arxiv: 2605.02832 · v1 · submitted 2026-05-04 · 💻 cs.AI · cs.HC· cs.SE

Recognition: 3 theorem links

· Lean Theorem

HAAS: A Policy-Aware Framework for Adaptive Task Allocation Between Humans and Artificial Intelligence Systems

Vicente Pelechanoa , Antoni Mestre , Manoli Albert , Miriam Gil

Authors on Pith no claims yet

Pith reviewed 2026-05-08 18:00 UTC · model grok-4.3

classification 💻 cs.AI cs.HCcs.SE

keywords human-AI task allocationadaptive governancecontextual banditsautonomy spectrumcognitive dimensionspolicy-aware systemsmanufacturing collaborationsoftware engineering workflows

0 comments

The pith

HAAS shows governance constraints act as tunable variables that shift AI tasks toward supervised human collaboration with measurable domain effects.

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

The paper develops HAAS to move beyond binary human-or-AI task splits by treating governance as adjustable rules that limit feasible actions before any learning begins. A rule-based expert system enforces those limits using five cognitive dimensions and a five-mode autonomy spectrum, while a contextual-bandit component learns which remaining modes work best from real outcome data in software engineering and manufacturing benchmarks. Tests reveal that raising governance intensity reliably replaces fully autonomous AI assignments with supervised joint modes. In manufacturing specifically, stronger rules simultaneously raise operational performance and lower reported fatigue. The results indicate that moderate governance settings gain ground once the learner has enough experience inside the allowed action set.

Core claim

Governance is not a binary on-off switch but a tunable design variable: tighter constraints predictably convert autonomous AI assignments into supervised collaborations, with domain-specific costs and benefits; in manufacturing, stronger governance can improve operational performance and reduce fatigue simultaneously.

What carries the argument

The HAAS framework, which couples a rule-based expert system enforcing governance constraints prior to learning with a contextual-bandit learner that selects among feasible collaboration modes on the basis of outcome feedback, all grounded in five auditable cognitive dimensions and a five-mode autonomy spectrum from human-only to fully autonomous.

If this is right

Tighter governance constraints systematically replace fully autonomous AI assignments with supervised human-AI collaboration modes.
In manufacturing settings, raising governance strength can increase operational performance while decreasing human fatigue at the same time.
No single governance intensity dominates every context; moderate levels gain relative advantage once the learner accumulates experience inside the constrained action space.
Domain-specific cost-benefit profiles emerge, so governance settings must be calibrated rather than chosen once for all tasks.

Where Pith is reading between the lines

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

The same tunable-governance structure could be tested in high-stakes domains like healthcare or logistics to determine whether the manufacturing workload-buffering pattern repeats.
Organizations could run HAAS-style simulations on historical task data to select governance policies that meet explicit targets for safety, throughput, or fatigue reduction before live deployment.
The pre-learning rule enforcement step suggests a general pattern for embedding auditability into any adaptive allocation system that must respect external regulations or ethical limits.

Load-bearing premise

The five cognitive dimensions and five-mode autonomy spectrum accurately represent task-agent fit and that benchmark results generalize beyond the tested software engineering and manufacturing scenarios.

What would settle it

Deploy HAAS in a third domain such as clinical decision support, apply the same governance levels, and check whether stronger constraints still produce simultaneous gains in performance and reductions in fatigue; failure to observe those effects would falsify the central claim.

Figures

Figures reproduced from arXiv: 2605.02832 by Antoni Mestre, Manoli Albert, Miriam Gil, Vicente Pelechanoa.

**Figure 1.** Figure 1: Three-layer architecture of HAAS. The governance layer filters feasible collaboration modes before the bandit selects an allocation; execution outcomes then update both reward and human state for the next cycle. (2000), translating qualitative task properties into a continuous AI affinity score. Each subtask s is characterised by a vector d(s) = (r, τ, c, a, h) ∈ [0, 1]5 of rubric-based scores assigned on… view at source ↗

**Figure 2.** Figure 2: Five-step execution loop connecting governed allocation to the benchmark view at source ↗

**Figure 3.** Figure 3: Screenshot of the Human–AI Symbiosis Studio dashboard (KPI summary view, manufacturing domain). The panel shows sprint-level KPIs, collaboration-mode distribution, and allocation history charts generated during a benchmark run. current normalised fatigue level (updated per the dynamics in Section 3.6). M(t) ∈ [0, 1] is the monotony signal, defined as the fraction of the last five subtasks assigned to the s… view at source ↗

