arxiv: 2604.18850 · v1 · submitted 2026-04-20 · 💻 cs.HC · cs.AI· cs.SI

Recognition: unknown

The Triadic Loop: A Framework for Negotiating Alignment in AI Co-hosted Livestreaming

Katherine Wang , Nadia Berthouze , Aneesha Singh

Authors on Pith no claims yet

Pith reviewed 2026-05-10 03:21 UTC · model grok-4.3

classification 💻 cs.HC cs.AIcs.SI

keywords Triadic LoopAI co-hostlivestreamingbidirectional adaptationmulti-party interactionstrategic misalignmentrelational evaluationcollaborative AI

0 comments

The pith

Alignment in AI co-hosted livestreaming arises from a cycle of mutual adaptations among the streamer, AI co-host, and audience.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper introduces the Triadic Loop to describe how alignment works when an AI joins a streamer and live audience in real time. It claims that each pair of actors continuously reshapes the others through feedback, so problems in any one pair can affect the whole setup. This view departs from standard AI alignment that treats the system as following commands from one user. A reader would care because livestreaming already involves group social cues that single-user models overlook, and the framework points to ways AI can participate in those cues instead of just responding to them.

Core claim

The Triadic Loop reconceptualizes alignment in AI co-hosted livestreaming as a temporally reinforced process of bidirectional adaptation among three actors: streamer ↔ AI co-host, AI co-host ↔ audience, and streamer ↔ audience. Unlike instruction-following paradigms, bidirectional alignment requires each actor to continuously reshape the others, meaning misalignment in any sub-loop can destabilize the broader system. AI co-hosts function not only as mediators but as performative participants and community members shaping collective meaning-making. The framework also proposes strategic misalignment as a mechanism for sustaining engagement and introduces three relational evaluation constructs.

What carries the argument

The Triadic Loop, a conceptual model of bidirectional adaptation across three interconnected actor pairs that treats misalignment in any pair as a threat to overall stability.

If this is right

Misalignment in any single sub-loop can destabilize the entire alignment process across the three actors.
AI co-hosts can serve as active participants that help shape collective meaning rather than only relaying information.
Strategic misalignment between actors can be used deliberately to keep audience engagement from dropping.
Design choices for AI co-hosts should focus on maintaining social coherence across all three relationships at once.
Relational evaluation constructs based on existing instruments can measure the health of these multi-party loops.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

The same three-party reinforcement idea might apply to AI in other live group settings such as collaborative online tools or virtual events.
Training data for such AI systems would need to capture how audience reactions feed back into streamer-AI exchanges.
Testing could involve temporarily breaking one loop in a real stream and checking whether engagement metrics drop as the model predicts.

Load-bearing premise

That the three-way structure and its reinforcing loops are the main drivers of alignment stability and that AI co-hosts can reliably act as performative community members without further empirical checks.

What would settle it

A controlled observation of AI co-hosted streams in which one actor pair is deliberately misaligned while the other two remain aligned, followed by measurement of whether overall engagement and meaning-making stay stable.

Figures

Figures reproduced from arXiv: 2604.18850 by Aneesha Singh, Katherine Wang, Nadia Berthouze.

read the original abstract

AI systems are increasingly embedded in multi-user social environments, yet most alignment frameworks conceptualize interaction as a dyadic relationship between a single user and an AI system. Livestreaming platforms challenge this assumption: interaction unfolds among streamers and audiences in real time, producing dynamic affective and social feedback loops. In this paper, we introduce the Triadic Loop, a conceptual framework that reconceptualizes alignment in AI co-hosted livestreaming as a temporally reinforced process of bidirectional adaptation among three actors: streamer $\leftrightarrow$ AI co-host, AI co-host $\leftrightarrow$ audience, and streamer $\leftrightarrow$ audience. Unlike instruction-following paradigms, bidirectional alignment requires each actor to continuously reshape the others, meaning misalignment in any sub-loop can destabilize the broader system. Drawing on literature from multi-party interaction, collaborative AI, and relational agents, we articulate how AI co-hosts function not only as mediators but as performative participants and community members shaping collective meaning-making. We further propose "strategic misalignment" as a mechanism for sustaining community engagement and introduce three relational evaluation constructs grounded in established instruments. The framework contributes a model of dynamic multi-party alignment, an account of cross-loop reinforcement, and design implications for AI co-hosts that sustain social coherence in participatory media environments.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

