Pith. sign in

REVIEW 4 major objections 4 minor 73 references

From Divergence to Consensus: Evaluating the Role of Large Language Models in Facilitating Agreement through Adaptive Strategies

T0 review · 4 major / 4 minor · reviewed 2026-08-09 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read The paper claims that ChatGPT 4.0 outperforms the two other tested LLMs as an automated consensus facilitator, achieving higher cosine-similarity alignment in fewer iterations.

desk verdict A useful pilot platform, but the ChatGPT-4.0 superiority claim rests on an unvalidated similarity metric that likely rewards echo, not consensus. read the letter →

arxiv 2503.15521 v1 pith:S2LQPCE7 submitted 2025-02-03 cs.HC cs.AIcs.CLcs.CY

Loukas Triantafyllopoulos , Dimitris Kalles This is my paper · ORCID
classification cs.HCcs.AIcs.CLcs.CY
keywords consensus-buildinglargelanguagemodelscosinesimilarityadaptivefacilitationstrategiescollaborativedecisionsupportsustainabledevelopmentgoalshuman-AIinteraction
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper sets out to establish that a large language model can act as an automated facilitator in a synchronous multi-user chat system, guiding participants from divergent written opinions to an accepted consensus proposal. It compares three models, ChatGPT 4.0, Mistral Large 2, and AI21 Jamba Instruct, in a purpose-built platform where the LLM selects among five facilitation strategies and iteratively revises its proposal after user feedback. The success measure is the cosine similarity between participants' initial opinions and the final accepted proposal, computed with sentence embeddings. On 75 sessions involving 30 students, ChatGPT 4.0 scored highest, averaging 0.701 versus 0.6126 for AI21 Jamba and 0.5807 for Mistral Large 2, and never needed more than three iterations to reach consensus. A sympathetic reader would care because automated LLM facilitation of this kind, if valid, could scale consensus-building to online discussions that human facilitators cannot cover.

What carries the argument

The load-bearing mechanism is an iterative propose-reject-revise loop in a custom-built chat system. After participants state their initial positions, the LLM drafts a consensus proposal; each participant accepts or rejects it; on rejection the LLM chooses one of five predefined facilitation strategies—clarify understanding, summarize the discussion, highlight common ground, propose a compromise, or reframe the question—and drafts a revised proposal. The evaluation machinery is cosine similarity between the sentence-embedding vectors of each participant's initial opinion and the accepted consensus proposal, supplemented by the same similarity computed across intermediate proposals to track convergence.

What would settle it

A control condition with a facilitator that simply outputs the mean embedding of the two participants' initial opinions would settle the matter: if that echo baseline reaches or exceeds ChatGPT 4.0's 0.701 average cosine similarity, then the reported advantage is an artifact of measuring proximity to initial positions rather than genuine consensus-building.

Watch

Extended reading notes

Core claim

The central claim is comparative: among the three tested LLMs, ChatGPT 4.0 is the most effective facilitator in this setup because its accepted consensus proposals lie closest to participants' original positions and are accepted after fewer rounds. The paper reports an average cosine similarity of 0.701 for ChatGPT 4.0, 0.6126 for AI21 Jamba, and 0.5807 for Mistral Large 2 across 124 measured occasions, and ChatGPT 4.0 was the only model whose discussions always concluded within three iterations. The authors interpret this as evidence that the model selects the most appropriate adaptive strategy from the five available options, producing proposals that participants recognize as sufficiently aligned with their views. They also report topic-level variation, with the highest average similarity on health and education discussions and lower values on climate and water topics, while ChatGPT 4.0 led on all four sustainability topics.

Load-bearing premise

The ranking depends on cosine similarity between initial opinions and the accepted proposal being a valid surrogate for consensus quality, so a facilitator that echoes participants' starting positions can score high without producing genuine agreement.

Editorial extensions

If this is right

  • If the central claim is right, ChatGPT 4.0 can be deployed as a faster consensus facilitator than the two alternatives in this two-participant chat setting, closing agreements in at most three rounds.
  • The five adaptive strategies give LLM facilitators a concrete action vocabulary, and the study implies that the choice of strategy affects how quickly a proposal becomes acceptable to the group.
  • Cosine similarity between initial opinions and final proposals can serve as a quantitative, repeatable evaluation metric for automated facilitation, allowing future systems to tune prompts and strategies against it.
  • The topic-level results suggest that facilitator performance will not be uniform across policy domains, since climate and water discussions produced lower average alignment than health and education discussions.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • A baseline facilitator that simply echoes the average of participants' initial statements would likely score well on cosine similarity; testing such a baseline would show whether ChatGPT 4.0's 0.701 reflects consensus skill or mere proximity to initial positions.
  • Because the sessions were conducted in Greek while two of the models were trained mainly on English and one mainly on French, the observed ranking may partly be a language-proficiency effect; repeating the study in each model's dominant language is a direct test.
  • The two-participant design means 'consensus' here is a binary acceptance by two students, so extending the framework to larger groups would require a different acceptance rule and would likely change both strategy selection and iteration counts.
  • The absence of a clear elbow point in the similarity trajectories suggests cosine similarity alone does not tell a facilitator when to stop iterating; a stopping rule would need an independent signal such as participant satisfaction or a stability threshold.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 4 minor