**Figure 4.** Figure 4: Governance Ladder — quality, fatigue, lead time, and cumulative regret per level view at source ↗

**Figure 5.** Figure 5: Collaboration mode redistribution across governance levels. view at source ↗

**Figure 6.** Figure 6: Best governance level per scenario (portability battery, 10 seeds, 8 cycles). Bar view at source ↗

**Figure 7.** Figure 7: Long-horizon stability (16 cycles, 30 seeds) for L0, L2, and L4 on four outcome view at source ↗

**Figure 8.** Figure 8: Multi-dimensional trade-off radar for L0–L4 (standard scenario per domain, 30 view at source ↗

read the original abstract

Deciding how to distribute work between humans and AI systems is a central challenge in organisational design. Most approaches treat this as a binary choice, yet the operational reality is richer: humans and AI routinely share tasks or take complementary roles depending on context, fatigue, and the stakes involved. Governing that distribution -- balancing efficiency, oversight, and human capability -- remains an open problem. This paper presents Human-AI Adaptive Symbiosis (HAAS), an implemented framework for adaptive task allocation in software engineering and manufacturing. HAAS combines two coupled components: a rule-based expert system that enforces governance constraints before any learning occurs, and a contextual-bandit learner that selects among feasible collaboration modes from outcome feedback. Task-agent fit is represented through five auditable cognitive dimensions and a five-mode autonomy spectrum -- from human-only to fully autonomous -- embedded in a reproducible benchmark spanning both domains. Three empirical findings emerge. First, governance is not a binary switch but a tunable design variable: tighter constraints predictably convert autonomous AI assignments into supervised collaborations, with domain-specific costs and benefits. Second, in manufacturing, stronger governance can improve operational performance and reduce fatigue simultaneously -- a workload-buffering effect that contradicts the usual framing of governance as pure overhead. Third, no single governance setting dominates across all contexts; moderate governance becomes increasingly competitive as the learner accumulates experience within the governed action space. Together, these findings position HAAS as a pre-deployment workbench for comparing and inspecting human--AI allocation policies before organisational commitment.

Editorial analysis

A structured set of objections, weighed in public.

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

Desk Editor's Note private letter to a colleague

HAAS pairs upfront rules with a contextual bandit to make human-AI task allocation tunable across five modes, with manufacturing benchmarks showing governance can sometimes cut fatigue while holding performance.

read the letter

The paper's core offering is an implemented HAAS framework that first applies a rule-based expert system to enforce governance limits, then uses a contextual bandit to select from five collaboration modes (human-only to full autonomy) based on observed outcomes. Task fit is encoded via five auditable cognitive dimensions, and the whole thing runs on a benchmark covering software engineering and manufacturing scenarios. Three main observations follow from the runs: governance acts as a continuous dial rather than an on-off switch, stronger governance in manufacturing can improve output while lowering fatigue, and moderate settings gain ground as the bandit learns within the constrained space.

Referee Report

2 major / 2 minor

Summary. The manuscript presents the HAAS framework for adaptive human-AI task allocation, combining a rule-based expert system that enforces governance constraints with a contextual-bandit learner that selects among feasible collaboration modes based on outcome feedback. Task-agent fit is encoded via five auditable cognitive dimensions and a five-mode autonomy spectrum (human-only to fully autonomous), evaluated in a reproducible benchmark spanning software engineering and manufacturing. Three empirical findings are reported: governance acts as a tunable design variable that predictably shifts assignments toward supervised modes with domain-specific trade-offs; stronger governance in manufacturing can simultaneously improve operational performance and reduce fatigue (workload-buffering effect); and moderate governance becomes more competitive as the learner gains experience within the constrained action space.

