Damadi, M. S.; Davoust, A. Fairness in social machines: a systematic review Journal Article In: Journal of Information, Communication and Ethics in Society, pp. 1–40, 2026, ISSN: 1477996X (ISSN). @article{damadi_fairness_2026,
title = {Fairness in social machines: a systematic review},
author = {M. S. Damadi and A. Davoust},
url = {https://www.scopus.com/inward/record.uri?eid=2-s2.0-105030531101&doi=10.1108%2FJICES-01-2025-0002&partnerID=40&md5=63e2f87ee3852ffe2d49c514e38cba1c},
doi = {10.1108/JICES-01-2025-0002},
issn = {1477996X (ISSN)},
year = {2026},
date = {2026-01-01},
journal = {Journal of Information, Communication and Ethics in Society},
pages = {1–40},
abstract = {Purpose – The purpose of the paper is to provide a systematic review of biases in social machines to better understand the general problem of fairness in these systems. It aims to identify and categorize phenomena described as biases toward specific demographic groups, frame them normatively as harmful and relate them to established fairness concepts originally defined for algorithmic systems. Design/methodology/approach – The phenomenon of algorithmic bias refers to systematic biases against identifiable demographic groups that occur in automated decisions systems. Such biases have mostly been studied in the context of black-box decision systems built using machine learning (ML). However, similar problems have also been reported in complex socio-technical systems such as Wikipedia and Airbnb, known more generally as social machines, where the observed biases cannot necessarily be attributed to specific automated decision systems. Instead, the biases may emerge as a result of complex processes involving numerous users and a computational infrastructure. To gain a better understanding of fairness in social machines, the authors select a representative sample of social machines from six distinct categories, and systematically review the literature reporting biases in these systems, covering 196 papers. The authors classify the reported bias phenomena, identify the affected demographic groups and relate the phenomena to established notions of harm from algorithmic fairness research. Finally, the authors identify the normative expectations of fairness associated with the different problems and discuss the applicability of existing criteria proposed for ML-driven decision systems. The analysis highlights the conceptual similarity of bias phenomena between algorithmic systems and social machines, allowing for a shared vocabulary to describe and compare phenomena across a broad class of systems. Findings – The paper identifies two key biases in social machines: representational harm, from underrepresentation or biased portrayal of disadvantaged groups, and allocative harm, from unfair decision processes, measurable via metrics like demographic parity. Gender bias is prevalent and easier to detect due to explicit markers, offering insights for identifying other biases. Unique biases arise from user categorizations, creating unintended discrimination linked to protected characteristics. These biases result from complex user interactions, not isolated algorithms. Addressing them requires redesigning social machines, focusing on computational infrastructure and interaction norms, such as visibility settings, to mitigate harmful outcomes. Originality/value – The paper’s originality lies in its systematic review of biases in social machines, offering a novel perspective on fairness in these systems. Unlike prior studies focusing solely on algorithmic fairness, this work examines the broader socio-technical interactions within social machines, identifying biases that emerge from user interactions and design choices. By linking these biases to established fairness concepts like demographic parity and representational harm, the paper bridges the gap between algorithmic fairness and social dynamics. © 2025 Emerald Publishing Limited},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Purpose – The purpose of the paper is to provide a systematic review of biases in social machines to better understand the general problem of fairness in these systems. It aims to identify and categorize phenomena described as biases toward specific demographic groups, frame them normatively as harmful and relate them to established fairness concepts originally defined for algorithmic systems. Design/methodology/approach – The phenomenon of algorithmic bias refers to systematic biases against identifiable demographic groups that occur in automated decisions systems. Such biases have mostly been studied in the context of black-box decision systems built using machine learning (ML). However, similar problems have also been reported in complex socio-technical systems such as Wikipedia and Airbnb, known more generally as social machines, where the observed biases cannot necessarily be attributed to specific automated decision systems. Instead, the biases may emerge as a result of complex processes involving numerous users and a computational infrastructure. To gain a better understanding of fairness in social machines, the authors select a representative sample of social machines from six distinct categories, and systematically review the literature reporting biases in these systems, covering 196 papers. The