Summary. The paper proposes a framework in which three large language models (ChatGPT 4.0, Mistral Large 2, and AI21 Jamba Instruct) act as automated facilitators in a custom multi-user chat system, applying one of five consensus-building strategies iteratively until participants accept a proposal. The evaluation uses cosine similarity, computed with the Universal Sentence Encoder, between participants' initial opinions and the final consensus proposal as a surrogate for consensus quality. Based on 75 two-participant sessions on sustainability topics, the paper reports that ChatGPT 4.0 achieves the highest average cosine similarity (0.701 versus 0.6126 for AI21 Jamba and 0.5807 for Mistral Large 2) and reaches consensus in fewer iterations. The paper concludes that ChatGPT 4.0 is the most effective facilitator and discusses limitations including the surrogate metric, participant homogeneity, and the Greek-language setting.

Significance. If the central claim were established, the paper would offer a useful comparison of LLM facilitators and a reusable platform for studying AI-mediated consensus building. The study has genuine strengths: it is a real deployment with human participants, it uses an ethics-approved protocol, it explicitly names its metric as a surrogate, and it reports per-topic and per-iteration analyses rather than only aggregate scores. However, the headline result rests entirely on an unvalidated surrogate whose direction of bias favors superficial agreement, and the quantitative support is weakened by sample-size inconsistencies and the absence of any inferential statistics. The paper also explicitly acknowledges that discussions were conducted in Greek while two of the models were primarily trained in English and one in French, adding a further confound to the comparison. Because the metric question is load-bearing, the current evidence does not support the claimed superiority of ChatGPT 4.0.

major comments (4)
  1. [§3.3 and §5.1] The central evaluation metric is cosine similarity between participants' initial opinions and the final accepted proposal. As the paper itself calls this a "surrogate for consensus," it is not established that higher similarity corresponds to better facilitation. A facilitator that simply restated each participant's own initial position would achieve near-perfect similarity without fostering deliberation, compromise, or genuine agreement. The discussion in §6 acknowledges that ChatGPT 4.0's proposals "aligned more closely with users' initial opinions," which is precisely the behavior that the metric rewards. The paper provides no validation of the surrogate against human judgments of consensus quality, and no comparison against a trivial echo baseline. Without such validation or a baseline, the claim of ChatGPT 4.0's superiority is unsupported; I ask the authors to add a baseline condition or an external validation of the metric against perceived consensus.
  2. [Table 3 and §4] The reported occasion counts do not add up. The paper states that 75 sessions were conducted with two participants each, which implies 150 participant-level observations; Table 3 reports 124 "occasions" (50 + 40 + 34 = 124), and Table 4's per-topic counts also sum to 124. The discrepancy is unexplained. Either the definition of an "occasion" differs from a participant's initial opinion, or data are missing, or some sessions had different participation structures. This is load-bearing because every average in Tables 3 and 4 depends on these counts, and the missing 26 observations could change the model ranking.
  3. [Tables 3 and 4, §5.1] The comparison of models rests on differences in average cosine similarity (e.g., 0.701 vs. 0.6126 vs. 0.5807) without any measure of variance, confidence intervals, or significance tests. The per-model samples are small (50, 40, 34) and some per-topic cells have only two observations (e.g., climate action for ChatGPT 4.0 and Mistral Large 2). With cosine similarities computed from embeddings that the paper itself notes are noisy in a non-primary language, these differences may be well within sampling error. The paper should report the distribution of similarities, include error bars, and apply an appropriate inferential test before claiming that ChatGPT 4.0 is superior.
  4. [§6 and §4] The discussion section states that the findings were consistent "across all discussions in this pilot study, which addressed topics spanning the 17 SDGs," but the study actually used six questions covering four SDGs (good health and well-being, climate action, quality education, and clean water and sanitation). This overstatement is not central to the method, but it misrepresents the scope of the evidence and should be corrected to refer to the four SDGs actually studied.
minor comments (4)
  1. [§3.3, Eq. (1)] The text after Eq. (1) contains a typo: "A and B A represent the user opinion" should be "A and B represent the user opinion and the consensus sentence," respectively.
  2. [Figures 3–5] The manuscript references Figs. 3, 4, and 5, but the figures themselves are not included in the text; only the captions are present. The reader cannot inspect the iteration-level curves or the elbow-point analysis, so the claims in §5.2 about ChatGPT 4.0 converging in at most three iterations cannot be verified from the submitted text.
  3. [Throughout] The third model is referred to inconsistently as "AI21 Jamba," "AI21 Jamba-Instruct," and "A21 Jamba" (e.g., Table 3). Please use one consistent name throughout.
  4. [§3.2.1] The paper states that ChatGPT 4.0 uses "approximately 1.76 trillion parameters" and cites reference [51]; this figure is not supported by OpenAI's published documentation and appears to be a hallucinated or misattributed specification. Please verify the model's parameter count or remove the claim.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the paper reports an empirical comparison, and the cosine-similarity surrogate is a stated validity assumption rather than a fitted input or self-referential derivation.

