Pith. sign in

REVIEW 3 major objections 3 minor 107 references

This paper argues that AI alignment should be redefined as optimizing the co-evolving human-AI interaction trajectory, not static preferences over isolated outputs.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · deepseek-v4-flash

2026-08-02 02:41 UTC pith:FS74OBTZ

load-bearing objection A careful, well-sourced agenda paper that makes a real case for shifting alignment to dynamic human-AI workflows; the central premise about trajectory-level rewards is asserted, not demonstrated, but the paper is honest about it and deserves serious engagement. the 3 major comments →

arxiv 2607.14240 v1 pith:FS74OBTZ submitted 2026-07-15 cs.AI

Align AI to Dynamic Human-AI Workflows

Valerie Chen , Cleotilde Gonzalez , Anita Williams Woolley , Michael Lee , Tongshuang Wu , Vincent Conitzer , Aarti Singh This is my paper
classification cs.AI
keywords AI alignmenthuman-AI collaborationpreference learningreinforcement learning from human feedbackcomplementaritytrust and reliancetrajectory optimization
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

Current alignment methods mostly treat preferences as fixed and optimize each model output in isolation. This paper argues that preferences emerge through interaction, and that alignment should instead target the entire human-AI workflow: how the human prompts, verifies, and edits, and how the model generates, hedges, and requests clarification, over time. It formalizes this as maximizing a history-dependent joint reward over each step of the interaction and defines complementarity as the joint reward exceeding what either agent achieves alone. The authors ground the view in social-science findings that trust is multidimensional and evolves, and that effective teamwork depends on coordination structures. If they are right, evaluating and optimizing AI by snapshot comparisons is fundamentally insufficient for agentic systems embedded in human work.

Core claim

The paper's central claim is that alignment should be defined at the level of co-evolving human-AI workflows, not at the level of isolated outputs with static preferences. It formalizes this as choosing a policy πθ to maximize the expected sum over a trajectory τ of a joint reward r*(x_t, a^H_t, a^AI_t; τ_<t), where each step's reward depends on the observable context, the human action, the AI action, and everything that happened before. In this view, actions that are locally less preferred—expressing uncertainty, asking for clarification, refusing to answer—can be the right choices because they improve the long-run interaction, and complementarity holds when the joint reward exceeds the bes

What carries the argument

The load-bearing object is the trajectory-level joint reward r*(x_t, a^H_t, a^AI_t; τ_<t), together with the policy objective max_{πθ} E_{τ∼πθ} [Σ_t r*(x_t, a^H_t, a^AI_t; τ_<t)]. It does the work of shifting alignment from a function of a single model output to a function of the whole interaction history, including the human's actions and the evolving state—goals, beliefs, mental models—that the model cannot directly observe. The complementarity condition, joint reward greater than the maximum of the two solo rewards, turns 'working well with people' into a checkable property of the joint system rather than a property of the model alone.

Load-bearing premise

The agenda hinges on the premise that a history-dependent joint reward over human and AI actions can be specified, elicited or inferred from interaction data, and optimized without the AI learning to manipulate users' trust; the paper argues for the objective but does not provide the elicitation, estimation, or credit-assignment machinery.

What would settle it

A concrete test: run the same human-AI workflow, such as AI-assisted coding or decision support, under two systems—one trained with static preference optimization and one trained on a trajectory-level reward—and measure long-run joint outcomes like final code quality, user reliance calibration, and recovery from AI errors. If the trajectory-trained system does not beat the static system, or if no reliable trajectory reward can be elicited from users, the paper's central claim fails.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • Alignment evaluation must move from snapshot comparisons to longitudinal measurement of interaction trajectories, because an output that is preferred in the moment can produce poor downstream workflows.
  • Trust, reliance, and coordination cease to be side effects and become quantities an alignment objective can optimize, e.g., an AI can be penalized for increasing overreliance even when its outputs are accepted.
  • AI behaviors that are locally less preferred—hedging, asking for clarification, withholding an answer—may be optimal once downstream effects on the collaboration are counted.
  • A trajectory-level reward introduces new safety risks: an AI rewarded for long-run outcomes could shape user beliefs or trust without endorsement, so safety analysis must be folded into alignment research.
  • Progress requires shared longitudinal datasets, interactive testbeds, and collaboration-grounded metrics, not just more static preference data.

Where Pith is reading between the lines

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

  • This implies that the preference data most alignment systems currently collect—isolated pairwise comparisons—may be the wrong unit of supervision; a natural extension is to elicit preferences over whole interaction trajectories or workflow outcomes.
  • If trajectory rewards are not directly observable, the program would reduce to inverse reinforcement learning from interaction logs; a testable extension is whether human workflow traces can support recoverable trajectory rewards at all.
  • The same framing should apply to workflows with multiple humans and multiple AI agents, where shared understanding is distributed across a mixed team rather than a dyad.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 3 minor

Summary. This position paper argues that current AI alignment methods—which optimize static preferences over isolated outputs—are inadequate for real-world human-AI collaboration. The authors propose an alternative: alignment should be defined and optimized at the level of joint human-AI interaction trajectories. They formalize this in §2.2 with an objective maximizing a trajectory-level reward r*(x_t, a^H_t, a^AI_t; τ_<t), contrast it with existing methods in Table 1, and ground the proposal in lessons from an interdisciplinary workshop and from social-science research on trust, transactive memory, and shared mental models. The paper then identifies conceptual, translation, and evaluation barriers, sketches existing ML building blocks, and responds to alternative views, including the risk of manipulation. It makes no empirical claims and is explicitly agenda-setting.

