[Paper Review] From Fixed to Flexible: Shaping AI Personality in Context-Sensitive Interaction

Shakyani Jayasiriwardene, Hongyu Zhou+7 more|arXiv (Cornell University)|Jan 13, 2026

Social Robot Interaction and HRI0 citations

TL;DR

The paper presents a real-time, user-driven system that lets users adjust an AI chatbot’s personality across eight dimensions in three task contexts, and analyzes how expectations and trajectories of personalization emerge.

ABSTRACT

Conversational agents are increasingly expected to adapt across contexts and evolve their personalities through interactions, yet most remain static once configured. We present an exploratory study of how user expectations form and evolve when agent personality is made dynamically adjustable. To investigate this, we designed a prototype conversational interface that enabled users to adjust an agent's personality along eight research-grounded dimensions across three task contexts: informational, emotional, and appraisal. We conducted an online mixed-methods study with 60 participants, employing latent profile analysis to characterize personality classes and trajectory analysis to trace evolving patterns of personality adjustment. These approaches revealed distinct personality profiles at initial and final configuration stages, and adjustment trajectories, shaped by context-sensitivity. Participants also valued the autonomy, perceived the agent as more anthropomorphic, and reported greater trust. Our findings highlight the importance of designing conversational agents that adapt alongside their users, advancing more responsive and human-centred AI.

Motivation & Objective

• Investigate how users form expectations of conversational AI personalities under different task conditions.
• Examine how user expectations evolve during context-sensitive interactions with dynamically adjustable personalities.
• Assess user perceptions of interfaces enabling real-time personality adjustment.
• Characterize baseline personality preferences and adjustment patterns across contexts.
• Understand the role of affective anthropomorphism in trust and interaction quality.

Proposed method

• Developed a GPT-4.1 powered chatbot with real-time personality tunability across eight dimensions: Decency, Profoundness, Instability, Vibrancy, Engagement, Neuroticism, Serviceability, Subservience.
• Implemented a five-point slider for each dimension and allowed one dimension to change per conversational turn to isolate effects.
• Used a structured system prompt to translate slider values into chatbot responses.
• Employed three task contexts (Informational, Emotional, Appraisal) based on Cutrona and Suhr (1992) with GPT-4.1 to generate scenarios.
• Collected online mixed-methods data from 60 participants including self-reports and interaction logs.
• Applied Latent Profile Analysis (LPA) to identify personality configuration classes per condition and stage, and Sankey diagrams to visualize transitions.
• Conducted trajectory analysis using k-means clustering to identify adjustment strategies (Steady Anchors, Adaptive Explorers, Reactive Shifters).
• Analyzed trait importance via post-task surveys and produced heat-maps of net changes across conditions.

Figure 1 . The main conversational interface. (A) Slider panel for configuring and fine-tuning the agent’s personality across eight dimensions. (B) Conversational area for task-based interaction with the agent. (C) Information panel providing step-by-step instructions and guidance for using the inte

Experimental results

Research questions

• RQ1 RQ1: How do users form personality expectations of conversational AI based on different task conditions?
• RQ2 RQ2: How do user expectations of conversational AI personality dynamically evolve during context-sensitive conversations?
• RQ3 RQ3: How do users perceive conversational AI interfaces with dynamically adjustable personality traits?

Key findings

• User expectations vary by context, aligning with different social roles (e.g., helper, guide, assistant).
• Engagement and Serviceability are baseline preferences across conditions; Decency is particularly salient in appraisal contexts.
• Across contexts, participants converge on common traits (Engagement, Serviceability) and express a need for fine-tunable dimensions.
• Three adjustment trajectories are observed: Steady Anchors, Adaptive Explorers, and Reactive Shifters, with context-dependent prevalence.
• Trait adjustments show that Engagement and Decency change least, while Neuroticism, Subservience, and Instability exhibit more movement.
• Affective anthropomorphism contributes to perceived trust alongside competence in adaptive interfaces.

Figure 2 . Overview of the user study procedure, from consent and pre-task measures through task-based chatbot interactions, post-task surveys, and post-study questionnaire.