This paper gives a clean conceptual model for three-way alignment in AI livestream co-hosting but stops short of any tests or examples.

read the letter

The main point is that alignment here is not just streamer-to-AI but runs through three loops: streamer-AI, AI-audience, and streamer-audience. Misalignment in one can ripple through the others, and the authors treat the AI as an active participant that helps shape group meaning rather than just following instructions. They also float strategic misalignment as a deliberate tactic to keep audiences engaged and tie three evaluation ideas to existing scales. That synthesis from multi-party and relational-agent work is the clearest addition; it reframes the problem in a way that fits live social platforms better than standard dyadic alignment talk. The framework hangs together without obvious internal contradictions or circular definitions, and the design implications follow logically from the loops they describe. The absence of data is the real limit. No stream logs, no pilot sessions, no comparison against simpler models, so the claims about cross-loop reinforcement and community stability remain plausible but unexamined. The evaluation constructs are referenced to prior instruments, which helps, yet without even a worked example it is hard to see how they would actually be applied or what they would reveal. Readers working on HCI for social media or collaborative AI will find the structure useful for scoping new projects or workshops. It is not a finished result but a map that could guide later empirical work. I would send it to referees with a request for at least one concrete scenario or small validation step, because the core idea is distinct enough to warrant that next round of scrutiny.

Referee Report

0 major / 2 minor

Summary. This paper claims to present the Triadic Loop as a new conceptual framework for alignment in AI co-hosted livestreaming. Alignment is described as bidirectional adaptation in three actor pairs (streamer-AI co-host, AI co-host-audience, streamer-audience), forming a temporally reinforced system where misalignment in one loop affects the whole. The work synthesizes literature on multi-party interaction, collaborative AI, and relational agents, argues for AI co-hosts as active community members, introduces the concept of strategic misalignment to maintain engagement, and proposes three relational evaluation constructs based on established instruments.

Significance. The significance lies in extending alignment research from dyadic to triadic multi-party settings in real-time social media. If the framework proves useful, it could influence the design of AI systems for livestreaming and similar participatory environments by emphasizing cross-actor adaptation and social coherence. The literature synthesis and design implications are valuable contributions to the HCI community, providing a foundation for future studies on dynamic alignment processes.

minor comments (2)

[Abstract] The abstract introduces 'strategic misalignment' without a brief definition or example, which may leave readers unclear on its role until the main text.
[Evaluation constructs] The three relational evaluation constructs are proposed but their grounding in specific established instruments (e.g., which questionnaires or scales) should be detailed to allow for immediate use by researchers.

Simulated Author's Rebuttal

0 responses · 0 unresolved

We thank the referee for their positive summary of the manuscript, recognition of its potential contributions to multi-party alignment research in HCI, and recommendation for minor revision. We appreciate the acknowledgment that the Triadic Loop framework extends dyadic alignment concepts to real-time participatory environments and offers valuable design implications.

Circularity Check

0 steps flagged

No significant circularity: conceptual synthesis from external literature

full rationale

The paper introduces the Triadic Loop as a new conceptual framework by synthesizing established external domains (multi-party interaction, collaborative AI, relational agents). No equations, parameters, empirical fits, or predictions exist that could reduce to self-inputs. The framework is defined through its own articulation but relies on independent prior literature without self-citation chains or definitional loops. This is a standard non-circular outcome for a purely conceptual contribution that remains self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 3 axioms · 2 invented entities

The contribution is the introduction of a new conceptual model. It rests on domain assumptions from cited fields rather than new data or derivations. The framework itself and strategic misalignment are the primary invented elements.

