positive Liberty, Negative Liberty, and Agency. Empirical validation through a video vignette study demonstrates that Positive Liberty strongly predicts AI acceptance, while Negative Liberty primarily influences the Sense of Agency. The findings highlight the importance of designing AI systems that align with user goals and values, rather than simply maximizing freedom or control. The framework offers actionable insights for responsible AI design and regulation, with potential applications across diverse AI domains.