Towards Effective Multidisciplinary Health and HCI Teams based on AI Framework

Diego G\'omez-Zar\'a; Mohammed Almutairi

REVIEW 4 major objections 5 minor 23 references

Towards Effective Multidisciplinary Health and HCI Teams based on AI Framework

T0 review · 4 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read Team formation in health-HCI research can be cast as a sequential choice over non-dominated teams, learned by a contextual bandit from user preferences.

desk verdict A clearly written workshop pitch that restates the authors' own framework and overclaims 'ensures engaged teams' with zero supporting evidence; fine for a discussion slot, not for peer review as a research paper. read the letter →

arxiv 2506.05226 v1 pith:RGCYGF5A submitted 2025-06-05 cs.HC

Mohammed Almutairi , Diego Gómez-Zará This is my paper

classification cs.HC

keywords multidisciplinaryteamformationhealth-HCIcollaborationNSGA-IIcontextualbanditsuserpreferencesnon-dominatedcompositionscross-disciplinary

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 argues that assembling multidisciplinary health and HCI research teams is itself a solvable design problem, not just an organizational accident. The authors propose an AI-driven team formation system that treats the task as a sequential decision process: an optimization stage precomputes a set of non-dominated (Pareto-optimal) team compositions balancing diversity, familiarity, and expertise, and a user-interaction stage lets the system learn an individual's preferences from repeated selections until an optimal team is recommended. The goal is a user-centric tool that reduces the design-reality gaps and collaboration failures common in health-HCI projects. If the framework works, teams would be formed by objective criteria plus personal preference, rather than by similarity-driven convenience.

What carries the argument

The central machinery is a two-stage pipeline. First, NSGA-II, a fast elitist multi-objective genetic algorithm, searches over possible team compositions and returns the non-dominated frontier, deliberately preventing bias such as favoring expert-heavy teams. Second, a contextual bandit (drawing on lil'UCB-style exploration) learns user preferences from iterative selections among this frontier, adjusting the recommendation set to highlight preferred teams and remove less desired ones. The sequential decision formulation connects these stages: each user interaction is a context that refines the team recommendation, and the optimal team is the composition at the end of this adaptive loop.

What would settle it

Run a controlled study where health-HCI teams are assembled by this system, by random matching, and by preference-only matching, then measure collaboration quality (e.g., task completion, member satisfaction, retention, output quality). If teams chosen through the bandit loop do not outperform the baselines, or if users' shown preferences do not correlate with post-formation team performance, the central claim that the approach ensures more engaged and satisfied teams would be falsified.

Watch

Extended reading notes

Core claim

The central claim is that optimal multidisciplinary teams for health-focused HCI research can be produced by combining multi-objective optimization with preference learning. An optimal team is defined as one where no objective is dominated by another: the team is well-connected (familiarity), drawn from diverse areas of expertise, and matched to the project's requirements and the user's preferences. The system first generates a pool of non-dominated team compositions using NSGA-II, then presents these to the user and uses a contextual bandit to adaptively update which teams are highlighted, converging to the desired optimal team. The authors state the approach ensures more engaged and satisfied multidisciplinary teams collaborate.

Load-bearing premise

The framework assumes that preferences a user expresses by choosing among suggested teams are a good signal of what makes a team actually work well, and that diversity, familiarity, and expertise capture the real drivers of collaboration success in health-HCI research.

Editorial extensions

If this is right

If the framework is correct, health-HCI teams can be assembled before formation, with diversity and familiarity treated as explicit, tunable objectives instead of afterthoughts.
User-centric feedback would let each researcher steer team composition toward their own priorities while remaining inside the non-dominated set, so no objective is silently sacrificed.
The two-stage design separates what good teams look like from what this user prefers, making the preference-learning stage reusable across projects and users.
Objective pre-filtering could reduce failures attributed to design-reality gaps, because teams are matched to project requirements rather than to existing social ties alone.