axioms (3)

domain assumption Most alignment frameworks conceptualize interaction as dyadic between single user and AI
Stated as the assumption challenged by livestreaming contexts.
domain assumption Livestreaming produces dynamic affective and social feedback loops among multiple actors
Core premise drawn from multi-party interaction literature.
domain assumption Bidirectional alignment requires continuous reshaping among actors and misalignment in one sub-loop can destabilize the system
Load-bearing claim of the Triadic Loop model.

invented entities (2)

Triadic Loop no independent evidence
purpose: Framework modeling alignment as temporally reinforced bidirectional adaptation among three actors
Newly proposed conceptual structure without external validation in the abstract.
strategic misalignment no independent evidence
purpose: Mechanism for sustaining community engagement
Introduced as part of the framework to explain how controlled deviation can support participation.

pith-pipeline@v0.9.0 · 5531 in / 1671 out tokens · 45564 ms · 2026-05-10T03:21:39.725394+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

58 extracted references · 5 canonical work pages · 1 internal anchor

[1]

Amanda Askell, Yuntao Bai, Anna Chen, Dawn Drain, Deep Ganguli, Tom Henighan, Andy Jones, Nicholas Joseph, Ben Mann, Nova DasSarma, et al. 2021. A general language assistant as a laboratory for alignment.arXiv preprint arXiv:2112.00861(2021). The Triadic Loop: A Framework for Negotiating Alignment in AI Co-hosted Livestreaming CHI ’26 BiAlign Workshop, Ap...

work page internal anchor Pith review arXiv 2021
[2]

Timothy Bickmore and Rosalind Picard. 2005. Establishing and Maintaining Long- Term Human-Computer Relationships.ACM Transactions on Computer-Human Interaction12, 2 (2005), 293–327

2005
[3]

Frank Biocca, Chad Harms, and Judee K Burgoon. 2003. The Core of Networked Presence: Self, Agency, and Embodiment.Presence: Teleoperators and Virtual Environments12, 5 (2003), 456–480

2003
[4]

Susanne Biundo, Daniel Höller, Bernd Schattenberg, and Pascal Bercher. 2016. Companion-technology: an overview.KI-Künstliche Intelligenz30, 1 (2016), 11– 20

2016
[5]

Anita L Blanchard. 2007. Developing a sense of virtual community measure. CyberPsychology & Behavior10, 6 (2007), 827–830

2007
[6]

Dan Bohus and Eric Horvitz. 2011. Multiparty Turn Taking in Situated Dialog: Study, Lessons, and Directions. InProceedings of SIGDIAL

2011
[7]

Simone Borsci, Alessio Malizia, Martin Schmettow, Frank Van Der Velde, Gunay Tariverdiyeva, Divyaa Balaji, and Alan Chamberlain. 2022. The chatbot usability scale: the design and pilot of a usability scale for interaction with AI-based conversational agents.Personal and ubiquitous computing26, 1 (2022), 95–119

2022
[8]

Cynthia Breazeal and Brian Scassellati. 2000. Infant-like social interactions between a robot and a human caregiver.Adaptive Behavior8, 1 (2000), 49–74

2000
[9]

Jie Cai, Sagnik Chowdhury, Hongyang Zhou, and Donghee Yvette Wohn. 2023. Hate raids on twitch: Understanding real-time human-bot coordinated attacks in live streaming communities.Proceedings of the ACM on Human-Computer Interaction7, CSCW2 (2023), 1–28

2023
[10]

Chun-Wei Chiang, Zhuoran Lu, Zhuoyan Li, and Ming Yin. 2024. Enhancing ai-assisted group decision making through llm-powered devil’s advocate. In Proceedings of the 29th International Conference on Intelligent User Interfaces. 103–119

2024
[11]

Paul T Costa and Robert R McCrae. 1992. Normal personality assessment in clinical practice: The NEO Personality Inventory.Psychological assessment4, 1 (1992), 5

1992
[12]

1990.Flow: The psychology of optimal experience

Mihaly Csikszentmihalyi and Mihaly Csikzentmihaly. 1990.Flow: The psychology of optimal experience. Vol. 1990. Harper & Row New York

1990
[13]