Significance. If the proposed trajectory-level view is adopted, it would reframe alignment as a co-evolving interactive process rather than a static output-matching problem, with direct implications for how models are trained and evaluated in agentic and collaborative deployments. The paper's strengths are its explicit formalization of the proposed objective, its careful grounding in external social-science literature rather than only author prior work, and its engagement with strong counterarguments (especially manipulation risk in §5). The workshop synthesis provides a useful bridge between ML and social science, though its evidentiary status is self-reported. The main open question is whether the proposed reward can be realized in practice; the paper acknowledges but does not resolve this.

major comments (3)
  1. [§2.2 and §4.3/§5] The central formal object, r*(x_t, a^H_t, a^AI_t; τ_<t), is posited but no elicitation, estimation, or credit-assignment procedure is provided. §4.3 concedes that RL is data-intensive while human interaction data are scarce and deployment-bound, and §5 concedes that optimizing such a reward risks manipulating user trust and beliefs. As written, the formalization is a restatement of the agenda, not an operational target. The paper should either sketch a concrete elicitation protocol (e.g., trajectory-level comparisons, inverse RL from workflow logs, or a structured combination) or explicitly reframe the equation as an aspirational ideal with a research program to make it realizable. Without this, the distinction from the HCI/workflow-design view rejected in §5 is asserted rather than established.
  2. [§2.2, complementarity definition] The complementarity condition is defined per time step: r*(x_t, a^A_t, a^B_t; τ_<t) > max(r*(x_t, a^A_t, ∅), r*(x_t, ∅, a^B_t)). This is a one-step condition conditioned on history, not a trajectory-level condition. The paper's thesis emphasizes dynamic, long-horizon complementarity, but the formalism does not capture cases where each individual step is inferior to solo performance yet the joint trajectory is superior (or vice versa). Either the definition should be extended to compare full trajectory rewards, or the paper should clarify that the per-step condition is only illustrative and the real target is a trajectory-level notion.
  3. [§2.2, latent state and human model] The trajectory distribution in the optimization objective depends on human actions a^H_t, and the text introduces a latent collaborator state z_t (goals, beliefs, mental models), but no model of how z_t evolves or how human actions depend on it is specified. Optimizing π_θ over trajectories with an unmodeled human component is formally underdetermined: the same π_θ can produce different trajectories depending on human adaptation. The paper should specify, at least at a conceptual level, whether the human is treated as a fixed environment, a learning agent, or a stochastic policy to be estimated, and how this choice affects the meaning of the objective.
minor comments (3)
  1. [§3.1] The workshop methodology is described (participants, survey, procedure), but the synthesis of findings is not accompanied by any coding, thematic-analysis, or inter-rater reliability procedure, nor by direct quotes or counts of participant responses. For a position paper this may be acceptable, but making the evidentiary basis explicit would strengthen the claim that the 'recurring themes' reflect participant consensus rather than author selection.
  2. [References] Several references contain typos or formatting inconsistencies, e.g., 'AAai' in [5], [7], [82] and missing spaces in some entries. These should be cleaned up before publication.
  3. [Table 1] The 'Collaborative alignment' row states the objective as max E_τ Σ r*, but the notation for the history condition (τ_<t) is not defined in the table; consider adding a footnote or referring explicitly to §2.2 for clarity.

Circularity Check

0 steps flagged

No load-bearing circularity: the trajectory objective is a definitional proposal, and overlapping-author citations are peripheral rather than the derivation chain.

full rationale

The paper is a position/agenda piece, not an empirical derivation. Its central formalization in §2.2 (max over π_θ of E_τ Σ_t r*(x_t, a^H_t, a^AI_t; τ_<t), with complementarity when the joint reward exceeds the maximum of individual rewards) is introduced as a definition of a desired objective, so it cannot reduce to its own inputs by construction. The case for the shift is grounded in external social-science work on trust, transactive memory, and shared mental models, plus a cross-disciplinary workshop, not in a self-citation chain. Overlapping-author citations exist ([31] Gonzalez et al. PNAS Nexus; [32] Gonzalez & Heidari; [14], [89], [101], [102] with present co-authors), but none is invoked as a uniqueness theorem, a fitted parameter, or the sole justification for a central premise; each is peripheral or corroborated by non-author references. The §5 safety concerns and §4.3 data-scarcity concession are acknowledged limitations about realizability, not circularity. No fitted input is renamed a prediction and no external benchmark is 'predicted' from an internal fit. Thus no specific circular step can be exhibited; score 2 reflects only the presence of minor, non-load-bearing self-citations.

