Mimetic Alignment with ASPECT: Evaluation of AI-inferred Personal Profiles
Summary: arXiv:2603.26922v1 Announce Type: cross
Abstract
AI agents that communicate on behalf of individuals need to capture how each person actually communicates. However, current approaches either require costly per-person fine-tuning, produce generic outputs from shallow persona descriptions, or optimize preferences without modeling communication style.
We present ASPECT (Automated Social Psychometric Evaluation of Communication Traits), a pipeline that directs LLMs (Large Language Models) to assess constructs from a validated communication scale against behavioral evidence from workplace data, without per-person training.
Introduction
The increasing reliance on AI agents to represent individuals in professional settings has raised concerns regarding the accuracy and authenticity of these representations. Traditional methods of developing AI communication profiles often necessitate extensive customization, which can be both expensive and time-consuming. Furthermore, these methods frequently lead to overly generalized communication styles that fail to reflect the unique traits of the individual.
ASPECT Overview
ASPECT offers a novel solution to these challenges by leveraging existing workplace data to create personalized communication profiles. Unlike conventional approaches, ASPECT does not require per-person fine-tuning. Instead, it utilizes validated scales to analyze communication traits, allowing for a more nuanced understanding of how individuals communicate in different scenarios.
Methodology
In a case study involving 20 participants, ASPECT generated profiles based on 1,840 paired item ratings and 600 scenario evaluations. The methodology focused on aligning AI-generated profiles with self-assessments, exploring the effectiveness of ASPECT-generated responses compared to generic and self-report baselines.
Findings
The results indicated that ASPECT-generated profiles achieved moderate alignment with participants’ self-assessments. Additionally, participants expressed a preference for ASPECT-generated responses over both generic outputs and self-report baselines. Notably, responses varied significantly across individuals and scenarios, highlighting the need for individualized assessments.
Profile Review and Recalibration
During the profile review phase, participants were able to examine linked evidence that supported the AI-generated assessments. This feature enabled them to identify any mischaracterizations in their profiles, recalibrate their self-ratings, and negotiate context-appropriate representations that better reflected their communication styles.
Implications for AI Communication Agents
The implications of this research are profound, particularly in the context of building inspectable and individually scoped communication profiles. ASPECT allows individuals to maintain control over how AI agents represent them in professional environments. This capability not only enhances the accuracy of AI communications but also fosters a sense of agency among users.
Conclusion
ASPECT represents a significant advancement in the development of AI communication agents. By providing a cost-effective and efficient means of generating personalized communication profiles, ASPECT has the potential to transform how AI agents interact on behalf of individuals, ensuring that these interactions are both authentic and contextually appropriate.
Future Directions
Further research is necessary to refine the ASPECT framework and explore its applications across diverse professional settings. As AI continues to evolve, tools like ASPECT will play a critical role in enhancing the quality of AI-mediated communications.