Significance. If the benchmark outcomes hold under scrutiny, the work supplies a practical, inspectable pre-deployment workbench for comparing human-AI allocation policies, moving beyond binary governance framings and demonstrating that policy constraints can yield net positive effects in specific domains. The implemented system and reproducible benchmark constitute clear strengths that could support follow-on empirical studies in organizational AI design.

major comments (2)

[Abstract and benchmark description] The three empirical findings rest on the claim that the five cognitive dimensions and five-mode autonomy spectrum accurately encode task-agent fit across domains (stated in the abstract and benchmark description). No sensitivity analysis, ablation, or external validation is supplied to show these axes dominate over omitted factors such as team dynamics or error propagation; without such checks the observed conversion from autonomous to supervised modes and the manufacturing performance-fatigue trade-off risk being specific to the chosen scenarios rather than general properties of policy-aware allocation.
[Empirical findings and results sections] The workload-buffering result in manufacturing (stronger governance improving both performance and reducing fatigue) is central to the second finding, yet the manuscript supplies no full methods, raw data, statistical tests, or controls for post-hoc analysis choices. This absence prevents verification that the effect is robust rather than an artifact of the specific benchmark instantiation.

minor comments (2)

[Framework architecture] Clarify whether the rule-based expert system and contextual-bandit components interact only sequentially or allow feedback loops, as the current description leaves the coupling mechanism underspecified.
[Discussion] Add explicit discussion of how the five-mode spectrum maps to real-world deployment constraints (e.g., regulatory or safety requirements) to strengthen the practical implications.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback. We address each major comment below, committing to revisions that enhance transparency and robustness without altering the core claims of the HAAS framework.

read point-by-point responses

Referee: [Abstract and benchmark description] The three empirical findings rest on the claim that the five cognitive dimensions and five-mode autonomy spectrum accurately encode task-agent fit across domains (stated in the abstract and benchmark description). No sensitivity analysis, ablation, or external validation is supplied to show these axes dominate over omitted factors such as team dynamics or error propagation; without such checks the observed conversion from autonomous to supervised modes and the manufacturing performance-fatigue trade-off risk being specific to the chosen scenarios rather than general properties of policy-aware allocation.

Authors: We acknowledge the absence of sensitivity analysis or ablation studies on the five dimensions in the current manuscript. These dimensions were derived from established cognitive task analysis literature and domain-expert input to prioritize auditability. In revision, we will add a dedicated sensitivity analysis subsection that perturbs dimension weights and reports resulting shifts in allocation distributions within the existing benchmark. We will also expand the limitations discussion to address omitted factors such as team dynamics and error propagation, noting that the current model focuses on single-task allocation. This will clarify that findings are benchmark-specific while still supporting the value of tunable governance. revision: partial
Referee: [Empirical findings and results sections] The workload-buffering result in manufacturing (stronger governance improving both performance and reducing fatigue) is central to the second finding, yet the manuscript supplies no full methods, raw data, statistical tests, or controls for post-hoc analysis choices. This absence prevents verification that the effect is robust rather than an artifact of the specific benchmark instantiation.

Authors: We agree that explicit statistical details and data availability are required for verification. The benchmark methods describe metric collection and analysis, with statistical tests (e.g., paired t-tests for performance and fatigue differences under varying governance levels) reported in the results. To strengthen the manuscript, we will expand the main-text methods section with full statistical reporting, controls for analysis choices, and a public repository link to anonymized raw data and scripts. This revision will allow independent confirmation of the workload-buffering effect. revision: yes

Circularity Check

0 steps flagged

No circularity: empirical framework results do not reduce to self-defined inputs or self-citations

full rationale

The paper presents HAAS as a composite framework (rule-based expert system enforcing governance constraints plus contextual-bandit learner) whose task-agent fit is encoded via explicitly chosen five cognitive dimensions and five-mode autonomy spectrum. The three main findings are reported as outcomes of a reproducible benchmark across software engineering and manufacturing scenarios. No derivation chain is claimed; the results are generated by running the implemented system on defined inputs rather than by algebraic reduction of a prediction to a fitted parameter or by load-bearing self-citation. The abstract and described structure contain no equations that equate a claimed prediction to its own construction, and no uniqueness theorem or ansatz is imported from prior author work to force the architecture. This is the normal case of an empirical systems paper whose central claims remain independent of the evaluation setup.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

Abstract-only review; the central claims rest on the unverified validity of the five cognitive dimensions and autonomy spectrum as complete descriptors of task fit, plus the assumption that the benchmark scenarios are representative of real organizational contexts.