full rationale

The paper's central claim—that ChatGPT 4.0 achieves higher alignment with participant opinions and requires fewer iterations—is an empirical measurement, not a derivation. Cosine similarity is used as an evaluation metric defined between participants' initial opinions and the LLM's accepted consensus proposal (Section 3.3), and the comparison among models is based on observed values of that metric (Section 5.1). No parameter is fitted to a subset of data and then renamed as a prediction, and no load-bearing step reduces to an equation already assumed as the conclusion. The metric's validity as a surrogate for genuine consensus is an important limitation that the authors themselves acknowledge in Section 6 ('while Cosine Similarity is a popular metric for assessing consensus facilitation by LLMs, it may not fully capture the qualitative aspects of participant agreement'), but this is a construct-validity concern, not circularity. There are no self-citations by the authors, no imported uniqueness theorem, and no ansatz smuggled in via citation. The comparison also includes an independent behavioral outcome—number of iterations required for participants to accept a proposal—which is not defined in terms of cosine similarity. Therefore no circular step is present, and the appropriate score is 0.

Assumptions & free parameters 0 free parameters · 5 assumptions · 0 invented entities

The central evaluation rests on several domain assumptions about the validity of cosine similarity as a consensus proxy, the adequacy of USE embeddings for Greek, and the representativeness of the sample. No numeric free parameters are fitted, but the choice of metric and the lack of a human-facilitator baseline are the main uncharged premises.

assumptions (5)
  • domain assumption Cosine similarity between initial positions and final proposal is a valid surrogate for consensus quality.
    Section 3.3; the central evaluation metric assumes higher alignment means better facilitation, but it may reward echoing initial views.
  • domain assumption The LLM's selection of the five adaptive strategies is appropriate and unbiased.
    Section 3.1; differences in outcomes are attributed to model capability, not to prompt variations or strategy selection behavior.
  • domain assumption Universal Sentence Encoder embeddings adequately capture semantic similarity for Greek-language text.
    Section 3.3; USE is trained mainly on English, and the discussions were in Greek, so embedding quality is unvalidated.
  • domain assumption Participants' initial text responses accurately represent their true positions.
    Section 3.1; initial opinions are collected as free text, and no validation of this elicitation is provided.
  • domain assumption The 30 self-selected Informatics students and dyadic sessions are representative of group deliberation settings.
    Section 4; acknowledged in limitations, generalizability is limited.

how reviews work

0 comments
Cite this review

Pith. "Pith review of From Divergence to Consensus: Evaluating the Role of Large Language Models in Facilitating Agreement through Adaptive Strategies." pith.science (2026). https://pith.science/paper/S2LQPCE7

@misc{pith2026250315521,
  author       = {Pith},
  title        = {Pith review of: From Divergence to Consensus: Evaluating the Role of Large Language Models in Facilitating Agreement through Adaptive Strategies},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/S2LQPCE7}},
  note         = {Machine review of arXiv:2503.15521}
}
read the original abstract

Achieving consensus in group decision-making often involves overcoming significant challenges, particularly in reconciling diverse perspectives and mitigating biases that hinder agreement. Traditional methods relying on human facilitators are often constrained by scalability and efficiency, especially in large-scale, fast-paced discussions. To address these challenges, this study proposes a novel framework employing large language models (LLMs) as automated facilitators within a custom-built multi-user chat system. Leveraging cosine similarity as a core metric, this approach evaluates the ability of three state-of-the-art LLMs- ChatGPT 4.0, Mistral Large 2, and AI21 Jamba Instruct- to synthesize consensus proposals that align with participants' viewpoints. Unlike conventional techniques, the system integrates adaptive facilitation strategies, including clarifying misunderstandings, summarizing discussions, and proposing compromises, enabling the LLMs to iteratively refine consensus proposals based on user feedback. Experimental results demonstrate the superiority of ChatGPT 4.0, which achieves higher alignment with participant opinions, requiring fewer iterations to reach consensus compared to its counterparts. Moreover, analysis reveals the nuanced performance of the models across various sustainability-focused discussion topics, such as climate action, quality education, good health and well-being, and access to clean water and sanitation. These findings highlight the transformative potential of LLM-driven facilitation for improving collective decision-making processes and underscore the importance of advancing evaluation metrics and cross-cultural adaptability in future research.