authors classify the reported bias phenomena, identify the affected demographic groups and relate the phenomena to established notions of harm from algorithmic fairness research. Finally, the authors identify the normative expectations of fairness associated with the different problems and discuss the applicability of existing criteria proposed for ML-driven decision systems. The analysis highlights the conceptual similarity of bias phenomena between algorithmic systems and social machines, allowing for a shared vocabulary to describe and compare phenomena across a broad class of systems. Findings – The paper identifies two key biases in social machines: representational harm, from underrepresentation or biased portrayal of disadvantaged groups, and allocative harm, from unfair decision processes, measurable via metrics like demographic parity. Gender bias is prevalent and easier to detect due to explicit markers, offering insights for identifying other biases. Unique biases arise from user categorizations, creating unintended discrimination linked to protected characteristics. These biases result from complex user interactions, not isolated algorithms. Addressing them requires redesigning social machines, focusing on computational infrastructure and interaction norms, such as visibility settings, to mitigate harmful outcomes. Originality/value – The paper’s originality lies in its systematic review of biases in social machines, offering a novel perspective on fairness in these systems. Unlike prior studies focusing solely on algorithmic fairness, this work examines the broader socio-technical interactions within social machines, identifying biases that emerge from user interactions and design choices. By linking these biases to established fairness concepts like demographic parity and representational harm, the paper bridges the gap between algorithmic fairness and social dynamics. © 2025 Emerald Publishing Limited |
Shangwe, C. N.; Davoust, A.; Khoury, R. Beyond Detection: Evaluating LLMs’ Semantic Understanding of Code Vulnerabilities Proceedings Article In: K., Adi; O., Nguena Timo; N., Boulahia-Cuppens; D., Espes; N., Stakhanova; M., Omar (Ed.): Lect. Notes Comput. Sci., pp. 19–34, Springer Science and Business Media Deutschland GmbH, 2026, ISBN: 03029743 (ISSN); 978-303220731-9 (ISBN), (Journal Abbreviation: Lect. Notes Comput. Sci.). @inproceedings{shangweDetectionEvaluatingLLMs2026,
title = {Beyond Detection: Evaluating LLMs’ Semantic Understanding of Code Vulnerabilities},
author = {C. N. Shangwe and A. Davoust and R. Khoury},
editor = {Adi K. and Nguena Timo O. and Boulahia-Cuppens N. and Espes D. and Stakhanova N. and Omar M.},
url = {https://www.scopus.com/pages/publications/105046105995?origin=resultslist},
doi = {10.1007/978-3-032-20732-6_2},
isbn = {03029743 (ISSN); 978-303220731-9 (ISBN)},
year = {2026},
date = {2026-01-01},
booktitle = {Lect. Notes Comput. Sci.},
volume = {16295 LNCS},
pages = {19–34},
publisher = {Springer Science and Business Media Deutschland GmbH},
abstract = {As Large Language Models (LLMs) become increasingly integrated into software development and analysis workflows, a critical question arises: do these models truly understand the semantics of code, or do they merely excel at pattern matching? Our goal is to assess the extent to which LLMs can back their predictions in vulnerability detection by correctly attributing the identified vulnerabilities to the violation of particular rules as proof that their decision is based on actual code semantics understanding. We employed the SVEN dataset, composed of function-level code snippets, to conduct a series of experiments that evaluate both the model’s ability to detect vulnerabilities and attribute predictions to the correct violated rule and measure LLMs’ performance under varying experimental setups. Our findings reveal that while LLMs achieve reasonable accuracy in vulnerability detection, a significant drop in performance is observed when correct rule attribution is also required, exposing a gap between perceived accuracy and actual accuracy. The difference between actual and perceived accuracy offers critical insight into the depth of code semantics understanding of LLMs in vulnerability detection. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2026.},
note = {Journal Abbreviation: Lect. Notes Comput. Sci.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
As Large Language Models (LLMs) become increasingly integrated into software development and analysis workflows, a critical question arises: do these models truly understand the semantics of code, or do they merely excel at pattern matching? Our goal is to assess the extent to which LLMs can back their predictions in vulnerability detection by correctly attributing the identified vulnerabilities to the violation of particular rules as proof that their decision is based on actual code semantics understanding. We employed the SVEN dataset, composed of function-level code snippets, to conduct a series of experiments that evaluate both the model’s ability to detect vulnerabilities and attribute predictions to the correct violated rule and measure LLMs’ performance under varying experimental setups. Our findings reveal that while LLMs achieve reasonable accuracy in vulnerability detection, a significant drop in performance is observed when correct rule attribution is also required, exposing a gap between perceived accuracy and actual accuracy. The difference between actual and perceived accuracy offers critical insight into the depth of code semantics understanding of LLMs in vulnerability detection. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2026. |