axioms (2)

domain assumption The five cognitive dimensions and five-mode autonomy spectrum provide an auditable and sufficient model of task-agent fit.
Invoked to represent tasks and select collaboration modes.
domain assumption The benchmark spanning software engineering and manufacturing is sufficient to support the reported domain-specific findings.
Required for generalizing the three empirical observations.

pith-pipeline@v0.9.0 · 5579 in / 1329 out tokens · 41094 ms · 2026-05-08T18:00:00.668923+00:00 · methodology

discussion (0)

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

Cost.FunctionalEquation / Foundation.AlphaCoordinateFixation washburn_uniqueness_aczel; J_uniquely_calibrated_via_higher_derivative unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

αAI(s) = w_r·r + w_τ·τ + w_c·(1−c) + w_a·(1−a) + w_h·(1−h) ... default weight vector w = (0.35, 0.25, 0.20, 0.10, 0.10) is grounded in the human–automation literature.
Foundation (whole forcing chain) reality_from_one_distinction unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

UCB1 ... LinUCB ... Thompson Sampling ... Discounted-UCB ... reward R = w_q·q + w_t·(1−t̄) + w_co·(1−c̄) + w_w·W with weights (0.30, 0.20, 0.10, 0.40)
Foundation (8-tick period theorems) n/a — benchmark horizon, not forced periodicity unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

common horizon of 8 cycles and 4 subtasks per cycle to preserve cross-domain comparability

What do these tags mean?

matches: The paper's claim is directly supported by a theorem in the formal canon.
supports: The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends: The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses: The paper appears to rely on the theorem as machinery.
contradicts: The paper's claim conflicts with a theorem or certificate in the canon.
unclear: Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

60 extracted references · 2 canonical work pages

[1]

International Journal of Human-Computer Studies , volume =

Gil, Miriam and Albert, Manoli and Fons, Joan and Pelechano, Vicente , title =. International Journal of Human-Computer Studies , volume =. 2019 , doi =

2019
[2]

and Wickens, Christopher D

Parasuraman, Raja and Sheridan, Thomas B. and Wickens, Christopher D. , title =. IEEE Transactions on Systems, Man, and Cybernetics -- Part A: Systems and Humans , year =
[3]

, title =

Sheridan, Thomas B. , title =
[4]

, title =

Cummings, Mary L. , title =. IEEE Intelligent Systems , year =
[5]

and Demir, Mustafa and Cooke, Nancy J

McNeese, Nathan J. and Demir, Mustafa and Cooke, Nancy J. and Myers, Christopher , title =. Human Factors , year =
[6]

and Robert, Lionel P

Ali, Arsha and Azevedo-Sa, Hebert and Tilbury, Dawn M. and Robert, Lionel P. , title =. Scientific Reports , year =
[7]

Procedia CIRP , year =

Petzoldt, Christoph and Niermann, Dario and Keiser, Dennis and Freitag, Michael , title =. Procedia CIRP , year =
[8]

Systems , year =

Urrea, Claudio , title =. Systems , year =
[9]

Artificial Intelligence for Engineering Design, Analysis and Manufacturing , year =

Kirgil-Budakli, Rukiye and Zeng, Yong and Akgunduz, Ali , title =. Artificial Intelligence for Engineering Design, Analysis and Manufacturing , year =
[10]

and de Vreede, Gert-Jan and de Vreede, Triparna and Elkins, Aaron and Maier, Ronald and Merz, Alexander B

Seeber, Isabella and Bittner, Eva and Briggs, Robert O. and de Vreede, Gert-Jan and de Vreede, Triparna and Elkins, Aaron and Maier, Ronald and Merz, Alexander B. and Oeste-Reiß, Sarah and Randrup, Nils and Schwabe, Gerhard and Söllner, Matthias , title =. Information & Management , year =
[11]

The International Journal of Robotics Research , year =

Soh, Harold and Xie, Yicheng and Chen, Min and Halpern, David , title =. The International Journal of Robotics Research , year =
[12]

and Billings, Dee R

Hancock, Peter A. and Billings, Dee R. and Schaefer, Kristin E. and Chen, Jessie Y. C. and de Visser, Ewart J. and Parasuraman, Raja , title =. Human Factors , year =
[13]

Hybrid Intelligence , journal =

Dellermann, Dominik and Ebel, Philipp and S. Hybrid Intelligence , journal =. 2019 , volume =

