ValueXplain

Research Foundations

Peer-reviewed research underpinning ValueXplain

FoundationJournal article

Toward explaining value concerns in software artifacts: A human-in-the-loop method

Information and Software Technology

Read publication
See details

Authors

Davoud Mougouei, Son Pham, Ahmad Azarnik, Mahdi Fahmideh, Arif Nurwidyantoro, Elahe Mougouei, Amanul Haque, Hoa Khanh Dam, Saima Rafi, Javed Ali Khan

Summary

Context: Violations of human values in software can lead to user dissatisfaction, economic harm, and loss of trust. Existing Software Engineering research has explored detecting value violations through manual or automated labeling (classification) that merely detects their presence or absence without explaining the underlying mechanisms of violations (the how). This makes it challenging to verify the reliability of such findings and limits their practicality. Objectives: This study introduces and evaluates a human-in-the-loop method that produces explanatory accounts of conditions that may lead to violations of human values in software, referred to as Value Concerns. Methods: The proposed human-in-the-loop method integrates the automated reasoning capabilities of a Large Language Model (LLM) – ChatGPT – with human assessment to explain value violations in software artifacts. A case study of 1200 Android and iOS APIs is used to examine the method’s ability to identify and articulate value concerns. Results: Our findings show that ChatGPT’s inferences about value concerns are reasonably accurate yet incomplete. Inaccuracies arise from unfounded explanations, extended reasoning, and persuasive linguistic effects. The human-in-the-loop cycles mitigate these issues by reviewing and refining the model’s outputs. The case study supports the use of LLMs for transparent reasoning about value violations. Conclusion: Large Language Models, when guided by human oversight, can function effectively as explanatory tools and reasoning partners for value violations in software artifacts. Rather than serving as classifiers, they can generate structured explanatory accounts of conditions that may lead to value violations in software.

FoundationJournal article

Exploring value concerns in social media issue reports: a human-LLM study of Bluesky and Mastodon

Behaviour & Information Technology

Read publication
See details

Authors

Davoud Mougouei, Ahmad Azarnik, Elahe Mougouei, Son Pham, Niamh Kinchin, Sherrica Anne Senewiratne

Summary

Design choices in social media platforms have repeatedly undermined human values, as highlighted by the recent U.S. Senate hearing on child safety failures in major platforms. Issue-tracking systems like GitHub Issues serve as primary venues for discussing these design decisions, revealing trade-offs that may affect user values. This paper examines the GitHub issue reports of two popular decentralised social media platforms, Bluesky and Mastodon, to identify scenarios that serve as emerging evidence of undermining values, i.e. Value Concerns. We select these platforms due to their focus on values like autonomy, transparency, and privacy. We utilise the reasoning capabilities of Large Language Models (LLMs) for inference of value concerns. The inferred concerns are iteratively refined and then evaluated by human evaluators from computing and social sciences; the endorsed concerns are categorised to reveal mechanisms of undermining values in social media, offering insights for mitigating them.

FoundationJournal article

Reasoning About Human Values in GitHub Issues: What Can a Large Language Model Reveal?

International Journal of Human–Computer Interaction

Read publication
See details

Authors

Davoud Mougouei, Amanul Haque, Arif Nurwidyantoro, Elahe Mougouei, Mahdi Fahmideh, Mojtaba Shahin, Andi Dharmawan, Triyogatama Wahyu Widodo, Hoa Khanh Dam

Summary

The recent U.S. Senate hearing on child safety failures in major social media platforms, gender bias in Amazon’s recruitment algorithms, and privacy lawsuits against Google over vague location tracking illustrate how software can neglect human values with societal and economic consequences. Addressing such oversights requires understanding how design choices impact values. GitHub Issues provide a key venue where these choices are documented: through feature requests, improvements, and bug reports, they capture both technical complexities and the tradeoffs reflecting stakeholders’ beliefs. They thus offer insights into how values such as privacy, fairness, and inclusivity influence and are shaped by software design. Yet inferring value alignments, often implied rather than explicit, demands complex reasoning beyond keyword searches and cannot be done manually at scale. In this paper, we use a Large Language Model (LLM) to infer (detect and explain) the alignment of GitHub Issues with human values across three open-source projects, and evaluate the accuracy of the findings through human evaluation.

FoundationJournal article

A Human-LLM study of value concerns in smart home app reviews

Internet of Things

Read publication
See details

Authors

Davoud Mougouei, Son Pham, Ahmad Azarnik, Elahe Mougouei, Arif Nurwidyantoro, Hilya Mudrika Arini, Fitri Trapsilawati, Rizqi Nur'aini A'yuninnisa

Summary

Amazon’s $31 million settlement over privacy violations involving its Alexa voice assistant and Ring cameras highlights how neglecting human values in smart-home platforms can cause financial, reputational, and societal harm. This study examines how breaches of human values manifest in everyday use of smart-home apps. We analyze app reviews of three major platforms – Google Home, Amazon Alexa, and Samsung SmartThings – to collect firsthand accounts of design choices that impact user values. Using a large language model (LLM), we infer structured scenarios, referred to as value concerns , that explain how and why potential value violations arise. The inferred value concerns are refined for consistency and assessed by evaluators from computing and social sciences to ensure their factuality, plausibility, and specificity. Our analysis of value concerns uncovers recurring mechanisms by which smart-home designs can threaten a wide range of human values, offering practical insights to mitigate these impacts.