Taufiq Daryanto, Xiaohan Ding, Kaike Ping, Lance T Wilhelm, Yan Chen, Chris Brown, and Eugenia H Rho. 2026. Human-Human-AI Triadic Programming: Uncovering the Role of AI Agent and the Value of Human Partner in Collaborative Learning.arXiv preprint arXiv:2601.12134(2026)

work page arXiv 2026
[14]

Emilie Delaherche, Mohamed Chetouani, Ammar Mahdhaoui, Catherine Saint- Georges, Sylvie Viaux, and David Cohen. 2012. Interpersonal synchrony: A survey of evaluation methods across disciplines.IEEE Transactions on Affective Computing3, 3 (2012), 349–365

2012
[15]

Dobai, L

I. Dobai, L. Rothkrantz, and C. van der Mast. 2005. Personality model for a companion AIBO. InProceedings of the 2005 ACM SIGCHI International Conference on Advances in Computer Entertainment Technology. 438–441

2005
[16]

I. Gabriel. 2020. Artificial intelligence, values, and alignment.Minds and Machines 30 (2020)

2020
[17]

W. A. Hamilton, A. Kerne, and T. L. Robbins. 2014. Streaming on Twitch: fostering participatory communities of play. InProc. CHI 2014

2014
[18]

Catherine Han, Joseph Seering, Deepak Kumar, Jeffrey T Hancock, and Zakir Durumeric. 2023. Hate raids on twitch: Echoes of the past, new modalities, and implications for platform governance.Proceedings of the ACM on Human- Computer Interaction7, CSCW1 (2023), 1–28

2023
[19]

Donna L Hoffman and Thomas P Novak. 2009. Flow online: lessons learned and future prospects.Journal of interactive marketing23, 1 (2009), 23–34

2009
[20]

Stephanie Houde, Kristina Brimijoin, Michael Muller, Steven I Ross, Dario Andres Silva Moran, Gabriel Enrique Gonzalez, Siya Kunde, Morgan A Foreman, and Justin D Weisz. 2025. Controlling AI Agent Participation in Group Conversations: A Human-Centered Approach. InProceedings of the 30th International Conference on Intelligent User Interfaces. 390–408

2025
[21]

Houtti, M

M. Houtti, M. Zhou, L. Terveen, and S. Chancellor. 2025. Observe, Ask, Intervene: Designing AI Agents for More Inclusive Meetings. InProc. CHI 2025

2025
[22]

Hu and G

Y. Hu and G. Freeman. 2026. Beyond a Conventional Chatbot: How AI Streamers Transcend Live Streaming Experiences from Viewers’ Perspectives. InProceedings of the 2026 CHI Conference on Human Factors in Computing Systems

2026
[23]

Susan A Jackson and Herbert W Marsh. 1996. Development and validation of a scale to measure optimal experience: The Flow State Scale.Journal of sport and exercise psychology18, 1 (1996), 17–35

1996
[24]

Shreya Jain, Dipankar Niranjan, Hemank Lamba, Neil Shah, and Ponnurangam Kumaraguru. 2019. Characterizing and detecting livestreaming chatbots. In Proceedings of the 2019 IEEE/ACM International Conference on Advances in Social Networks Analysis and Mining. 683–690

2019
[25]

Jiaming Ji, Tianyi Liu, Jiayi Dai, Xuehai Pan, Chi Zhang, Ce Bian, Boyuan Chen, Ruiyang Sun, Yizhou Wang, and Yaodong Yang. 2023. AI Alignment: A Compre- hensive Survey.arXiv preprint arXiv:2310.19852(2023)

work page arXiv 2023
[26]

Mark R Johnson. 2024. Humour and comedy in digital game live streaming.New Media & Society26, 6 (2024), 3045–3067

2024
[27]

Hanbyul Joo, Tomas Simon, Mina Cikara, and Yaser Sheikh. 2019. Towards social artificial intelligence: Nonverbal social signal prediction in a triadic interaction. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. 10873–10883

2019
[28]

Minseong Kim and Hyung-Min Kim. 2022. What online game spectators want from their twitch streamers: Flow and well-being perspectives.Journal of Retail- ing and Consumer Services66 (2022), 102951

2022
[29]

