Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

More Than 1v1: Human-AI Alignment in Early Developmental Communities with Multimodal LLMs

Weiyan Shi, Kenny Tsu Wei Choo

TL;DR

It is argued that alignment in developmental settings should be treated as a community-governed process rather than an individual optimisation problem.

Abstract

In early developmental contexts, particularly in parent-child interaction analysis, alignment involves families and professionals such as speech-language pathologists (SLPs) who interpret children's everyday interactions from different roles. When multimodal large language models (MLLMs) are introduced to support this process, alignment becomes a question of how authority, responsibility, and emotional risk are distributed across stakeholders. Through a three-part study with five families and three SLPs, we trace how MLLM-generated outputs move from expert-facing analysis to parent-facing feedback. We propose layered community alignment: grounding representations in expert-aligned structures, mediating translation through professional guardrails, and enabling family-level adaptation within those boundaries. We argue that alignment in developmental settings should be treated as a community-governed process rather than an individual optimisation problem.

More Than 1v1: Human-AI Alignment in Early Developmental Communities with Multimodal LLMs

TL;DR

It is argued that alignment in developmental settings should be treated as a community-governed process rather than an individual optimisation problem.

Abstract

In early developmental contexts, particularly in parent-child interaction analysis, alignment involves families and professionals such as speech-language pathologists (SLPs) who interpret children's everyday interactions from different roles. When multimodal large language models (MLLMs) are introduced to support this process, alignment becomes a question of how authority, responsibility, and emotional risk are distributed across stakeholders. Through a three-part study with five families and three SLPs, we trace how MLLM-generated outputs move from expert-facing analysis to parent-facing feedback. We propose layered community alignment: grounding representations in expert-aligned structures, mediating translation through professional guardrails, and enabling family-level adaptation within those boundaries. We argue that alignment in developmental settings should be treated as a community-governed process rather than an individual optimisation problem.
Paper Structure (14 sections, 2 figures)

This paper contains 14 sections, 2 figures.

Table of Contents

  1. Introduction
  2. Empirical Lens: Probing Alignment Tensions
  3. Design Overview
  4. Analytic Approach
  5. Tracing Alignment Tensions Across Layers
  6. Layer 1: Expert-Readable Representation — Alignment as Structured Visibility
  7. Layer 2: Expert-Mediated Translation — Alignment as Emotional Safeguarding
  8. Layer 3: Family-Level Adaptation — Alignment as Contextual Negotiation
  9. Conclusion and Discussion
  10. Alignment and Professional Authority
  11. Alignment and Emotional Risk
  12. Alignment, Personalisation, and Epistemic Drift
  13. Alignment as Distributed Responsibility
  14. From Community Alignment to Societal Infrastructure

Figures (2)

  • Figure 1: Overview of the three-part study. Part I collected naturalistic parent--child interaction videos and generated expert-aligned SLP-facing analyses. Part II involved SLP evaluation and steering of expert outputs. Part III evaluated a parent-facing prototype grounded in practitioner-informed guidance.
  • Figure 2: System Overview. Both interfaces are grounded in the same segment-level MLLM analysis of parent–child interaction (parent--child interaction) videos. SLPView (left) provides structured expert-oriented behavioural assessment with quality labels and aligned timelines. ParentView (right) reframes this analysis into descriptive summaries and supportive suggestions, removing evaluative framing while preserving segment-level continuity.