Figures

Figures reproduced from arXiv: 2503.15521 by the authors.

Figure 1
Figure 1. Consensus Formation Workflow Participants then indicate their acceptance or rejection of the consensus proposal. In cases of rejection, the LLM collects feedback to understand the reasons, while participants who agree await the feedback process. The LLM considers the conversation history and user feedback to select one of five predefined strategies. Each strategy guides the LLM in creating a revised proposal using t… view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

73 extracted references · 48 canonical work pages

  1. [1]

    Hanson, M. P. (2005). Clues to Achieving Consensus: A Leader's Guide to Navigating Collaborative Problem Solving. R&L Education

    work page 2005

  2. [2]

    Neves, M.P. (2016). Consensus. In: ten Have, H. (eds) Encyclopedia of Global Bioethics. Springer, Cham. doi: https://doi.org/10.1007/978-3-319-09483-0_119

    work page doi:10.1007/978-3-319-09483-0_119 2016

  3. [3]

    (2023, November)

    Ding, S., & Ito, T. (2023, November). Self-agreement: a framework for fine-tuning language models to find agreement among diverse opinions. In Pacific Rim International Conference on Artificial Intelligence (pp. 298 -309). Singapore: Springer Nature Singapore. doi: https://doi.org/10.1007/978-981-99-7022-3_26

    work page doi:10.1007/978-981-99-7022-3_26 2023

  4. [4]

    P., & Wang, T

    Foroughi, M., de Andrade, B., Roders, A. P., & Wang, T. (2023). Public participation and consensus-building in urban planning from the lens of heritage planning: A systematic literature review. Cities, 135, 104235. doi: https://doi.org/10.1016/j.cities.2023.104235

    work page arXiv 2023

  5. [5]

    V., Weisswange, T

    Pham, T. V., Weisswange, T. H., & Hassenzahl, M. (2024, July). Embodied Mediation i n Group Ideation –A Gestural Robot Can Facilitate Consensus - Building. In Proceedings of the 2024 ACM Designing Interactive Systems Conference (pp. 2611-2632). doi: https://doi.org/10.1145/3643834.3660696

    arXiv 2024

  6. [6]

    Tversky, A., Kahneman, D., & Slovic, P. (1982). Judgment under uncertainty: Heuristics and biases (pp. 3-20)

    work page 1982

  7. [7]

    T., Yang, M., & Dow, S

    Liu, W., Xiao, S., Browne, J. T., Yang, M., & Dow, S. P. (2018). ConsensUs: Supporting multi -criteria group decisions by visualizing points of disagreement. ACM Transactions on Social Computing , 1(1), 1 -26. doi: https://doi.org/10.1145/3159649

    work page doi:10.1145/3159649 2018

  8. [8]

    Toma, C., & Butera, F. (2009). Hidden profiles and concealed information: Strategic information sharing and use in group decision making. Personality and Social Psychology Bulletin , 35(6), 793 -806. doi: https://doi.org/10.1177/0146167209333176

    work page doi:10.1177/0146167209333176 2009

Show all 73 references
  1. [9]

    Straus, D. (2002). How to make collaboration work: Powerful ways to build consensus, solve problems, and make decisions. Berrett-Koehler Publishers

    2002

  2. [10]

    Pound, D. I. A. N. A. (2015). Designing and facilitating consensus-building–keys to success. Conflicts in Conservation: Navigating towards Solutions; Redpath, SM, Gutierrez, RJ, Wood, KA, Young, JC, Eds, 240-256

    2015

  3. [11]

    Cowell, S., Begg, K., & Clift, R. (2006). Support for sustainable development policy decisions a case study from highway maintenance. The I nternational Journal of Life Cycle Assessment , 11, 29 -39. doi: https://doi.org/10.1065/lca2006.04.009

    2006 doi

  4. [12]

    (2024, May)

    Govers, J., Velloso, E., Kostakos, V., & Goncalves, J. (2024, May). AI -Driven Mediation St rategies for Audience Depolarisation in Online Debates. In Proceedings of the CHI Conference on Human Factors in Computing Systems (pp. 1-18). Doi: https://doi.org/10.1145/3613904.3642322

    2024

  5. [13]

    Dong, Y., Ding, S., & Ito, T. (2024). An Automated Multi-Phase Facilitation Agent Based on LLM. IEICE TRANSACTIONS on Information and Systems , 107(4), 426-433. doi: https://doi.org/10.1587/transinf.2023IHP0011

    2024 doi

  6. [14]

