Measuring Inclusion in Interaction: Inclusion Analytics for Human-AI Collaborative Learning

Jaeyoon Choi, Nia Nixon · 2026

Read Measuring Inclusion in Interaction: Inclusion Analytics for Human-AI Collaborative Learning by Jaeyoon Choi, Nia Nixon (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2602-09269.txt.

Introduction

Belonging in team learning is usually judged by rough head counts or after-the-fact surveys, missing how it unfolds moment by moment. Jaeyoon Choi and one co-author bring belonging stats, a talk-based setup that treats belonging as a live, back-and-forth process in team problem solving. Three lenses shape the view: fair speaking, feeling climate, and knowledge fairness.

Scalable talk-level gauges make those ideas plain in real talk, using both fake chats and observed human-AI teaming data. The study bares patterns of who speaks, how ties grow, and whose ideas get taken up, patterns hidden from total or after-the-fact review. As a trial run, it points toward step-focused gauging of belonging in human-AI learning.

Catalog record

CREATORJaeyoon Choi, Nia Nixon
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE37 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2602-09269

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Measuring Inclusion in Interaction: Inclusion Analytics for Human-AI Collaborative Learning Jaeyoon Choi, Nia Nixon Abstract . Inclusion, equity, and access are widely valued in AIED, yet are often assessed through coarse sample descriptors or post-hoc self-reports that miss how inclusion is shaped moment by moment in collaborative problem solving (CPS). In this proof-of-concept paper, we introduce inclusion analytics , a discourse-based framework for examining inclusion as a dynamic, interactional process in CPS. We conceptualize inclusion along three complementary dimensions —participation equity, affective climate, and epistemic equity—and demonstrate how these constructs can be made analytically visible using scalable, interaction-level measures. Using both simulated conversations and empirical data from human-AI teaming experiments, we illustrate how inclusion analytics can surface patterns of participation, relational dynamics, and idea uptake that remain invisible to aggregate or post-hoc evaluations. This work represents an initial step toward process-oriented approaches to measuring inclusion in human-AI collaborative learning environments. 1. Introduction The AIED community has long articulated inclusion, equity, and access as core values guiding the design and evaluation of intelligent learning technologies ( 25 ) . These commitments are reflected in research agendas, society mission statements, recurring conference themes, equity-focused workshops, and community initiatives and awards. Collectively, these signals emphasize a broad consensus that learning technologies should support diverse learners and reduce inequities rather than amplify them. However, while inclusion is widely valued, it remains far less clear how it should be understood, operationalized, and measured, particularly within the dynamics of learning activity itself . In practice, inclusion has often been operationalized through relatively coarse indicators, such as representation counts, demographic breakdowns of study samples, or aggregate pre-post measures of attitudes like belonging. While these approaches are important and informative, they risk reducing inclusion to a checkbox for evaluation: something that can be verified and "set aside", rather than a lived experience that unfolds during learning activity itself. This limitation is particularly consequential in collaborative problem-solving (CPS) contexts ( 20 ) . CPS depends not only on cognitive coordination, but also on social and relational conditions: who participates, whose contributions are acknowledged, how disagreement is handled, and whether learners feel safe contributing ( 11 ; 5 ) . Prior work consistently demonstrates that inclusive climates and psychological safety shape participation, coordination, and team performance ( 18 ; 26 ) . Particularly, in STEM settings, establishing inclusive team climates remains a persistent challenge, as women and underrepresented racial minorities often report fewer opportunities to contribute, diminished recognition of their ideas, and reduced interpersonal power in group interactions. These moment-to-moment experiences accumulate over time, shaping learners’ sense of belonging and long-term engagement in STEM fields ( 7 ) . The growing presence of AI in collaborative learning introduces an additional layer to this challenge. With recent advances in large language models and agentic AI systems, AI is no longer merely a computational tool. Instead, AI can participate directly in group activity as a teammate – suggesting ideas, evaluating options, coordinating decisions, and shaping the flow of collaboration. This shift creates both opportunities and risks. AI teammates may help groups work more effectively, but they may also reshape social dynamics in ways that reinforce