Reading between the lines

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

A strong but untested implication is that revealed preferences from the bandit loop (which teams users click) are a reliable ground-truth signal for collaboration quality; a field study could test whether selections predict post-formation outcomes like satisfaction, retention, or publication success.
The framework could be extended to other multidisciplinary domains, such as engineering, education, or policy teams, since neither the NSGA-II nor the bandit machinery is health-specific; only the attribute definitions would change.
The paper does not report an evaluation; a direct extension would benchmark the system against random team assignment and against preference-only formation in controlled health-HCI collaborations.
Because NSGA-II's non-dominated set is computed once per candidate pool, the expensive search can be amortized, suggesting the approach is scalable to large rosters of researchers.

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.

Desk Editor's Note

read the letter

Here's the short version: this is a four-page workshop position paper that describes a team-formation framework already introduced in the authors' own prior work [3]. It makes a strong predictive claim in Section 4—"Our approach ensures that more engaged and satisfied multidisciplinary teams collaborate"—and provides no data, simulation, or user study to back it up. The motivation is real and the literature review is competent, but as a research contribution the paper is thin.

On the positive side, the authors identify a genuine problem: assembling effective multidisciplinary health/HCI teams is hard, and existing team-formation literature focuses mostly on post-formation dynamics. The two-stage pipeline—NSGA-II to precompute a Pareto set over diversity, familiarity, and expertise, and a contextual bandit to learn user preferences from iterative selections—is a reasonable use of established techniques. The paper is clearly written, appropriately short for a workshop, and transparent that the system comes from their prior work. The related-work coverage spans team science, organizational structure, and similarity-attraction, which is appropriate for a position piece.

The load-bearing problem is that the central claim does not follow from the described method. The bandit reward is user preference expressed as selections or ratings of candidate teams, but nothing ties that signal to actual post-formation engagement, satisfaction, or collaboration quality. The authors themselves note in Section 3 that preference-based formation can lead to homogeneous teams via similarity-attraction; the proposed loop still treats user preference as the terminal reward, so the 'ensures' language is unjustified. Second, the description is too high-level to assess correctness: no state space, action space, reward definition, or NSGA-II encoding is given, so "optimal team" is not a checkable statement. Third, there is no new result here—the framework is reference [3] restated, and the current paper adds only the health/HCI framing. There are also small issues: the URL in reference [3] is misspelled ('teamfromation.me'), and the ChatGPT citation in [17] feels tangential.

Who is this for? Someone thinking about submitting to a CHI workshop or looking for a strawman in a reading group on AI-assisted team formation. As a full paper, it would not deserve a serious referee in its current form because the central claim is a prediction without evidence. If the authors add even a small simulation or pilot user study that connects preference feedback to team outcomes, the framework could become worth engaging with. For now, I would not accept it for peer review beyond the workshop-discussion level, and I would not cite it.

Referee Report

4 major / 5 minor

Summary. This short paper proposes an AI-driven framework for forming multidisciplinary teams in health-focused HCI research. The authors motivate the problem with challenges from the team science and HCI-health literature, then sketch a two-stage system: stage one uses NSGA-II to generate a set of non-dominated team compositions over objectives such as diversity, familiarity, and expertise; stage two uses a contextual bandit to iteratively learn user preferences from selections among those compositions. The paper claims that this approach 'ensures that more engaged and satisfied multidisciplinary teams collaborate' (Section 4), but it presents no formal derivation, simulation, dataset, user study, or any other empirical evidence supporting this claim. The technical description is high-level, and the core algorithm is attributed to the authors' own prior work [3], which is not reproduced here.

Significance. If the claimed framework were shown to reliably improve collaboration outcomes in health-HCI teams, it would address a real and well-documented problem: multidisciplinary teams frequently struggle with divergent goals, methods, and incentive structures. The paper usefully assembles relevant literature on team formation, similarity-attraction, and organizational barriers, and it identifies a plausible mechanism (precomputed Pareto-optimal compositions plus preference learning) worth investigating. However, the significance as it stands is limited because the central claim is a prediction without evidence. The paper provides no machine-checked proofs, no reproducible code, no parameter-free derivation, and no falsifiable empirical result; it is essentially a proposal that delegates its technical content to a self-citation. The contribution is therefore not assessable on the merits, and the strong 'ensures' claim goes far beyond what the manuscript can support.

major comments (4)