2019
[14]

2020 , url =

Lattimore, Tor and Szepesvári, Csaba , title =. 2020 , url =

2020
[15]

Finite-time analysis of the multiarmed bandit problem , journal =

Auer, Peter and Cesa-Bianchi, Nicol\`. Finite-time analysis of the multiarmed bandit problem , journal =. 2002 , volume =

2002
[16]

, title =

Li, Lihong and Chu, Wei and Langford, John and Schapire, Robert E. , title =. Proceedings of the 19th International Conference on World Wide Web (WWW) , year =
[17]

and Van Roy, Benjamin and Kazerouni, Abbas and Osband, Ian and Wen, Zheng , title =

Russo, Daniel J. and Van Roy, Benjamin and Kazerouni, Abbas and Osband, Ian and Wen, Zheng , title =. Foundations and Trends in Machine Learning , year =
[18]

European Conference on Machine Learning (ECML) , year =

Kocsis, Levente and Szepesvári, Csaba , title =. European Conference on Machine Learning (ECML) , year =
[19]

Algorithmic Learning Theory (ALT) , year =

Garivier, Aurélien and Moulines, Eric , title =. Algorithmic Learning Theory (ALT) , year =
[20]

Monarch, Robert , title =
[21]

International Journal of Human-Computer Interaction , year =

Shneiderman, Ben , title =. International Journal of Human-Computer Interaction , year =
[22]

2024 , type =

Regulation (EU) 2024/1689 of the European Parliament and of the Council laying down harmonised rules on artificial intelligence (Artificial Intelligence Act) , institution =. 2024 , type =

2024
[23]

2016 , number =

Robots and robotic devices -- Collaborative robots , institution =. 2016 , number =

2016
[24]

2018 , url =

Dafoe, Allan , title =. 2018 , url =

2018
[25]

and Dodson, John D

Yerkes, Robert M. and Dodson, John D. , title =. Journal of Comparative Neurology and Psychology , year =
[26]

, title =

Parasuraman, Raja and Wickens, Christopher D. , title =. Human Factors , year =
[27]

Automatica , year =

Bainbridge, Lisanne , title =. Automatica , year =
[28]

, title =

Endsley, Mica R. , title =. Human Factors , year =
[29]

, title =

Crandall, Beth and Klein, Gary and Hoffman, Robert R. , title =
[30]

2023 , doi =

Brynjolfsson, Erik and Li, Danielle and Raymond, Lindsey , title =. 2023 , doi =

2023
[31]

and Van Mieghem, Jan A

Gijsbrechts, Joren and Boute, Robert N. and Van Mieghem, Jan A. and Zhang, Dennis J. , title =. Manufacturing & Service Operations Management , year =
[32]

Machine Learning , year =

Ben-David, Shai and Blitzer, John and Crammer, Koby and Kulesza, Alex and Pereira, Fernando and Vaughan, Jennifer , title =. Machine Learning , year =
[33]

Science , year =

Noy, Shakked and Zhang, Whitney , title =. Science , year =
[34]

Mollick, Hila Lifshitz-Assaf, Katherine Kellogg, Saran Rajendran, Lisa Krayer, François Candelon, and Karim R

Dell'Acqua, Fabrizio and McFowland, Edward and Mollick, Ethan R. and Lifshitz-Assaf, Hila and Kellogg, Katherine C. and Rajendran, Saran and Krayer, Lisa and Candelon, Fran. Navigating the Jagged Technological Frontier: Field Experimental Evidence of the Effects of. 2023 , number =. doi:10.2139/ssrn.4573321 , url =

work page doi:10.2139/ssrn.4573321 2023
[35]

The Impact of AI on Developer Productivity: Evidence from GitHub Copilot

Peng, Sida and Kalliamvakou, Eirini and Cihon, Peter and Demirer, Mert , title =. 2023 , howpublished =. doi:10.48550/arXiv.2302.06590 , url =

work page Pith review doi:10.48550/arxiv.2302.06590 2023
[36]

and See, Katrina A

Lee, John D. and See, Katrina A. , title =. Human Factors , year =
[37]

Artificial Intelligence , year =

Miller, Tim , title =. Artificial Intelligence , year =
[38]