existing inequities, particularly for participants whose voices are already less likely to be recognized. Understanding these effects requires analytic tools that can capture inclusion as it unfolds during interaction, rather than only as a retrospective perception. Despite the importance of these concerns, progress has been constrained by a persistent measurement gap: inclusion and sense of belonging are typically conceptualized as affective states and assessed through self-report instruments. Although valuable, these measures provide limited visibility into how inclusion unfolds moment-by-moment during collaboration—who speaks, whose ideas are taken up, how disagreement is navigated, and how affective tone shapes participation. This limitation is particularly problematic in human–AI teaming contexts, where AI participation may redistribute attention, authority, and epistemic weight in ways that are not easily captured by surveys alone. Ironically, this gap persists despite the AIED community’s unique capacity to model learning as a dynamic and interactional process, suggesting an opportunity—and perhaps an obligation—to develop theoretically grounded, analytic approaches that surface the mechanisms and temporal trajectories through which inclusion is produced or undermined during collaborative activity. To address this gap, we introduce the concept of inclusion analytics : a discourse-based measurement framework that operationalizes inclusion through observable interaction patterns. We conceptualize inclusion along three complementary dimensions, namely participation equity, affective climate, and epistemic equity – each capturing a distinct aspect of learners’ lived experiences in collaborative activity. Using empirical data from human-AI teaming experiments, complemented by simulated collaborative discourse designed to isolate and illustrate specific inclusion mechanisms, we present an initial proof of concept demonstrating how inclusion analytics can surface inequities that remain invisible to aggregated or post-hoc measures. More broadly, we argue that inclusion analytics represents a methodological direction for the AIED community, one that leverages its theoretical foundations and technical capabilities to move beyond checklists and self-reports toward a richer, process-oriented understanding of inclusion in human-AI collaboration. 2. Inclusion Analytics: A Multi-Dimensional Framework In this paper, we conceptualize inclusion in CPS as an interactional construct that is enacted through observable patterns of discourse rather than solely experienced as an internal state. Specifically, we operationalize inclusion along three complementary dimensions: participation equity (i.e., who has access to the conversational floor), affective climate (i.e., whether interactional norms support social safety and mutual respect), and epistemic equity (i.e., whose ideas are taken up and shape the group’s reasoning). 2.1. Participation Equity Balanced participation in CPS is widely recognized as an important condition for successful collaboration. Prior work in collaborative learning consistently shows that collaborative outcomes are more productive when participation is more evenly distributed across group members, as this allows learners to more fully leverage one another’s knowledge and skills ( 28 ) . For example, groups in which conversational participation is dominated by a small subset of members are less collectively intelligent than those characterized by more equal distributions of turn-taking. Accordingly, whether learners in group have comparably distributed access to the conversational floor, also known as participation equity, represents a critical dimension for examining the effectiveness of collaborative activity. However, participation equity is more than a performance-related concern; it is also a fundamentally an inclusion issue. CPS is an interactional process through which ideas are externalized, negotiated, and integrated through discussions. When participation is imbalanced, learners who speak less have fewer opportunities to verbalize their reasonings, explain their ideas, or engage in interactive sense-making, hence limiting opportunities for learning ( 13 ) . Moreover, persistent participation imbalances can result in some learners’ contributions remaining unheard or unacknowledged, fostering experiences of exclusion and undermining a sense of belonging, particularly for students from marginalized or underrepresented identities ( 17 ) . For these reasons, we include participation equity as a core dimension of inclusion analytics, conceptualizing it as an interactional property of discourse. Consistent with prior work, we operationalize participation equity using observable indicators such as turn counts and word counts, which are aggregated into group-level indices that capture the degree of participation imbalance ( 1 ) . 2.2. Affective Climate Affective climate plays an important role in CPS as it shapes whether learners feel safe to participate in group interactions.

Read the complete text