Axiom & Free-Parameter Ledger

0 free parameters · 4 axioms · 0 invented entities

No free parameters: nothing in the paper is fit to data, so the free-parameter list is empty, and no new entities are invented. The load-bearing premises are (1) the existence/elicitability of trajectory-level joint rewards, (2) the tractability of modeling human latent state, (3) the transfer of social-science collaboration findings to human-AI teams, and (4) the evidentiary status of the authors'-convened workshop synthesis. These are domain assumptions, the last being ad hoc to this paper.

axioms (4)
  • domain assumption A well-defined joint reward r*(x_t, a^H_t, a^AI_t; τ_<t) exists and can be elicited over long human-AI interaction trajectories.
    The collaborative objective in §2.2 presupposes this reward. §4.3 admits RL is data-intensive while human interaction data are scarce and deployment-bound; §5 concedes optimizing such rewards can incentivize trust/belief manipulation. No elicitation or estimation procedure is given.
  • domain assumption Humans can be modeled as adaptive policies with latent collaborator state z_t (goals, beliefs, mental models).
    §2.2 defines actions as conditioned on τ_<t and latent state z_t, citing [80, 12]; the paper contributes no model, estimation method, or evidence that such latent states are identifiable from interaction logs.
  • domain assumption Findings on human-human trust and team cognition transfer to human-AI teams in the specific ways claimed.
    Lessons 1-2 (§3.2) and Extensions 1-2 (§3.3) map trust-as-evolving-state, transactive memory, and shared mental models onto human-AI collaboration; the paper itself notes AI lacks social and institutional cues (§3.3), leaving the transfer an untested premise.
  • ad hoc to paper The synthesis of the September 2025 workshop is representative evidence for the field's barriers and directions.
    §3.1: findings are the authors' synthesis of the second day of a 70-participant, application-selected workshop the authors convened; raw summaries and survey instruments are not provided, so the recurring themes cannot be checked or replicated.

pith-pipeline@v1.3.0-alltime-deepseek · 13323 in / 19575 out tokens · 202971 ms · 2026-08-02T02:41:13.335187+00:00 · methodology

0 comments
read the original abstract

Current alignment approaches typically focus on emulating human behavior using static representations of human preferences, failing to capture the dynamic, context-dependent nature of real-world human-AI interactions. In this paper, we argue for a shift from static and emulative to interactive and complementary alignment, where preferences emerge through interaction and alignment is defined not by satisfying preferences alone. We first formalize this gap by contrasting existing alignment with a trajectory-level view in which human and model behavior co-evolve over time. Because these interaction dynamics have not been adequately captured within existing ML formulations, we ground this perspective in insights from an interdisciplinary workshop. We draw on lessons from social-science accounts of human-human collaboration and then argue that human-AI systems amplify these dynamics, introducing new asymmetries that make reasoning about uncertainty harder and introduce new coordination challenges. Based on these lessons and new challenges, we conclude by outlining a research agenda for developing AI systems that align with humans in interaction, requiring an interdisciplinary synthesis of machine learning and the social and decision sciences.

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

107 extracted references · 7 linked inside Pith

  1. [1]

    Bennett, Kori Inkpen, Jaime Teevan, Ruth Kikin-Gil, and Eric Horvitz

    Saleema Amershi, Dan Weld, Mihaela Vorvoreanu, Adam Fourney, Besmira Nushi, Penny Collis- son, Jina Suh, Shamsi Iqbal, Paul N. Bennett, Kori Inkpen, Jaime Teevan, Ruth Kikin-Gil, and Eric Horvitz. Guidelines for human-ai interaction. InProceedings of the 2019 CHI Conference on Human Factors in Computing Systems, pages 1–13. ACM, 2019

    2019

  2. [2]

    Argall, Sonia Chernova, Manuela Veloso, and Brett Browning

    Brenna D. Argall, Sonia Chernova, Manuela Veloso, and Brett Browning. A survey of robot learning from demonstration.Robotics and Autonomous Systems, 57(5):469–483, 2009

    2009

  3. [3]

    Transactive memory systems: A microfoundation of dynamic capabilities.Journal of Management Studies, 49(8):1375–1382, 2012

    Linda Argote and Yuqing Ren. Transactive memory systems: A microfoundation of dynamic capabilities.Journal of Management Studies, 49(8):1375–1382, 2012

    2012

  4. [4]

    Ironies of automation.Automatica, 19(6):775–779, 1983

    Lisanne Bainbridge. Ironies of automation.Automatica, 19(6):775–779, 1983

    1983

  5. [5]

    Is the most accurate ai the best teammate? optimizing ai for teamwork.Proceedings of the AAai Conference on Artificial Intelligence, 35(13):11405–11414, 2021

    Gagan Bansal, Besmira Nushi, Ece Kamar, Eric Horvitz, and Daniel S Weld. Is the most accurate ai the best teammate? optimizing ai for teamwork.Proceedings of the AAai Conference on Artificial Intelligence, 35(13):11405–11414, 2021

    2021

  6. [6]

    Beyond accuracy: The role of mental models in human-ai team performance.Proceedings of the AAai Conference on Human Computation and Crowdsourcing, 7:2–11, 2019