To Trust or to Think: Cognitive Forcing Functions Can Reduce Overreliance on

Bu. To Trust or to Think: Cognitive Forcing Functions Can Reduce Overreliance on. Proceedings of the. 2021 , volume =

2021
[39]

Proceedings of the 25th International Joint Conference on Artificial Intelligence (

Kamar, Ece , title =. Proceedings of the 25th International Joint Conference on Artificial Intelligence (. 2016 , pages =

2016
[40]

Proceedings of the 2021

Bansal, Gagan and Wu, Tongshuang and Zhou, Joyce and Fok, Raymond and Nushi, Besmira and Kamar, Ece and Ribeiro, Marco Tulio and Weld, Daniel , title =. Proceedings of the 2021. 2021 , pages =

2021
[41]

and Parasuraman, Raja and Matthews, Gerald , title =

Warm, Joel S. and Parasuraman, Raja and Matthews, Gerald , title =. Human Factors , year =
[42]

, title =

Parasuraman, Raja and Manzey, Dietrich H. , title =. Human Factors , year =
[43]

and Fisk, Arthur D

Beer, Jenay M. and Fisk, Arthur D. and Rogers, Wendy A. , title =. Journal of Human-Robot Interaction , year =
[44]

2011 , number =

Robots and robotic devices -- Safety requirements for industrial robots -- Part 1: Robots , institution =. 2011 , number =

2011
[45]

, title =

Fitts, Paul M. , title =. 1951 , note =

1951
[46]

and Kaber, David B

Endsley, Mica R. and Kaber, David B. , title =. Ergonomics , year =
[47]

and Endsley, Mica R

Kaber, David B. and Endsley, Mica R. , title =. Theoretical Issues in Ergonomics Science , year =
[48]

Handbook of Cognitive Task Design , editor =

Inagaki, Toshiyuki , title =. Handbook of Cognitive Task Design , editor =. 2003 , pages =

2003
[49]

Expert Systems with Applications , year =

Liao, Shu-Hsien , title =. Expert Systems with Applications , year =
[50]

Turban, Efraim , title =
[51]

and Riley, Gary D

Giarratano, Joseph C. and Riley, Gary D. , title =
[52]

Academy of Management Review , year =

Raisch, Sebastian and Krakowski, Sebastian , title =. Academy of Management Review , year =
[53]

and Inkpen, Kori and Teevan, Jaime and Kiber, Ruth and Horvitz, Eric , title =

Amershi, Saleema and Weld, Dan and Vorvoreanu, Mihaela and Fourney, Adam and Nushi, Besmira and Collisson, Penny and Suh, Jina and Iqbal, Shamsi and Bennett, Paul N. and Inkpen, Kori and Teevan, Jaime and Kiber, Ruth and Horvitz, Eric , title =. Proceedings of the 2019. 2019 , pages =

2019
[54]

Hemmer, Patrick and Schemmer, Max and V. Human-. 2021 , howpublished =

2021
[55]

2022 , doi =

Shneiderman, Ben , title =. 2022 , doi =

2022
[56]

Minds and Machines , year =

Floridi, Luciano and Cowls, Josh and Beltrametti, Monica and Chatila, Raja and Chazerand, Patrice and Dignum, Virginia and Luetge, Christoph and Madelin, Robert and Pagallo, Ugo and Rossi, Francesca and Schafer, Burkhard and Valcke, Peggy and Vayena, Effy , title =. Minds and Machines , year =
[57]

Nature Human Behaviour , year =

Vaccaro, Michelle and Almaatouq, Abdullah and Malone, Thomas , title =. Nature Human Behaviour , year =
[58]

and Flathmann, Christopher and McNeese, Nathan J

Hauptman, Allyson I. and Flathmann, Christopher and McNeese, Nathan J. , title =. Applied Ergonomics , year =
[59]

and Heidari, Hoda and Jalali, Mohammad S

Gonzalez, Cleotilde and Donahue, Kate and Goldstein, Daniel G. and Heidari, Hoda and Jalali, Mohammad S. and Schelble, Beau and Singh, Aarti and Woolley, Anita Williams , title =. PNAS Nexus , year =
[60]

Advanced Engineering Informatics , year =

Wang, Jingfei and Yan, Yan and Hu, Yaoguang and Yang, Xiaonan , title =. Advanced Engineering Informatics , year =