Tiwari, A.; Arrabito, R.; Davoust, A.; Falk, T. H. Bias in Physiology-Based Cognitive State Detection for Human–Autonomy Teaming: A Comprehensive Survey Proceedings Article In: R.A., Sottilare; J., Schwarz (Ed.): Lect. Notes Comput. Sci., pp. 139–158, Springer Science and Business Media Deutschland GmbH, 2026, ISBN: 03029743 (ISSN); 978-303230014-0 (ISBN), (Journal Abbreviation: Lect. Notes Comput. Sci.). @inproceedings{tiwariBiasPhysiologyBasedCognitive2026,
title = {Bias in Physiology-Based Cognitive State Detection for Human–Autonomy Teaming: A Comprehensive Survey},
author = {A. Tiwari and R. Arrabito and A. Davoust and T. H. Falk},
editor = {Sottilare R.A. and Schwarz J.},
url = {https://www.scopus.com/pages/publications/105044000677?origin=resultslist},
doi = {10.1007/978-3-032-30015-7_9},
isbn = {03029743 (ISSN); 978-303230014-0 (ISBN)},
year = {2026},
date = {2026-01-01},
booktitle = {Lect. Notes Comput. Sci.},
volume = {16737 LNCS},
pages = {139–158},
publisher = {Springer Science and Business Media Deutschland GmbH},
abstract = {Advances in artificial intelligence (AI) are changing the role of adaptive instructional systems (AIS) from passive tools, that deliver personalized learning pathways by adapting to student feedback, to active team members and co-learners that dynamically adapt alongside human partners. In this emerging landscape of human–autonomy teaming (HAT), assessing the cognitive states of individuals interacting with highly intelligent systems has become a critical design consideration for improving team performance, collaboration, and learning outcomes. Physiological signals have emerged as a popular method for measurement of cognitive states in the past few decades. However, these signals can be strongly influenced by different user demographics, including age and biological sex. If unaccounted for, these differences could introduce systematic biases into AIS based learning pathways for different demographic groups. These biases can lead to discriminative performance of instructional systems and may also leave them vulnerable to adversarial exploitation. Aside from physiological signal-induced biases, there may also be major behavioural differences between different sex and/or age groups when interacting with and teaming with AIS, thus further confounding cognitive state monitoring. In this survey, we review emerging human-autonomy teaming studies that incorporate physiological data and analyze where biological sex- and age-related physiological or behavioral differences were documented and/or analyzed. We highlight evidence demonstrating that differences do exist during interaction with autonomous systems. We then discuss how biases can accumulate across study design and analysis modelling pipelines, and provide practical guidelines for mitigating these biases in future human-autonomy teaming research and applications. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2026.},
note = {Journal Abbreviation: Lect. Notes Comput. Sci.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Advances in artificial intelligence (AI) are changing the role of adaptive instructional systems (AIS) from passive tools, that deliver personalized learning pathways by adapting to student feedback, to active team members and co-learners that dynamically adapt alongside human partners. In this emerging landscape of human–autonomy teaming (HAT), assessing the cognitive states of individuals interacting with highly intelligent systems has become a critical design consideration for improving team performance, collaboration, and learning outcomes. Physiological signals have emerged as a popular method for measurement of cognitive states in the past few decades. However, these signals can be strongly influenced by different user demographics, including age and biological sex. If unaccounted for, these differences could introduce systematic biases into AIS based learning pathways for different demographic groups. These biases can lead to discriminative performance of instructional systems and may also leave them vulnerable to adversarial exploitation. Aside from physiological signal-induced biases, there may also be major behavioural differences between different sex and/or age groups when interacting with and teaming with AIS, thus further confounding cognitive state monitoring. In this survey, we review emerging human-autonomy teaming studies that incorporate physiological data and analyze where biological sex- and age-related physiological or behavioral differences were documented and/or analyzed. We highlight evidence demonstrating that differences do exist during interaction with autonomous systems. We then discuss how biases can accumulate across study design and analysis modelling pipelines, and provide practical guidelines for mitigating these biases in future human-autonomy teaming research and applications. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2026. |