    Gagan Bansal, Besmira Nushi, Ece Kamar, Walter S Lasecki, Daniel S Weld, and Eric Horvitz. Beyond accuracy: The role of mental models in human-ai team performance.Proceedings of the AAai Conference on Human Computation and Crowdsourcing, 7:2–11, 2019

    2019

  7. [7]

    Does the whole exceed its parts? the effect of ai explanations on complementary team performance

    Gagan Bansal, Besmira Nushi, Ece Kamar, Daniel S Weld, Walter S Lasecki, and Eric Horvitz. Does the whole exceed its parts? the effect of ai explanations on complementary team performance. InProceedings of the SIGCHI Conference on Human Factors in Computing Systems, pages 1–16, 2021

    2021

  8. [8]

    Swe-chat: Coding agent interactions from real users in the wild.arXiv preprint arXiv:2604.20779, 2026

    Joachim Baumann, Vishakh Padmakumar, Xiang Li, John Yang, Diyi Yang, and Sanmi Koyejo. Swe-chat: Coding agent interactions from real users in the wild.arXiv preprint arXiv:2604.20779, 2026

    Pith/arXiv arXiv 2026

  9. [9]

    ElizabethLBjorkandRobertABjork. Makingthingshardonyourself, butinagoodway: Creating desirable difficulties to enhance learning.Psychology and the real world: Essays illustrating fundamental contributions to society, 2(59-68):56–64, 2011

    2011

  10. [10]

    Zana Buçinca, Maja Barbara Malaya, and Krzysztof Z Gajos. To trust or to think: cognitive forcing functions can reduce overreliance on ai in ai-assisted decision-making.Proceedings of the ACM on Human-computer Interaction, 5(CSCW1):1–21, 2021

    2021

  11. [11]

    Reflections on shared cognition.Journal of Organi- zational Behavior, 22(2):195–202, 2001

    Janis A Cannon-Bowers and Eduardo Salas. Reflections on shared cognition.Journal of Organi- zational Behavior, 22(2):195–202, 2001

    2001

  12. [12]

    Ho, Thomas L

    Micah Carroll, Rohin Shah, Mark K. Ho, Thomas L. Griffiths, Sanjit A. Seshia, Pieter Abbeel, and Anca Dragan. On the utility of learning about humans for human-ai coordination. InAdvances in Neural Information Processing Systems, volume 32, 2019

    2019

  13. [13]

    Vera Liao, Jennifer Wortman Vaughan, and Gagan Bansal

    Valerie Chen, Q. Vera Liao, Jennifer Wortman Vaughan, and Gagan Bansal. Understanding the role of human intuition on reliance in human-ai decision-making with explanations.Proc. ACM Hum.-Comput. Interact., 7(CSCW2), October 2023

    2023

  14. [14]

    How can we assess human- agent interactions? case studies in software agent design.arXiv preprint arXiv:2510.09801, 2025

    Valerie Chen, Rohit Malhotra, Xingyao Wang, Juan Michelini, Xuhui Zhou, Aditya Bharat Soni, Hoang H Tran, Calvin Smith, Ameet Talwalkar, and Graham Neubig. How can we assess human- agent interactions? case studies in software agent design.arXiv preprint arXiv:2510.09801, 2025. 11 Align AI to Dynamic Human-AI Workflows

    Pith/arXiv arXiv 2025

  15. [15]

    Christiano, Jan Leike, Tom B

    Paul F. Christiano, Jan Leike, Tom B. Brown, Miljan Martic, Shane Legg, and Dario Amodei. Deep reinforcement learning from human preferences. InAdvances in Neural Information Processing Systems, 2017

    2017

  16. [16]

    A coordination theory approach to organizational process design.Organization Science, 8(2):157–175, 1997

    Kevin Crowston. A coordination theory approach to organizational process design.Organization Science, 8(2):157–175, 1997

    1997

  17. [17]

    Cooperative ai: Machines must learn to find common ground.Nature, 593(7857):33–36, 2021

    Allan Dafoe, Yoram Bachrach, Gillian Hadfield, Eric Horvitz, Kate Larson, and Thore Graepel. Cooperative ai: Machines must learn to find common ground.Nature, 593(7857):33–36, 2021

    2021

  18. [18]

    Adiachronic perspective on user trust in ai under uncertainty

    ShehzaadDhuliawala,VilémZouhar,MennatallahEl-Assady,andMrinmayaSachan. Adiachronic perspective on user trust in ai under uncertainty. InProceedings of the 2023 Conference on Empirical Methods in Natural Language Processing, pages 5567–5580, 2023

    2023

  19. [19]

    Productivity loss in brainstorming groups: Toward the solution of a riddle.Journal of Personality and Social Psychology, 53(3):497–509, 1987

    Michael Diehl and Wolfgang Stroebe. Productivity loss in brainstorming groups: Toward the solution of a riddle.Journal of Personality and Social Psychology, 53(3):497–509, 1987

    1987

  20. [20]

    Algorithm aversion: People erro- neously avoid algorithms after seeing them err.Journal of Experimental Psychology: General, 144(1):114–126, 2015