[Section 4, 'Our approach ensures ...'] The central claim that the approach 'ensures that more engaged and satisfied multidisciplinary teams collaborate' is unsupported. The manuscript contains no user study, no simulation, no offline evaluation, and no formal argument establishing that the contextual bandit's reward signal (user selections or ratings of proposed team compositions) is correlated with post-formation team engagement, satisfaction, or collaboration success. As written, the 'optimal team' is optimal only with respect to the three proxy objectives and the user's expressed preferences, not with respect to any measured team outcome. A concrete test would be a randomized comparison of teams formed by this system versus a baseline (e.g., expert-assembled or preference-only teams) on validated collaboration and satisfaction instruments; absent such evidence, the 'ensures' claim should be removed or explicitly reframed as a hypothesis.
[Section 4 and Figure 1] The algorithmic description is too underspecified to assess correctness or reproducibility. The manuscript does not define the state space, action space, reward function, or bandit algorithm beyond naming references [9] and [13]; it does not define the three objectives (diversity, familiarity, expertise) formally; and it does not state how NSGA-II's population size, generations, or crossover/mutation operators are configured. Moreover, the definition of 'optimal team' at the end of Section 3 is effectively Pareto optimality ('no objective is dominated by another'), which is a mathematical property of a set of solutions, not a guarantee of team success. The authors should either provide the formal details of the framework or clearly state that this is a position paper with no new technical machinery.
[Section 3 versus Section 4] There is an internal tension between the paper's warning about similarity-attraction and its proposed reliance on user preferences. Section 3 correctly notes that teams formed solely on member preferences may become homogeneous and miss cultural nuances, citing similarity-attraction theory [16]. Yet Section 4's user-centric loop learns from user selections of team compositions and adapts to 'highlight more preferred teams,' which could reinforce the very homogeneity the paper warns against. The manuscript does not explain how the precomputed non-dominated set or the bandit's exploration strategy prevents the system from converging to teams that satisfy expressed preferences but lack diversity. This is a load-bearing issue because the claimed benefit depends on the system balancing diversity against preference, and no mechanism for that balance is specified.
[References [3] and [4]] The core framework is attributed to the authors' own prior work [3], which is not described in sufficient detail in this manuscript. The reader cannot verify that the current paper adds any independent derivation, implementation, or evaluation beyond a restatement of the earlier system. Similarly, reference [4] is cited as background but is a poster presentation with no associated technical substance. For a paper whose stated contribution is a framework, relying on an unpublished or non-reproducible self-citation for the central mechanism makes the contribution effectively unassessable. The authors should include either the full algorithmic details here or a substantive summary with formal notation, and they should clarify what is new in this manuscript relative to [3].

minor comments (5)

[Abstract and Section 1] The abstract and Section 1 use first-person singular ('As a Ph.D. student...', 'My research...') despite listing two authors; this should be harmonized to reflect joint authorship or explicit first-person acknowledgment of the student's perspective.
[Section 1, first sentence] The sentence 'My research on developing team compositions that involve multidisciplinary members from fields including education, academia, and health [4]' is a fragment; it lacks a main verb and should be rewritten.
[Reference [3]] The URL in reference [3] reads 'http://teamfromation.me', which appears to be a typo for 'teamformation.me'; please verify the URL.
[Section 4] The citation to 'lil\'ucb' [13] appears without any context explaining how that bandit algorithm is used (e.g., expected arm, exploration parameter, context vector); adding one or two sentences describing the learning update would clarify the proposal.
[CCS Concepts] The CCS Concepts line contains a semicolon-separated list that is difficult to parse; consider formatting each concept on its own line following ACM guidelines.

Circularity Check

2 steps flagged · score 6.0 of 10

The system's promised outcome ('more engaged and satisfied teams') is the optimization objective itself, and the proposed solution is the authors' own prior work [3] with no external validation.

self citation load bearing [Section 4, heading and opening paragraph]
"4 PROPOSED SOLUTION: AI-DRIVEN TEAM FORMATION SYSTEM [3]. Our proposed solution is designed to address the challenges of assembling multidisciplinary teams by utilizing team members' information to recommend the optimal team. ... Our approach ensures that more engaged and satisfied multidisciplinary teams collaborate."