FoundationConference paper

A First Look at AI Trends in Value-Aligned Software Engineering Publications: Human-LLM Insights

2025 IEEE/ACM 47th International Conference on Software Engineering: Software Engineering in Society (ICSE-SEIS)

Read publication
See details

Authors

Davoud Mougouei, Ahmad Azarnik, Mahdi Fahmideh, Elahe Mougouei, Hoa Khanh Dam, Arif Ali Khan, Saima Rafi, Javed Ali Khan, Aakash Ahmad

Summary

Recent criticism of social media platforms by the U.S. Senate Judiciary Committee for neglecting child safety exemplifies how software can undermine human values. This is further complicated by the growing integration of Artificial Intelligence (AI) in software, which introduces inherent challenges such as biases and limited transparency. However, AI also presents opportunities to embed human values into software. To explore these opportunities, we have utilized the reasoning abilities of ChatGPT, a large language model (LLM), in combination with human expertise, to study the use of AI in publications that address human values, across some of the leading software engineering (SE) venues from 2022 to 2023. Our findings confirm the use of AI concepts - mainly General Machine Learning - in around 33 % of these valuealigned publications. The value alignments largely concern pragmatic aspects of Achievement and (personal) Security, while the majority of the values receive less attention. The socially focused values of Conformity and Tradition and the personally focused value of Hedonism are rarely addressed in the SE publications.

FoundationJournal article

Venturing ChatGPT's lens to explore human values in software artifacts: a case study of mobile APIs

Behaviour & Information Technology

Read publication
See details

Authors

Davoud Mougouei, Saima Rafi, Mahdi Fahmideh, Elahe Mougouei, Javed Ali Khan, Hoa Khanh Dam, Arif Nurwidyantoro, Michel Chaudron

Summary

Software is designed for humans and must account for their values. However, current research and practice focus on a narrow range of well-explored values, e.g. security, overlooking a more comprehensive perspective. Those exploring a broader array of values rely on manual identification, which is labour-intensive and prone to human bias. Moreover, existing methods offer limited reliability as they fail to explain their findings. In this paper, we propose leveraging the reasoning capabilities of Large Language Models (LLMs) for automated inference about values. This allows for not only detecting values but also explaining how they are expressed in the software. We aim to examine the effectiveness of LLMs, specifically ChatGPT (Chat Generative Pre-Trained Transformer), in automated detection and explanation of values in software artifacts. Using ChatGPT, we investigate how mobile APIs align with human values based on their documentation. Human evaluation of ChatGPT's findings shows a reciprocal shift in understanding values, with both ChatGPT and experts adjusting their assessments through dialogue. While experts recognise ChatGPT's potential for revealing values, emphasis is placed on human involvement to enhance the accuracy of the findings by detecting and eliminating convincing but inaccurate explanations provided by the language model due to potential hallucinations or confabulations.

FoundationConference paper

Engineering Human Values in Software through Value Programming

Proceedings of the IEEE/ACM 42nd International Conference on Software Engineering Workshops

Read publication
See details

Authors

Davoud Mougouei

Summary

Ignoring human values in software development may disadvantage users by breaching their values and introducing biases in software. This can be mitigated by informing developers about the value implications of their choices and taking initiatives to account for human values in software. To this end, we propose the notion of Value Programming with three principles: (P1) annotating source code and related artifacts with respect to values; (P2) inspecting source code to detect conditions that lead to biases and value breaches in software, i.e., Value Smells; and (P3) making recommendations to mitigate biases and value breaches. To facilitate value programming, we propose a framework that allows for automated annotation of software code with respect to human values. The proposed framework lays a solid foundation for inspecting human values in code and making recommendations to overcome biases and value breaches in software.

FoundationConference paper

Operationalizing human values in software: a research roadmap

Proceedings of the 2018 26th ACM Joint Meeting on European Software Engineering Conference and Symposium on the Foundations of Software Engineering

Read publication
See details

Authors

Davoud Mougouei, Harsha Perera, Waqar Hussain, Rifat Shams, Jon Whittle

Summary

Software influences several aspects of people's lives and therefore needs to reflect their values. However, existing software engineering methods fail to account for human values, which may result in breaching those values in software and, therefore, dissatisfaction of users and loss of profit and reputation. To avoid such negative consequences, human values need to be integrated -- in a verifiable way -- into software. We refer to this as Operationalizing Human Values in Software. But this is not easy to achieve due to three main obstacles: first, human values are hard to define in a way that can be put into practice; second, existing software design decisions are mainly ignorant of values; finally, values are hard to determine and quantify in software. This paper aims to establish a research roadmap for overcoming these obstacles. The proposed roadmap focuses on (i) establishing practical definitions for human values, (ii) integrating values into software design, and (iii) measuring values in the software development life cycle.