    G., Zamfirescu, C

    Filip, F. G., Zamfirescu, C. B., Ciurea, C., Filip, F. G., Zamfirescu, C. B., & Ciurea, C. (2017). Collaborative activities and methods. Computer-Supported Collaborative Decision-Making, 71-120. doi: https://doi.org/10.1007/978-3-319- 47221-8_3

    2017 doi

  7. [15]

    Ito, T., Hadfi, R., & Suzuki, S. (2022). An agent that facilitates crowd discussion: A crowd discussion support system based on an automated facilitation agent. Group Decision and Negotiation , 31(3), 621 -647. doi: https://doi.org/10.1007/s10726-021-09765-8

    2022 doi

  8. [16]

    Zhou, L., Farag, Y., & Vlachos, A. (2024). An LLM Feature -based Framework for Dialogue Constructiveness As sessment. arXiv preprint arXiv:2406.14760 . doi: https://doi.org/10.48550/arXiv.2406.14760

    2024 doi arXiv

  9. [17]

    W., Lopez -Lopez, E., Hechtlinger, S., Rahwan, Z., Aeschbach, S., Bakker, M

    Burton, J. W., Lopez -Lopez, E., Hechtlinger, S., Rahwan, Z., Aeschbach, S., Bakker, M. A., ... & Hertwig, R. (2024). How large language models can reshape collective intelligence. Nature Human Behaviour , 1 -13. doi: https://doi.org/10.1038/s41562-024-01959-9

    2024 doi

  10. [18]

    Abdelnabi, S., Gomaa, A., Sivaprasad, S., Schönherr, L., & Fritz, M. (2023). Llm- deliberation: Evaluating llms with interactive multi -agent negotiation games. arXiv preprint arXiv:2309.17234 . doi: https://doi.org/10.48550/arXiv.2309.17234

    2023 doi arXiv

  11. [19]

    Westermann, H., Savelka, J., & Benyekhlef, K. (2023). Llmediator: Gpt-4 assisted online dispute resolution. arXiv preprint arXiv:2307.16732 . doi: https://doi.org/10.48550/arXiv.2307.16732

    2023 doi arXiv

  12. [20]

    Xu, S., Wu, Z., Zhao, H., Shu, P., Liu, Z., Liao, W., ... & Li, X. (2024). Reasoning before comparison: LLM -enhanced semantic similarity metrics for domain specialized text analysis. arXiv preprint arXiv:2402. 11398. doi: https://doi.org/10.48550/arXiv.2402.11398

    2024 doi arXiv

  13. [21]

    Zhang, L., Jijo, K., Setty, S., Chung, E., Javid, F., Vidra, N., & Clifford, T. (2024). Enhancing Large Language Model Performance To A nswer Questions and Extract Information More Accurately. arXiv preprint arXiv:2402.01722 . doi: https://doi.org/10.48550/arXiv.2402.01722

    2024 doi arXiv

  14. [22]

    T., del Moral, M

    Chiclana, F., García, J. T., del Moral, M. J., & Her rera-Viedma, E. (2015). Analyzing consensus measures in group decision making. Procedia Computer Science, 55, 1000-1008. doi: https://doi.org/10.1016/j.procs.2015.07.103

    2015 doi

  15. [23]

    Bakhshi, F., & Ashtiani, M. (2024). An approach for reaching consensus in large- scale group decision -making focusing on dimension reduction. Complex & Intelligent Systems, 10(3), 4223-4251. doi: https://doi.org/10.1007/s40747-024- 01377-4

    2024 doi

  16. [24]

    Ingram, S., & Munzner, T. (2015). Dimensionality reduction for documents with nearest neighbor queries. Neurocomputing, 150, 557 -569. Doi: https://doi.org/10.1016/j.neucom.2014.07.073

    2015 doi

  17. [25]

    & Amiryar, R

    Haqbeen, J., Ito, T., Hadfi, R., Nishida, T., Sahab, Z., Sahab, S., ... & Amiryar, R. (2020, June). Promoting discussion with AI -based facilitation: Urban dialogue with Kabul city. In Proceedings of the 8th ACM Collective Int elligence, ACM Collective Intelligence Conference ...

    2020

  18. [26]

    Haqbeen, J., Sahab, S., & Ito, T. (2023). A Case Study on the Comparison of AI - facilitated Threaded Conversation versus Threaded Conversation. IIAI Letters on Informatics and Interdisciplinary Research , 4. doi: https://doi.org/10.52731/liir.v004.179

    2023 doi

  19. [27]

    Hadfi, R., Okuhara, S., Haqbeen, J., Sahab, S., Ohnuma, S., & Ito, T. (2023). Conversational agents enhance women's contribution in online debates. Scientific Reports, 13(1), 14534. doi: https://doi.org/10.1038/s41598-023-41703-3

    2023 doi

  20. [28]

    Sahab, S., Haqbeen, J., & Ito, T. (2024). Conversational AI as a Facilitator Improves Participant Engagement and Problem -Solving in Online Discussion: Sharing Evidence from Five Cities in Afghanistan. IEICE TRANSACTIONS on Information and Systems , 107(4), 434 -442. doi: http...