The heading identifies the proposed solution as the authors' own prior work [3] (Almautairi and Gómez-Zará 2024). The paper restates that framework (NSGA-II non-dominated teams plus contextual bandit user feedback) and asserts its benefits, but it provides no derivation, simulation, user study, or external comparison. The central claim that the system ensures engaged and satisfied teams is therefore supported only by the authors' unverified self-citation, not by any independent evidence.
self definitional [Section 3 (definition of optimal team) and Section 4 (claimed outcome)]
"We define an optimal team in HCI and Health as one that achieves an optimal balance between the desired individual attributes of team members and success factors, such that no objective is dominated by another, aiming to create a team that is well-connected and formed from diverse areas of expertise. ... Our approach ensures that more engaged and satisfied multidisciplinary teams collaborate."

The paper defines 'optimal team' as a non-dominated balance over proxy attributes, and the system achieves that balance by learning user preferences from iterative selections. The conclusion 'ensures more engaged and satisfied teams' follows only if preference feedback is the same construct as engagement and satisfaction, which the paper never establishes. Section 3 itself warns that teams formed solely from preferences may be homogeneous and overlook cultural nuances, so the claimed outcome is built into the objective rather than derived from evidence.

full rationale

This is a short workshop position paper, so absence of a full evaluation is not by itself circularity. However, the load-bearing claim is the final sentence of Section 4, which asserts that the approach 'ensures that more engaged and satisfied multidisciplinary teams collaborate.' The system's reward signal is the user's expressed preference over candidate teams; no downstream measure of actual collaboration success is defined, measured, or cited. The 'optimal team' is optimal only with respect to the non-dominated objectives and the learned preferences, so the conclusion is an assumption built into the optimization target, not a prediction about real-world team outcomes. The paper's own Section 3 cautions that preference-based formation can produce homogeneous teams via similarity-attraction, undercutting the premise that preference maximization equals collaboration quality. Compounding this, Section 4's heading labels the proposed solution as the authors' own prior work [3], and the framework is described only by reference to that self-citation; there is no independent derivation in this paper. For these reasons, the central claim partially reduces to the system's own objective and to the authors' unverified prior work, giving a circularity score of 6.

Assumptions & free parameters 0 free parameters · 4 assumptions · 0 invented entities

The framework is built on domain assumptions about the measurability of team attributes, the validity of diversity/familiarity as objectives, the reliability of user feedback, and the appropriateness of the chosen algorithms. No free parameters are defined because no quantitative model is specified, and no new entities are introduced.

assumptions (4)

domain assumption Individual attributes and preferences of team members can be quantitatively represented and optimized.
The entire framework depends on encoding team member characteristics as optimization variables, but the paper never defines these attributes or how they are measured (Section 4).
domain assumption Diversity and familiarity are sufficient and reliable proxies for team effectiveness in health and HCI collaboration.
The paper asserts that balancing diversity and familiarity yields optimal teams, citing theories like similarity-attraction, but provides no empirical evidence linking these objectives to actual collaborative outcomes in this domain (Sections 3 and 4).
domain assumption User feedback during iterative selection is a valid signal for learning team composition preferences.
The human-in-the-loop mechanism assumes users can meaningfully compare and choose among non-dominated teams, and that their choices reflect true preferences that improve team quality (Section 4, Figure 1).
domain assumption NSGA-II and lil'UCB are appropriate algorithms for this problem setting.
The paper assumes these established algorithms will work well when combined for team formation, but provides no simulation or theoretical justification for the combination (Section 4).

how reviews work

0 comments

Cite this review

Pith. "Pith review of Towards Effective Multidisciplinary Health and HCI Teams based on AI Framework." pith.science (2026). https://pith.science/paper/RGCYGF5A

@misc{pith2026250605226,
  author       = {Pith},
  title        = {Pith review of: Towards Effective Multidisciplinary Health and HCI Teams based on AI Framework},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/RGCYGF5A}},
  note         = {Machine review of arXiv:2506.05226}
}

read the original abstract

As a Ph.D. student with a diverse background in both public and private sectors, I have encountered numerous challenges in cross-disciplinary and multi-stakeholder team projects. My research on developing team compositions that involve multidisciplinary members from fields including education, academia, and health. Along with my advisor, we are focused on exploring how HCI can help individuals assemble more effective teams. This effort involves developing socio-technical systems that guide and inform individuals of the potential teams that they can assemble. We employ state-of-the-art algorithms that prioritize inclusion among team members from diverse areas of expertise and familiarity between the team members. Our goal for attending this workshop is to engage in meaningful dialogues with scholars and researchers, leveraging these interactions to refine our approach to building an AI-driven team composition system to foster effective, interdisciplinary collaboration in health-focused HCI research.