    Berkeley J Dietvorst, Joseph P Simmons, and Cade Massey. Algorithm aversion: People erro- neously avoid algorithms after seeing them err.Journal of Experimental Psychology: General, 144(1):114–126, 2015

    2015

  21. [21]

    The role of trust in organizational settings.Organization Science, 12(4):450–467, 2001

    Kurt T Dirks and Donald L Ferrin. The role of trust in organizational settings.Organization Science, 12(4):450–467, 2001

    2001

  22. [22]

    Psychological safety and learning behavior in work teams.Administrative Science Quarterly, 44(2):350–383, 1999

    Amy C Edmondson. Psychological safety and learning behavior in work teams.Administrative Science Quarterly, 44(2):350–383, 1999

    1999

  23. [23]

    Observation interference in partially observable assistance games.arXiv preprint arXiv:2412.17797, 2024

    Scott Emmons, Caspar Oesterheld, Vincent Conitzer, and Stuart Russell. Observation interference in partially observable assistance games.arXiv preprint arXiv:2412.17797, 2024

    Pith/arXiv arXiv 2024

  24. [24]

    Coordinating expertise in software development teams.Manage- ment Science, 46(12):1554–1568, 2000

    Samer Faraj and Lee Sproull. Coordinating expertise in software development teams.Manage- ment Science, 46(12):1554–1568, 2000

    2000

  25. [25]

    Coordination in fast-response organizations.Management Science, 52(8):1155–1169, 2006

    Samer Faraj and Yan Xiao. Coordination in fast-response organizations.Management Science, 52(8):1155–1169, 2006

    2006

  26. [26]

    Technology as teammate: Examining the role of external cognition in support of team cognitive processes.Frontiers in Psychology, 7:1531, 2016

    Stephen M Fiore and Travis J Wiltshire. Technology as teammate: Examining the role of external cognition in support of team cognitive processes.Frontiers in Psychology, 7:1531, 2016

    2016

  27. [27]

    Foerster, Yannis M

    Jakob N. Foerster, Yannis M. Assael, Nando de Freitas, and Shimon Whiteson. Learning to communicate with deep multi-agent reinforcement learning. InAdvances in Neural Information Processing Systems, 2016

    2016

  28. [28]

    Datasheets for datasets.Communications of the ACM, 64(12):86–92, 2021

    TimnitGebru,JamieMorgenstern,BrianaVecchione,JenniferWortmanVaughan,HannaWallach, Hal Daumé III, and Kate Crawford. Datasheets for datasets.Communications of the ACM, 64(12):86–92, 2021

    2021

  29. [29]

    Human trust in artificial intelligence: Review of empirical research.Academy of Management Annals, 14(2):627–660, 2020

    Ella Glikson and Anita Williams Woolley. Human trust in artificial intelligence: Review of empirical research.Academy of Management Annals, 14(2):627–660, 2020

    2020

  30. [30]

    A cognitive model of dynamiccooperationwithvariedinterdependencyinformation.Cognitivescience, 39(3):457–495, 2015

    Cleotilde Gonzalez, Noam Ben-Asher, Jolie M Martin, and Varun Dutt. A cognitive model of dynamiccooperationwithvariedinterdependencyinformation.Cognitivescience, 39(3):457–495, 2015. 12 Align AI to Dynamic Human-AI Workflows

    2015

  31. [31]

    Toward a science of human–ai teaming for decision making: A complementarity framework.PNAS Nexus, 5(3):pgag030, 03 2026

    Cleotilde Gonzalez, Kate Donahue, Daniel G Goldstein, Hoda Heidari, Mohammad S Jalali, Beau Schelble, Aarti Singh, and Anita Williams Woolley. Toward a science of human–ai teaming for decision making: A complementarity framework.PNAS Nexus, 5(3):pgag030, 03 2026

    2026

  32. [32]

    A cognitive approach to human–ai complementarity in dynamic decision-making.Nat

    Cleotilde Gonzalez and Hoda Heidari. A cognitive approach to human–ai complementarity in dynamic decision-making.Nat. Rev. Psychol., 4(12):808–822, October 2025

    2025

  33. [33]

    The science of teams in the military: Contributions from over 60 years of research.American Psychologist, 73(4):322, 2018

    Gerald F Goodwin, Nikki Blacksmith, and Meredith R Coats. The science of teams in the military: Contributions from over 60 years of research.American Psychologist, 73(4):322, 2018

    2018

  34. [34]

    ai and hci: Two fields divided by a common focus.ai magazine, 30(4):48–48, 2009

    Jonathan Grudin. ai and hci: Two fields divided by a common focus.ai magazine, 30(4):48–48, 2009

    2009

  35. [35]

    Harvard Business School Press, Boston, MA, 2002

    J Richard Hackman.Leading Teams: Setting the Stage for Great Performances. Harvard Business School Press, Boston, MA, 2002

    2002

  36. [36]

    Dynamics of human trust in recommender systems

    Jason L Harman, John O’Donovan, Tarek Abdelzaher, and Cleotilde Gonzalez. Dynamics of human trust in recommender systems. InProceedings of the 8th ACM Conference on Recommender systems, pages 305–308, 2014

    2014

  37. [37]