    2024 doi

  21. [29]

    Kunz, W., & Rittel, H. W. (1970). Issues as elements of information systems (Vol. 131, p. 14). Oakland: Institute of Urban and Regional Development, University of California

    1970

  22. [30]

    A., & Ito, T

    Sahab, S., Haqbeen, J. A., & Ito, T. (2024). Comparative analysis of AI facilitator impact in online discussions: A cross-cultural study. In INFORMATIK 2024 (pp. 715-725). Gesellschaft für Informatik eV. doi: https://doi.org/10.18420/inf2024_59

    2024 doi

  23. [31]

    (2024, May)

    Dong, Y. (2024, May). The Multi -agent System based on LLM for Online Discussions. In Proceedings of the 23rd International Conference on Autonomous Agents and Multiagent Systems (pp. 2731 -2733). doi: https://doi.org/10.5555/3635637.3663269

    2024

  24. [32]

    (2024, June)

    Nomura, M., Ito, T., & Ding, S. (2024, June). Towards Collaborative Brain - storming among Humans and AI Agents: An Implementation of the IBIS -based Brainstorming Support System with Multiple AI Agents. In Proceedings of the ACM Collective Intelligence Conference (pp. 1 -9). ...

    2024

  25. [33]

    Song, T., Tan, Y., Zhu, Z., Feng, Y., & Lee, Y. C. (2024). Multi-Agents are Social Groups: Investigating Social Influence of Multiple Agents in Human -Agent Interactions. arXiv preprint arXiv:2411.04578 . doi: https://doi.org/10.48550/arXiv.2411.04578

    2024 doi

  26. [34]

    H., Bakker, M

    Tessler, M. H., Bakker, M. A., Jarrett, D., Sheahan, H., Chadwick, M. J., Koster, R., ... & Summerfield, C. (2024). AI can help humans find common ground in democratic deliberati on. Science, 386(6719), eadq2852. doi: https://doi.org/10.1126/science.adq2852

    2024 doi

  27. [35]

    J., Cabrerizo, F

    Pérez, I. J., Cabrerizo, F. J., Alonso, S., Dong, Y. C., Chiclana, F., & Herrera - Viedma, E. (2018). On dynamic c onsensus processes in group decision making problems. Information Sciences , 459, 20 -35. doi: https://doi.org/10.1016/j.ins.2018.05.017

    2018 doi

  28. [36]

    (2022, May)

    Hadfi, R., & Ito, T. (2022, May). Augmented Democratic Deliberation: Can Conversational Agents Boost Deliberation in Social Media?. In proceedings of the 21st international conference on autonomous agents and multiagent systems (pp. 1794-1798). doi: https://doi.org/10.5555/353...

    2022

  29. [37]

    de Rooij, A., Broek, S. V. D., Bouw, M., & de Wit, J. (2024). Co -Creating with a Robot Facilitator: Robot Expressions Cause Mood Contagion Enhancing Collaboration, Satisfaction, and Performance. International J ournal of Social Robotics, 1-20. Doi: https://doi.org/10.1007/s12...

    2024 doi

  30. [38]

    Kim, S., Eun, J., Seering, J., & Lee, J. (2021). Moderator chatbot for deliberative discussion: Effects of discussion structure and discussant facilitation. Proceedings of the ACM on Human -Computer Interaction, 5(CSCW1), 1-26. doi: https://doi.org/10.1145/3449161

    2021 doi

  31. [39]

    Sahab, S., Haqbeen, J., & Ito, T. (2023). Facilitating Collaborative Consensus Building in Web-based Discussion through Collective Task-Based Roles: A Case Study. IIAI Letters on Informatics and Interdisciplinary Research , 4. doi: https://doi.org/10.52731/liir.v004.180

    2023 doi

  32. [40]

    L., Zobel, C., Osorio, J., Cobb, T

    Bosch, D., Pease, J., Wolfe, M. L., Zobel, C., Osorio, J., Cobb, T. D., & Evanylo, G. (2012). Community DECISIONS: Stakeholder focused watershed planning. Journal of environmental management , 112, 226 -232, doi: https://doi.org/10.1016/j.jenvman.2012.07.031

    2012 doi

  33. [41]

