I, Robot? Socio-Technical Implications of Ultra-Personalized AI-Powered AAC; an Autoethnographic Account

2025-09-16

AI Summary

This study addresses the challenge that generic AI-based message prediction in augmentative and alternative communication (AAC) devices fails to reflect user-specific linguistic identity, resulting in frequent manual editing. To investigate the technical feasibility and sociotechnical implications of personalized language modeling, we conducted the first autoethnographic study embedding hyper-personalized AI into an AAC system. Over seven months, we collected personal communication data and fine-tuned a language model; subsequently, we deployed it for three months in real-world use, supplemented by interaction logging and reflective journaling. Results demonstrate that the customized model significantly reduces editing effort and enhances communicative efficiency. However, it simultaneously surfaces critical tensions—including blurred identity boundaries, trade-offs in privacy relinquishment, contested authorship, and threats to expressive autonomy. This work provides the first systematic empirical account of the dual nature of AI personalization in AAC contexts, offering foundational evidence and theoretical insights for human-centered design and ethical governance of assistive AI.

Technology Category

• Humans and AI: AI for Accessibility
• Natural Language Processing: Ethics — Bias, Fairness, Transparency & Privacy
• Philosophy and Ethics of AI: Artificial General Intelligence

Application Category

• User Modeling, Personalization and Recommendation: Accountability, Transparency, and Ethics for personalization
• Social Networks and Social Media: Generative AI / large language models and their impact on social systems
• Search and Retrieval-Augmented AI: Personalized, context-aware and across-device search

Abstract

Generic AI auto-complete for message composition often fails to capture the nuance of personal identity, requiring significant editing. While harmless in low-stakes settings, for users of Augmentative and Alternative Communication (AAC) devices, who rely on such systems for everyday communication, this editing burden is particularly acute. Intuitively, the need for edits would be lower if language models were personalized to the communication of the specific user. While technically feasible, such personalization raises socio-technical questions: what are the implications of logging one's own conversations, and how does personalization affect privacy, authorship, and control? We explore these questions through an autoethnographic study in three phases: (1) seven months of collecting all the lead author's AAC communication data, (2) fine-tuning a model on this dataset, and (3) three months of daily use of personalized AI suggestions. We reflect on these phases through continuous diary entries and interaction logs. Our findings highlight the value of personalization as well as implications on privacy, authorship, and blurring the boundaries of self-expression.

Problem

• Personalizing AI for AAC users' communication needs
• Exploring privacy and authorship in personalized AI systems
• Assessing socio-technical impacts of ultra-personalized AI

Innovation

• Personalized AI fine-tuning for AAC users
• Autoethnographic data collection and analysis
• Privacy-aware conversational AI personalization