Figures

Figures reproduced from arXiv: 2506.05226 by the authors.

**Figure 1.** The figure presents an AI framework for forming multidisciplinary teams, structured in two main stages. The initial stage [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

23 extracted references · 22 canonical work pages

[3]

Mohammed Almautairi and Diego Gómez-Zará. 2024. Optimizing Team Formation Dynamics with NSGA-II and User Feedback: A Contextual Bandit Approach. (2024). http://teamfromation.me

work page 2024
[17]

Malik Sallam. 2023. ChatGPT utility in healthcare education, research, and practice: systematic review on the promising perspectives and valid concerns. In Healthcare, Vol. 11. MDPI, 887

work page 2023
[9]

Kalyanmoy Deb, Amrit Pratap, Sameer Agarwal, and TAMT Meyarivan. 2002. A fast and elitist multiobjective genetic algorithm: NSGA-II. IEEE transactions on evolutionary computation 6, 2 (2002), 182–197

2002
[13]

Kevin Jamieson, Matthew Malloy, Robert Nowak, and Sébastien Bubeck. 2014. lil’ucb: An optimal exploration algorithm for multi-armed bandits. In Conference on Learning Theory . PMLR, 423–439

work page 2014
[16]

R Matthew Montoya and Robert S Horton. 2013. A meta-analytic investigation of the processes underlying the similarity-attraction effect. Journal of Social and Personal Relationships 30, 1 (2013), 64–94

work page 2013
[4]

Mohammed Almutairi and Diego Gómez-Zará. 2023. AI-Enhanced Team Formation in Online Learning Environments. Poster Presentation. https://mohammedkalmutairi.github.io/UCB_Poster/

work page 2023
[1]

Elena Agapie, Shefali Haldar, and Sharmaine G Poblete. 2022. Using HCI in Cross-Disciplinary Teams: A Case Study of Academic Collaboration in HCI-Health Teams in the US Using a Team Science Perspective. Proceedings of the ACM on Human-Computer Interaction 6, CSCW2 (2022), 1–35

work page 2022
[2]

Walid Al-Ahmad, Khalid Al-Fagih, Khalid Khanfar, Khalid Alsamara, Saleem Abuleil, and Hani Abu-Salem. 2009. A taxonomy of an IT project failure: root causes. International Management Review 5, 1 (2009), 93

work page 2009

Show all 23 references

[5]

Leonidas Anthopoulos, Christopher G Reddick, Irene Giannakidou, and Nikolaos Mavridis. 2016. Why e-government projects fail? An analysis of the Healthcare. gov website. Government information quarterly 33, 1 (2016), 161–173

2016
[6]

Garen Asatourian, Kingsley Maramot, Diana Mejia Reyes, and Javad Seif. 2021. A Multidisciplinary Team Formation Problem for Projects in the Aerospace Industry. In IIE Annual Conference. Proceedings. Institute of Industrial and Systems Engineers (IISE), 914–919. 3 CHI ’24, May ...