    J., Cruze, E., Fox, D., Roy, S., Bakewell, T., & Coleman, C

    Dworetzky, B., Paladino, M. J., Cruze, E., Fox, D., Roy, S., Bakewell, T., & Coleman, C. (2024). Family Engagement in Systems Change: Use of a New Assessment Tool in Quality Improvement. Pediatrics, 153(Supplement 1). doi: https://doi.org/10.1542/peds.2023-063424D

    2024 doi

  34. [42]

    (2024, May)

    Chosokabe, M., Koie, S., & Oyamada , Y. (2024, May). Examining the Effect of ChatGPT on Small Group Ideation Discussions. In International Conference on Group Decision and Negotiation (pp. 142 -153). Cham: Springer Nature Switzerland. doi: https://doi.org/10.1007/978-3-031-59373-4_12

    2024 doi

  35. [43]

    Nieto-Romero, M., Milcu, A., Leventon, J., Mikulcak, F., & Fischer, J. (2016). The role of scenarios in fostering collective action for sustainable development: Lessons from central Romania. Land Use Policy , 50, 156 -168. doi: https://doi.org/10.1016/j.landusepol.2015.09.013

    2016 doi

  36. [44]

    Opricovic, S. (2009). A compromise solution in water resources planning. Water resources management, 23, 1549-1561. doi: https://doi.org/10.1007/s11269-008- 9340-y

    2009 doi

  37. [45]

    Basuki, A. (2016). Sustainable strategies selection in SMEs using MCDM approach. In MATEC Web of Conferences (Vol. 58, p. 0200 7). EDP Sciences. doi: https://doi.org/10.1051/matecconf/20165802007

    2016

  38. [46]

    J., Hanspach, J., Abson, D

    Kettenburg, A. J., Hanspach, J., Abson, D. J., & Fischer, J. (2018). From disagreements to dialogue: unpacking the G olden Rice debate. Sustainability Science, 13, 1469-1482. doi: https://doi.org/10.1007/s11625-018-0577-y

    2018 doi

  39. [47]

    M., Sourav, S., & Tamás, O

    Easin, A. M., Sourav, S., & Tamás, O. (2024, September). An intelligent llm - powered personalized assistant for digital banking using langgraph and chain of thoughts. In 2024 IEEE 22nd Jubilee International Symposium on Intelligent Systems and Informatics (SISY) (pp. 625-630). IEEE

    2024

  40. [48]

    Vaswani, A. (2017). Attention is all you need. Advances in Neural Information Processing Systems

    2017

  41. [49]

    Hariri, W. (2023). Unlocking the potential of ChatGPT: A comprehensive exploration of its applications, advantages, limitations, and future directions in natural language processing. arXiv preprint arXiv:2 304.02017. doi: https://doi.org/10.48550/arXiv.2304.02017

    2023 doi arXiv

  42. [50]

    So what if ChatGPT wrote it?

    Dwivedi, Y. K., Kshetri, N., Hughes, L., Slade, E. L., Jeyaraj, A., Kar, A. K., ... & Wright, R. (2023). Opinion Paper:“So what if ChatGPT wrote it?” Multidisciplinary perspectives on opportunities, challenges and implications of generative conversational AI for research, prac...

    2023

  43. [51]

    Hao, X., Demir, E., & Eyers, D. (2024). Exploring collaborative decision-making: A quasi-experimental study of human and Generative AI interaction. Technology in Society, 78, 102662. doi: https://doi.org/10.1016/j.techsoc.2024.102662

    2024

  44. [52]

    I., & Tselikas, N

    Roumeliotis, K. I., & Tselikas, N. D. (2023). Chatgpt and open -ai models: A preliminary review. Future Internet , 15(6), 192. doi: https://doi.org/10.3390/fi15060192

    2023 doi

  45. [53]

    Mistral AI. (n.d.). Mistral Large: Introducing our advanced model . Retrieved January 7, 2025, from https://mistral.ai/news/mistral-large-2407/

    2025

  46. [54]

    T., Rademaker, A., Lionti, K., Giro, R., Lima, G., Fiorini, S.,

    da Silva, V. T., Rademaker, A., Lionti, K., Giro, R., Lima, G., Fiorini, S., ... & Steiner, M. (2024). Automated, LLM enabled extraction of synthesis details for reticular materials from scientific literature. arXiv preprint arXiv:2411.03484 . doi: https://doi.org/10.48550/arX...