Kimani, D

E. Kimani, D. Parmar, P. Murali, and T. Bickmore. 2021. Sharing the Load Online: Virtual Presentations with Virtual Co-Presenter Agents. InCHI EA 2021

2021
[30]

Andrea Krausman, Catherine Neubauer, Daniel Forster, Shan Lakhmani, An- thony L Baker, Sean M Fitzhugh, Gregory Gremillion, Julia L Wright, Jason S Metcalfe, and Kristin E Schaefer. 2022. Trust measurement in human-autonomy teams: Development of a conceptual toolkit.ACM Transactions on Human-Robot Interaction (THRI)11, 3 (2022), 1–58

2022
[31]

Mareike Kritzler, Jack Hodges, Dan Yu, Kimberly Garcia, Hemant Shukla, and Flo- rian Michahelles. 2019. Digital companion for industry. InCompanion Proceedings of The 2019 World Wide Web Conference. 663–667

2019
[32]

Soohwan Lee, Seoyeong Hwang, Dajung Kim, and Kyungho Lee. 2025. Conver- sational Agents as Catalysts for Critical Thinking: Challenging Social Influence in Group Decision-making. InProceedings of the Extended Abstracts of the CHI Conference on Human Factors in Computing Systems. 1–12

2025
[33]

SooHwan Lee, Mingyu Kim, Seoyeong Hwang, Dajung Kim, and Kyungho Lee
[34]

InCompanion Proceedings of the 30th Interna- tional Conference on Intelligent User Interfaces

Amplifying Minority Voices: AI-Mediated Devil’s Advocate System for Inclusive Group Decision-Making. InCompanion Proceedings of the 30th Interna- tional Conference on Intelligent User Interfaces. 17–21
[35]

John Zhang

Xinxin Li, Lorin M Hitt, and Z. John Zhang. 2024. Exploring the impact of AI assistants on consumer behavior in livestream shopping.Journal of Interactive Marketing(2024)

2024
[36]

Yi Li and Yi Peng. 2021. What drives gift-giving intention in live streaming? The perspectives of emotional attachment and flow experience.International Journal of Human–Computer Interaction37, 14 (2021), 1317–1329

2021
[37]

Wenhan Lyu, Yimeng Wang, Yifan Sun, and Yixuan Zhang. 2025. Will your next pair programming partner be human? an empirical evaluation of generative ai as a collaborative teammate in a semester-long classroom setting. InProceedings of the Twelfth ACM Conference on Learning@ Scale. 83–94

2025
[38]

T. Mok, C. M. Au Yueng, A. Tang, and L. Oehlberg. 2020. Talk Like Somebody is Watching: Understanding and Supporting Novice Live Streamers. InProc. IMX 2020

2020
[39]

Michael Muller, Stephanie Houde, Gabriel Gonzalez, Kristina Brimijoin, Steven I Ross, Dario Andres Silva Moran, and Justin D Weisz. 2024. Group brainstorming with an ai agent: Creating and selecting ideas. InInternational conference on computational creativity. 10

2024
[40]

Long Ouyang, Jeffrey Wu, Xu Jiang, Diogo Almeida, Carroll Wainwright, Pamela Mishkin, Chong Zhang, Sandhini Agarwal, Katarina Slama, Alex Ray, et al. 2022. Training language models to follow instructions with human feedback.NeurIPS 35 (2022)

2022
[41]

A. Rogge. 2023. Defining, designing and distinguishing artificial companions: A systematic literature review.International Journal of Social Robotics15, 9 (2023), 1557–1579

2023
[42]

Seering et al

J. Seering et al. 2020. It takes a village: Integrating an adaptive chatbot into an online gaming community. InProc. CHI 2020

2020
[43]

Seering, J

J. Seering, J. P. Flores, S. Savage, and J. Hammer. 2018. The social roles of bots: evaluating impact of bots on discussions in online communities.Proc. ACM Hum.-Comput. Interact.2, CSCW (2018)

2018
[44]

Seering, M

J. Seering, M. Luria, G. Kaufman, and J. Hammer. 2019. Beyond dyadic interactions: Considering chatbots as community members. InProc. CHI 2019