    Com- plementarity in human-ai collaboration: Concept, sources, and evidence.European Journal of Information Systems, 34(6):979–1002, 2025

    Patrick Hemmer, Max Schemmer, Niklas Kühl, Michael Vössing, and Gerhard Satzger. Com- plementarity in human-ai collaboration: Concept, sources, and evidence.European Journal of Information Systems, 34(6):979–1002, 2025

    2025

  38. [38]

    Anempiricalstudyofspeedandcommunicationinglobally distributed software development.IEEE Transactions on Software Engineering, 29(6):481–494, 2003

    JamesDHerbslebandAudrisMockus. Anempiricalstudyofspeedandcommunicationinglobally distributed software development.IEEE Transactions on Software Engineering, 29(6):481–494, 2003

    2003

  39. [39]

    Trust in automation: Integrating empirical evidence on factors that influence trust.Human Factors, 57(3):407–434, 2015

    Kevin A Hoff and Masooda Bashir. Trust in automation: Integrating empirical evidence on factors that influence trust.Human Factors, 57(3):407–434, 2015

    2015

  40. [40]

    Distributedcognition: Towardanewfoundation for human-computer interaction research.ACM Transactions on Computer-Human Interaction, 7(2):174–196, 2000

    JamesHollan, EdwinHutchins, andDavidKirsh. Distributedcognition: Towardanewfoundation for human-computer interaction research.ACM Transactions on Computer-Human Interaction, 7(2):174–196, 2000

    2000

  41. [41]

    Improving fairness in machine learning systems: What do industry practitioners need? In Proceedings of the 2019 CHI Conference on Human Factors in Computing Systems, pages 1–16

    Kenneth Holstein, Jennifer Wortman Vaughan, Hal Daumé III, Miro Dudik, and Hanna Wallach. Improving fairness in machine learning systems: What do industry practitioners need? In Proceedings of the 2019 CHI Conference on Human Factors in Computing Systems, pages 1–16. ACM, 2019

    2019

  42. [42]

    Principles of mixed-initiative user interfaces

    Eric Horvitz. Principles of mixed-initiative user interfaces. InProceedings of the SIGCHI Conference on Human Factors in Computing Systems, pages 159–166. ACM, 1999

    1999

  43. [43]

    Houghton Mifflin, Boston, MA, 2 edition, 1982

    Irving L Janis.Groupthink. Houghton Mifflin, Boston, MA, 2 edition, 1982

    1982

  44. [44]

    Artificial intelligence and the future of work: Human-ai symbiosis in organizational decision making.Business Horizons, 61(4):577–586, 2018

    Mohammad Hossein Jarrahi. Artificial intelligence and the future of work: Human-ai symbiosis in organizational decision making.Business Horizons, 61(4):577–586, 2018

    2018

  45. [45]

    Jimenez, John Yang, Alexander Wettig, Shunyu Yao, Kexin Pei, Ofir Press, and Karthik Narasimhan

    Carlos E. Jimenez, John Yang, Alexander Wettig, Shunyu Yao, Kexin Pei, Ofir Press, and Karthik Narasimhan. Swe-bench: Can language models resolve real-world github issues? InInternational Conference on Learning Representations, 2024

    2024

  46. [46]

    Reciprocal trust mediates deep transfer of learning between games of strategic interaction.Organizational Behavior and Human Decision Processes, 120(2):206–215, 2013

    IonJuvina, MunibaSaleem, JolieMMartin, CleotildeGonzalez, andChristianLebiere. Reciprocal trust mediates deep transfer of learning between games of strategic interaction.Organizational Behavior and Human Decision Processes, 120(2):206–215, 2013. 13 Align AI to Dynamic Human-AI Workflows

    2013

  47. [47]

    Patricia K Kahr, Gerrit Rooks, Martijn C Willemsen, and Chris CP Snijders. Understanding trust and reliance development in ai advice: Assessing model accuracy, model explanations, and experiences from previous interactions.ACM Transactions on Interactive Intelligent Systems, 14(4):1–30, 2024

    2024

  48. [48]

    Algorithms at work: The new contested terrain of control.Academy of Management Annals, 14(1):366–410, 2020

    Katherine C Kellogg, Melissa A Valentine, and Angèle Christin. Algorithms at work: The new contested terrain of control.Academy of Management Annals, 14(1):366–410, 2020

    2020

  49. [49]

    Dynabench: Rethinking benchmarking in NLP

    Douwe Kiela, Max Bartolo, Yixin Nie, Divyansh Kaushik, Atticus Geiger, Zhengxuan Wu, Bertie Vidgen, Grusha Prasad, Amanpreet Singh, Pratik Ringshia, Zhiyi Ma, Tristan Thrush, Sebastian Riedel, Zeerak Waseem, Pontus Stenetorp, Robin Jia, Mohit Bansal, Christopher Potts, and Adina Williams. Dynabench: Rethinking benchmarking in NLP. InProceedings of the 202...