    2024 doi arXiv

  47. [55]

    Pawitan, Y., & Holmes, C. (2024). Confidence in the Reasoning of Large Language Models. arXiv preprint arXiv:2412.15296 . doi: https://doi.org/10.48550/arXiv.2412.15296

    2024 doi arXiv

  48. [56]

    Kanepajs, A., Ivanov, V., & Moulange, R. (2024). Towards Safe Multilingual Frontier AI. arXiv preprint arXiv:2409.13708 . doi: https://doi.org/10.48550/arXiv.2409.13708

    2024 doi arXiv

  49. [57]

    Z., Collins, K

    Qiu, Z., Liu, W., Feng, H., Liu, Z., Xiao, T. Z., Collins, K. M., ... & Schölkopf, B. (2024). Can Large Language Models Understand Symbolic Graphics Programs?. arXiv preprint arXiv:2408.08313. doi: https://doi.org/10.48550/arXiv.2408.08313

    2024 doi arXiv

  50. [58]

    Q., Sablayrolles, A., Mensch, A., Bamford, C., Chaplot, D

    Jiang, A. Q., Sablayrolles, A., Mensch, A., Bamford, C., Chaplot, D. S., Casas, D. D. L., ... & Sayed, W. E. (2023). Mistral 7B. arXiv preprint arXiv:2310.06825. doi: https://doi.org/10.48550/arXiv.2310.06825

    2023 doi arXiv

  51. [59]

    AI21 Labs. (n.d.). Announcing Jamba-Instruct. Retrieved January 7, 2025, from https://www.ai21.com/blog/announcing-jamba-instruct

    2025

  52. [60]

    & Shoham, Y

    Team, J., Lenz, B., Arazi, A., Bergman, A., Manevich, A., Peleg, B., ... & Shoham, Y. (2024). Jamba-1.5: Hybrid transformer-mamba models at scale. arXiv preprint arXiv:2408.12570. doi: https://doi.org/10.48550/arXiv.2408.12570

    2024 doi arXiv

  53. [61]

    & Shoham, Y

    Lieber, O., Lenz, B., Bata, H., Cohen, G., Osin, J., Dalmedigos, I., ... & Shoham, Y. (2024). Jamba: A hybrid transformer-mamba language model. arXiv preprint arXiv:2403.19887. doi: https://doi.org/10.48550/arXiv.2403.19887

    2024 doi arXiv

  54. [62]

    (2019, December)

    Shahmirzadi, O., Lugowski, A., & Younge, K. (2019, December). Text similarity in vector space models: a comparative study. In 2019 18th IEEE international conference on machine learning and applications (ICMLA) (pp. 659-666). IEEE. doi: https://doi.org/10.1109/ICMLA.2019.00120

    2019

  55. [63]

    (2008, April)

    Huang, A. (2008, April). Similarity measures for text document clustering. In Proceedings of the sixth new zealand computer science research student conference (NZCSRSC2008), Christchurch, New Zealand (Vol. 4, pp. 9-56)

    2008

  56. [64]

    Li, B., & Han, L. (2013). Distance weighted cosine similarity measure for text classification. In Intelligent Data Engineering and Automated Learning –IDEAL 2013: 14th International Conference, IDEAL 2013, Hefei, China, October 20-23,

    2013

  57. [65]

    M., & Ye, L

    Kalhori, H., Alamdari, M. M., & Ye, L. (2018). Automated algorithm for impact force identification using cosine similarity searching. Measurement, 122, 648 -

    2018

  58. [66]

    (2023, September)

    Mishra, H., & Soundarajan, S. (2023, September). BalancedQR: A Framework for Balanced Query Recommendation. In Joint European Conference on Machine Learning and Knowledge Discovery in Databases (pp. 420-435). Cham: Springer Nature Switzerland. doi: https://doi.org/10.1007/978-...

    2023 doi

  59. [67]

    TensorFlow. (n.d.). Semantic similarity with TF Hub Universal E ncoder. Retrieved January 7, 2025, https://www.tensorflow.org/hub/tutorials/semantic_similarity_with_tf_hub_univ ersal_encoder

    2025

  60. [68]

    Sakamoto, M., Yamaguchi, E., Yamamoto, T., & Wakabayashi, K. (2021). An intervention study on students’ decision -making towards consensus building on socio-scientific issues. International Journal of Science Education , 43(12), 1965-

    2021

  61. [69]

    & LokmicTomkins, Z

    LevettJones, T., Catling, C., Cheer, S., Fields, L., Foster, A., Maguire, J., ... & LokmicTomkins, Z. (2024). Achieving consensus on the essential knowledge and skills needed by nursing students to promote planetary health and sustainable healthcare: A Delphi study. Journal of...

    2024 doi

  62. [70]

    Thorndike, R. L. (1953). Who belongs in the family?. Psychometrika, 18(4), 267- 276

    1953

  63. [657]

    doi: https://doi.org/10.1016/j.measurement.2018.01.016

    2018 doi

  64. [1983]

    doi: https://doi.org/10.1080/09500693.2021.1947541

    2021

  65. [2013]

    611 -618)

    Proceedings 14 (pp. 611 -618). Springer Berlin Heidelberg. doi: https://doi.org/10.1007/978-3-642-41278-3_74

    doi

Pith tools

Reviewed August 9, 2026 · model on record in the stance chip above.