2019
[45]

Hafsah Sheikh, Saad Ahmed, Junaid Qadir, and Ala Al-Fuqaha. 2024. The Role of AI Agents in Supporting Collaborative Learning in the Classroom. InProceedings of the 2024 CHI Conference on Human Factors in Computing Systems

2024
[46]

Tianhao Shen, Renren Jin, Yufei Huang, Chuang Liu, Weilong Dong, Ziyi Guo, Xinwei Wu, Yan Liu, and Deyi Xiong. 2024. Position: Towards Bidirectional Human-AI Alignment: A Systematic Survey.arXiv preprint arXiv:2406.09264 (2024)

work page arXiv 2024
[47]

Strohmann, D

T. Strohmann, D. Siemon, B. Khosrawi-Rad, and S. Robra-Bissantz. 2023. Toward a design theory for virtual companionship.Human–Computer Interaction38, 3-4 (2023), 194–234

2023
[48]

Trinh, L

H. Trinh, L. Ring, and T. Bickmore. 2015. DynamicDuo: Co-presenting with Virtual Agents. InProc. CHI 2015

2015
[49]

Urs and S

A. Urs and S. Kaiser. 2025. Backseat Gaming Permitted: Meet the AI Co-Streamer from Streamlabs. InSIGGRAPH Real-Time Live!

2025
[50]

Wang et al

L. Wang et al. 2025. Artificial Intelligence (AI) Assistant in Online Shopping: A Randomized Field Experiment on a Livestream Selling Platform.Information Systems Research(2025)

2025
[51]

Liwei Wang, Yongda Zhou, Ang Li, and Alex Luo. 2021. CASS: Towards Building a Social-Support Chatbot for Online Health Community.Proc. ACM Hum.-Comput. Interact.5, CSCW1 (2021)

2021
[52]

Anna Xygkou, Chee Siang Ang, Panote Siriaraya, Jonasz Piotr Kopecki, Alexan- dra Covaci, Eiman Kanjo, and Wan-Jou She. 2024. MindTalker: Navigating the complexities of AI-enhanced social engagement for people with early-stage de- mentia. InProceedings of the 2024 CHI Conference on Human Factors in Computing Systems. 1–15. CHI ’26 BiAlign Workshop, April 1...

2024
[53]

Jiaxun Ye, Yuxiang Cai, Yuhan Liu, and Zhicong Lu. 2025. My Favorite Streamer is an LLM: Discovering, Bonding, and Co-Creating in AI VTuber Fandom.arXiv preprint arXiv:2509.10427(2025)

work page arXiv 2025
[54]

Ryan Yen, Li Feng, Brinda Mehra, Ching Christie Pang, Siying Hu, and Zhicong Lu. 2023. Storychat: Designing a narrative-based viewer participation tool for live streaming chatrooms. InProceedings of the 2023 CHI Conference on Human Factors in Computing Systems. 1–18

2023
[55]

J Diego Zamfirescu-Pereira, Richmond Y Wong, Bjoern Hartmann, and Qian Yang
[56]

InProceedings of the 2023 CHI conference on human factors in computing systems

Why Johnny can’t prompt: how non-AI experts try (and fail) to design LLM prompts. InProceedings of the 2023 CHI conference on human factors in computing systems. 1–21

2023
[57]

Rui Zhang, Wen Duan, Christopher Flathmann, Nathan McNeese, Bart Knijnen- burg, and Guo Freeman. 2024. Verbal vs. Visual: How Humans Perceive and Collaborate with AI Teammates Using Different Communication Modalities in Various Human-AI Team Compositions.Proceedings of the ACM on Human- Computer Interaction8, CSCW2 (2024), 1–34

2024
[58]

Sijin Zhu, Zheng Wang, Yuan Zhuang, Yuyang Jiang, Mengyao Guo, Xiaolin Zhang, and Ze Gao. 2024. Exploring the impact of ChatGPT on art creation and collaboration: Benefits, challenges and ethical implications.Telematics and Informatics Reports14 (2024), 100138. Received 20 February 2007; revised 12 March 2009; accepted 5 June 2009

2024