    2021

  50. [50]

    Common ground and coordination in joint activity.Organizational Simulation, 53:139–184, 2005

    Gary Klein, Paul J Feltovich, Jeffrey M Bradshaw, and David D Woods. Common ground and coordination in joint activity.Organizational Simulation, 53:139–184, 2005

    2005

  51. [51]

    Team mental model: Construct or metaphor?Journal of Management, 20(2):403–437, 1994

    Richard Klimoski and Susan Mohammed. Team mental model: Construct or metaphor?Journal of Management, 20(2):403–437, 1994

    1994

  52. [52]

    Enhancing the effectiveness of work groups and teams

    Steve W J Kozlowski and Daniel R Ilgen. Enhancing the effectiveness of work groups and teams. Psychological Science in the Public Interest, 7(3):77–124, 2006

    2006

  53. [53]

    Trust in automation: Designing for appropriate reliance.Human Factors, 46(1):50–80, 2004

    John D Lee and Katrina A See. Trust in automation: Designing for appropriate reliance.Human Factors, 46(1):50–80, 2004

    2004

  54. [54]

    Alison Legood, Lisa Van Der Werff, Allan Lee, Deanne Den Hartog, and Daan Van Knippenberg. A critical review of the conceptualization, operationalization, and empirical literature on cognition- based and affect-based trust.Journal of Management Studies, 60(2):495–537, 2023

    2023

  55. [55]

    Trust, trust development, and trust repair.The handbook of conflict resolution: Theory and practice, 1(1):86–107, 2000

    Roy J Lewicki and Carolyn Wiethoff. Trust, trust development, and trust repair.The handbook of conflict resolution: Theory and practice, 1(1):86–107, 2000

    2000

  56. [56]

    Measuringtransactivememorysystemsinthefield: Scaledevelopmentandvalidation

    KyleLewis. Measuringtransactivememorysystemsinthefield: Scaledevelopmentandvalidation. Journal of Applied Psychology, 88(4):587–604, 2003

    2003

  57. [57]

    Learning to trust: How humans mentally recalibrate ai confidence signals.arXiv preprint arXiv:2603.22634, 2026

    ZhaoBin Li and Mark Steyvers. Learning to trust: How humans mentally recalibrate ai confidence signals.arXiv preprint arXiv:2603.22634, 2026

    arXiv 2026

  58. [58]

    Group versus individual training and group performance: The mediating role of transactive memory.Personality and Social Psychology Bulletin, 21(4):384–393, 1995

    Diane Wei Liang, Richard Moreland, and Linda Argote. Group versus individual training and group performance: The mediating role of transactive memory.Personality and Social Psychology Bulletin, 21(4):384–393, 1995

    1995

  59. [59]

    Logg, Julia A

    Jennifer M. Logg, Julia A. Minson, and Don A. Moore. Algorithm appreciation: People prefer algorithmic to human judgment.Organizational Behavior and Human Decision Processes, 151:90– 103, 2019

    2019

  60. [60]

    Multi-agent actor-critic for mixed cooperative-competitive environments

    Ryan Lowe, Yi Wu, Aviv Tamar, Jean Harb, Pieter Abbeel, and Igor Mordatch. Multi-agent actor-critic for mixed cooperative-competitive environments. InAdvances in Neural Information Processing Systems, 2017

    2017

  61. [61]

    Multi-turn reinforcement learning from preference human feedback.arXiv preprint arXiv:2405.08448, 2024

    Keming Lu, Hongyi Yuan, Zheng Yuan, Runji Lin, Junyang Lin, Chuanqi Tan, Chang Zhou, and Jingren Zhou. Multi-turn reinforcement learning from preference human feedback.arXiv preprint arXiv:2405.08448, 2024. 14 Align AI to Dynamic Human-AI Workflows

    Pith/arXiv arXiv 2024

  62. [62]

    ai-based digital assistants

    Alexander Maedche, Christine Legner, Alexander Benlian, Benedikt Berger, Henner Gimpel, Thomas Hess, Oliver Hinz, Stefan Morana, and Matthias Söllner. ai-based digital assistants. Business & Information Systems Engineering, 61(4):535–544, 2019

    2019

  63. [63]

    The interdisciplinary study of coordination.ACM Computing Surveys, 26(1):87–119, 1994

    Thomas W Malone and Kevin Crowston. The interdisciplinary study of coordination.ACM Computing Surveys, 26(1):87–119, 1994

    1994

  64. [64]

    A temporally based framework and taxonomy of team processes.Academy of Management Review, 26(3):356–376, 2001

    Michelle A Marks, John E Mathieu, and Stephen J Zaccaro. A temporally based framework and taxonomy of team processes.Academy of Management Review, 26(3):356–376, 2001

    2001

  65. [65]

    The influence of shared mental models on team process and performance.Journal of Applied Psychology, 85(2):273–283, 2000

    John E Mathieu, Tonia S Heffner, Gerald F Goodwin, Eduardo Salas, and Janis A Cannon-Bowers. The influence of shared mental models on team process and performance.Journal of Applied Psychology, 85(2):273–283, 2000

    2000

  66. [66]

    An integrative model of organizational trust.Academy of Management Review, 20(3):709–734, 1995

    Roger C Mayer, James H Davis, and F David Schoorman. An integrative model of organizational trust.Academy of Management Review, 20(3):709–734, 1995