2021
[7]

R Meredith Belbin. 2011. Management teams: Why they succeed or fail. Human Resource Management International Digest 19, 3 (2011)

2011
[8]

Massimo Bergami and Richard P Bagozzi. 2000. Self-categorization, affective commitment and group self-esteem as distinct aspects of social identity in the organization. British journal of social psychology 39, 4 (2000), 555–577

2000
[10]

William M Foster and Marvin Washington. 2009. Organizational structure and home team performance. Team Performance Management: An International Journal 15, 3/4 (2009), 158–171

2009
[11]

Roger Guimera, Brian Uzzi, Jarrett Spiro, and Luis A Nunes Amaral. 2005. Team assembly mechanisms determine collaboration network structure and team performance. Science 308, 5722 (2005), 697–702

2005
[12]

J Richard Hackman and Nancy Katz. 2010. Group behavior and performance. Handbook of social psychology 2 (2010), 1208–1251

2010
[14]

Molly Kilcullen, Jennifer Feitosa, and Eduardo Salas. 2022. Insights from the virtual team science: Rapid deployment during COVID-19. Human Factors 64, 8 (2022), 1429–1440

2022
[15]

Marcel Maurer, Norbert Bach, and Simon Oertel. 2022. Forced to go virtual. Working-from-home arrangements and their effect on team communication during COVID-19 lockdown. German Journal of Human Resource Management 36, 3 (2022), 238–269

2022
[18]

Ahmad Sedaghat. 2018. Factors affecting the team formation and work in project based learning (PBL) for multidisciplinary engineering subjects. Journal of Problem Based Learning in Higher Education 6, 2 (2018)

2018
[19]

Greg L Stewart. 2006. A meta-analytic review of relationships between team design features and team performance. Journal of management 32, 1 (2006), 29–55

2006
[20]

Daniel Stokols, Kara L Hall, Richard P Moser, Annie X Feng, Shalini Misra, and Brandie K Taylor. 2010. Cross-disciplinary team science initiatives: Research, training, and translation. na

2010
[21]

Christian Tröster, Ajay Mehra, and Daan van Knippenberg. 2014. Structuring for team success: The interactive effects of network structure and cultural diversity on team potency and performance. Organizational Behavior and Human Decision Processes 124, 2 (2014), 245–255

2014
[22]

Bruce W Tuckman. 1965. Developmental sequence in small groups. Psychological bulletin 63, 6 (1965), 384

1965
[23]

Ilze Zigurs. 2003. Leadership in virtual teams: Oxymoron or opportunity? Organizational dynamics (2003). 4

2003

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.

[1] [3]

Mohammed Almautairi and Diego Gómez-Zará. 2024. Optimizing Team Formation Dynamics with NSGA-II and User Feedback: A Contextual Bandit Approach. (2024). http://teamfromation.me

work page 2024

[2] [17]

Malik Sallam. 2023. ChatGPT utility in healthcare education, research, and practice: systematic review on the promising perspectives and valid concerns. In Healthcare, Vol. 11. MDPI, 887

work page 2023

[3] [9]

Kalyanmoy Deb, Amrit Pratap, Sameer Agarwal, and TAMT Meyarivan. 2002. A fast and elitist multiobjective genetic algorithm: NSGA-II. IEEE transactions on evolutionary computation 6, 2 (2002), 182–197

2002

[4] [13]

Kevin Jamieson, Matthew Malloy, Robert Nowak, and Sébastien Bubeck. 2014. lil’ucb: An optimal exploration algorithm for multi-armed bandits. In Conference on Learning Theory . PMLR, 423–439

work page 2014

[5] [16]

R Matthew Montoya and Robert S Horton. 2013. A meta-analytic investigation of the processes underlying the similarity-attraction effect. Journal of Social and Personal Relationships 30, 1 (2013), 64–94

work page 2013

[6] [4]

Mohammed Almutairi and Diego Gómez-Zará. 2023. AI-Enhanced Team Formation in Online Learning Environments. Poster Presentation. https://mohammedkalmutairi.github.io/UCB_Poster/

work page 2023

[7] [1]

Elena Agapie, Shefali Haldar, and Sharmaine G Poblete. 2022. Using HCI in Cross-Disciplinary Teams: A Case Study of Academic Collaboration in HCI-Health Teams in the US Using a Team Science Perspective. Proceedings of the ACM on Human-Computer Interaction 6, CSCW2 (2022), 1–35

work page 2022

[8] [2]

Walid Al-Ahmad, Khalid Al-Fagih, Khalid Khanfar, Khalid Alsamara, Saleem Abuleil, and Hani Abu-Salem. 2009. A taxonomy of an IT project failure: root causes. International Management Review 5, 1 (2009), 93

work page 2009