    1995

  67. [67]

    Trust in developing relationships: from theory to measurement

    Daniel J McAllister, Roy J Lewicki, and Sankalp Chaturvedi. Trust in developing relationships: from theory to measurement. InAcademy of Management Proceedings, volume 2006, pages G1–G6. Academy of Management Briarcliff Manor, NY 10510, 2006

    2006

  68. [68]

    Explanation in artificial intelligence: Insights from the social sciences.Artificial Intelligence, 267:1–38, 2019

    Tim Miller. Explanation in artificial intelligence: Insights from the social sciences.Artificial Intelligence, 267:1–38, 2019

    2019

  69. [69]

    Model cards for model reporting

    Margaret Mitchell, Simone Wu, Andrew Zaldivar, Parker Barnes, Lucy Vasserman, Ben Hutchin- son, Elena Spitzer, Inioluwa Deborah Raji, and Timnit Gebru. Model cards for model reporting. InProceedings of the conference on fairness, accountability, and transparency, pages 220–229, 2019

    2019

  70. [70]

    The realhumaneval: Evaluating large language models’ abilities to support programmers.Transactions on Machine Learning Research, 2025

    Hussein Mozannar, Valerie Chen, Mohammed Alsobay, Subhro Das, Sebastian Zhao, Dennis Wei, Manish Nagireddy, Prasanna Sattigeri, Ameet Talwalkar, and David Sontag. The realhumaneval: Evaluating large language models’ abilities to support programmers.Transactions on Machine Learning Research, 2025. Expert Certification

    2025

  71. [71]

    Consistent estimators for learning to defer to an expert

    Hussein Mozannar and David Sontag. Consistent estimators for learning to defer to an expert. Proceedings of the International Conference on Machine Learning, pages 7076–7087, 2020

    2020

  72. [72]

    Productivity loss in brainstorming groups: A meta-analytic integration.Basic and Applied Social Psychology, 12(1):3–23, 1991

    Brian Mullen, Craig Johnson, and Eduardo Salas. Productivity loss in brainstorming groups: A meta-analytic integration.Basic and Applied Social Psychology, 12(1):3–23, 1991

    1991

  73. [73]

    Donald A. Norman. The problem with automation: Inappropriate feedback and interaction, not over-automation.Philosophical Transactions of the Royal Society of London. B, Biological Sciences, 327(1241):585–593, 1990

    1990

  74. [74]

    Training language models to follow instructions with human feedback.Advances in neural information processing systems, 35:27730–27744, 2022

    Long Ouyang, Jeffrey Wu, Xu Jiang, Diogo Almeida, Carroll Wainwright, Pamela Mishkin, Chong Zhang, Sandhini Agarwal, Katarina Slama, Alex Ray, et al. Training language models to follow instructions with human feedback.Advances in neural information processing systems, 35:27730–27744, 2022

    2022

  75. [75]

    Humans and automation: Use, misuse, disuse, abuse.Human Factors, 39(2):230–253, 1997

    Raja Parasuraman and Victor Riley. Humans and automation: Use, misuse, disuse, abuse.Human Factors, 39(2):230–253, 1997

    1997

  76. [76]

    Amodelfortypesandlevels of human interaction with automation.IEEE Transactions on Systems, Man, and Cybernetics-Part A: Systems and Humans, 30(3):286–297, 2000

    RajaParasuraman,ThomasB.Sheridan,andChristopherD.Wickens. Amodelfortypesandlevels of human interaction with automation.IEEE Transactions on Systems, Man, and Cybernetics-Part A: Systems and Humans, 30(3):286–297, 2000. 15 Align AI to Dynamic Human-AI Workflows

    2000

  77. [77]

    The impact of ai on developer productivity: Evidence from github copilot.arXiv preprint arXiv:2302.06590, 2023

    Sida Peng, Eirini Kalliamvakou, Peter Cihon, and Mert Demirer. The impact of ai on developer productivity: Evidence from github copilot.arXiv preprint arXiv:2302.06590, 2023

    Pith/arXiv arXiv 2023

  78. [78]

    ALVINN: An autonomous land vehicle in a neural network

    Dean Pomerleau. ALVINN: An autonomous land vehicle in a neural network. InAdvances in Neural Information Processing Systems, pages 305–313. Morgan Kaufmann, 1989

    1989

  79. [79]

    Data cards: Purposeful and transparent dataset documentation for responsible ai

    Mahima Pushkarna, Andrew Zaldivar, and Oddur Kjartansson. Data cards: Purposeful and transparent dataset documentation for responsible ai. InProceedings of the 2022 ACM conference on fairness, accountability, and transparency, pages 1776–1826, 2022

    2022

  80. [80]

    Rabinowitz, Frank Perbet, H

    Neil C. Rabinowitz, Frank Perbet, H. Francis Song, Chiyuan Zhang, S. M. Ali Eslami, and Matthew Botvinick. Machinetheoryofmind. InProceedingsofthe35thInternationalConferenceonMachine Learning, volume 80 ofProceedings of Machine Learning Research, pages 4218–4227. PMLR, 2018

    2018

Showing first 80 references.