Machine Learning and Transparency: A Scoping Exercise

Artificially Intelligent (AI) systems have become part of society. They support decision-making processes and take decisions autonomously. Due to their ability to analyze very large data sets, they can be superior to humans in certain aspects and they will have a major impact on people's lives. Tensions with societal, legal norms or constitutional values have already arisen. They can be observed in different scenarios, ranging from credit scoring and autonomous driving to social profiling and predictive policing. The lack of transparency of AI systems, which were considered as “black-boxes,” has already been identified as a problem. These so-called “black boxes” have been a weak spot for those who develop or use an AI system. This paper demystifies AI by discussing the role of transparency and explainability of machine learning. The paper describes different aspects of transparency and explains methods to increase our understanding of the behavior of AI systems.

The current discourse on fairness, accountability, and transparency in machine learning is driven by two competing narratives: sociotechnical dogmatism, which holds that society is full of inefficiencies and imperfections that can only be solved by better algorithms; and sociotechnical skepticism, which opposes many instances of automation on principle. Both perspectives, we argue, are reductive and unhelpful. In this chapter, we review a large, diverse body of literature in an attempt to move beyond this restrictive duality, toward a pragmatic synthesis that emphasizes the central role of context and agency in evaluating new and emerging technologies. We show how epistemological and ethical considerations are inextricably intertwined in contemporary debates on algorithmic bias and explainability. We trace the dialectical interplay between dogmatic and skeptical narratives across disciplines, merging insights from social theory and philosophy. We review a number of theories of explanation, ultimately endorsing a sociotechnical pragmatism that combines elements of Floridi’s levelism and Mayo’s reliabilism to place a special emphasis on notions of agency and trust. We conclude that this hybrid does more to promote fairness, accountability, and transparency in machine learning than dogmatic or skeptical alternatives.

The integration of machine learning (ML) in marine engineering has been increasingly subjected to stringent regulatory scrutiny. While environmental regulations aim to reduce harmful emissions and energy consumption, there is also a growing demand for the interpretability of ML models to ensure their reliability and adherence to safety standards. This research highlights the need to develop models that are both transparent and comprehensible to domain experts and regulatory bodies. This paper underscores the importance of transparency in machine learning through a use case involving a VLGC ship two-stroke propulsion engine. By adhering to the CRISP-DM standard, we fostered close collaboration between marine engineers and machine learning experts to circumvent the common pitfalls of automated ML. The methodology included comprehensive data exploration, cleaning, and verification, followed by feature selection and training of linear regression and decision tree models that are not only transparent but also highly interpretable. The linear model achieved an RMSE of 23.16 and an MRAE of 14.7%, while the accuracy of decision trees ranged between 96.4% and 97.69%. This study demonstrates that machine learning models for predicting propulsion engine fuel consumption can be interpretable, adhering to regulatory requirements, while still achieving adequate predictive performance.

This paper intends to provide a better understanding of how these models produce predictions, particularly in complex medical diagnoses, and at the same time bridging the gap between technical model outputs and clinical applications. This study addresses the critical problem of transparency of machine learning (ML) models in health care where interpretability is an essential aspect for ethical decision making and trust building.The goal of this paper is to present a clearer understanding of how these models generate predictions, especially in important fields like complex medical diagnoses, thereby bridging the gap between technical model outputs and clinical applications. The crucial issue of transparency in machine learning (ML) models within healthcare is addressed in this study where interpretability plays a vital role in ethical decision-making and fostering trust. The focus of the research is enhancing model transparency by using SHapley Additive exPlanations (SHAP) and Deep Learning Important FeaTures (DeepLIFT), two crucial methods that are designed to elucidate the decision-making processes of ML models.This mode of approach helps to have more distributed comprehension of the decision pathways by models thus aiding in knowing how each feature contributed to the last prediction. It is this method that has been employed to showcase the efficiency of predicting melanoma and also diabetic retinopathy which are two vital medical diagnostic areas. In healthcare, SHAP along with DeepLIFT has improved the models’ explainability and trustworthiness significantly and hence making them easy for those in the field. The advanced interpretability methods presented in this document enhances ML model transparency especially when dealing with health issues. As a result, interpretability becomes an even bigger issue and they are supposed to be able to use these tools for reliable and open decisions when it comes to medical specialists.

The fact that machine learning is growing more and more entrenched in almost every aspect of society, combined with the opacity of various of its algorithms has induced the relatively young research area of transparent machine learning. The aim of this domain is to provide explanations for automated decisions to increase public trust. In my thesis, I am going to consider certain problems that arise from this research agenda. Particularly, I so far considered, the dilemma of conflicting explanations and the issue of privacy concerns arising from transparency.

Regulation of AI and AI Crimes

The convergence of AI and cultural representation raises critical ethical concerns in the digital age, particularly in culturally diverse and complex societies like Nigeria. This paper takes a conceptual methodological stance to explore the intersection of artificial intelligence (AI) technologies, such as language models and translation systems, in relation to Nigerian cultural and linguistic realities by highlighting the power dynamics of digital language preservation and the risk of reinforcing stereotypes or erasing nuance. Using a bipartite framework involving Floridi’s Information Ethics and the Fairness, Accountability, and Transparency in Machine Learning (FAT/ML) model, this study examines how artificial intelligence (AI) systems, including ChatGPT and Google Translate, may misrepresent or distort Nigerian languages and cultures, such as Yoruba, Hausa, Igbo, and Pidgin. The study advocates for the development of AI that is culturally aware and ethically grounded to promote inclusive frameworks that prioritize linguistic equity and cultural sustainability in the digital representation of Nigerian identities. This research contributes to the development of responsible AI practices that respect and promote cultural diversity in the digital age by exploring the global relevance of these issues.

With the rapid growth of artificial intelligence applications for the analysis of data and the design and operations of information systems that support the delivery of public services, public administrators are faced with some daunting challenges and novel opportunities. In order to understand what these challenges and opportunities are, they must first have a sense of some of the promises and policy discussions that are shaping present and future applications of artificial intelligence (AI). This chapter provides an overview of the fundamental features of one form of AI, machine learning, and the greater potential applications of AI for public policy and service delivery. An overview of several national-level policy documents (including the United States, European Union, and United Kingdom) are reviewed and synthesized. The forms of AI systems in relation to human interactions and role of administrative discretion are discussed, and recommendations for practice are provided.

From the Editor's Desk: Systems Science 2010-2020, and Post-COVID-19.

The rapid evolution of e-commerce has propelled personalization to the forefront of digital marketing strategies. This study investigates the transformative impact of Artificial Intelligence (AI) and Machine Learning (ML) on e-commerce personalization, examining their effects on customer engagement, sales performance, and long-term market dynamics. We conducted a comprehensive econometric analysis using panel data from diverse e-commerce platforms over multiple years. Employing difference-in-differences models and instrumental variable approaches, we isolated the specific impact of AI and ML-driven personalization on key performance metrics. Our research encompassed various product categories and market segments to assess heterogeneous effects across the e-commerce landscape. The implementation of AI and ML-driven personalization strategies led to statistically significant increases in conversion rates (10-15%) and customer lifetime value (20-30%). These insights have significant implications for e-commerce strategy, investment decisions, and regulatory frameworks in an increasingly AI-driven economy. This research contributes to the growing body of literature on AI economics, providing empirical evidence of its impact in e-commerce. It offers valuable guidance for practitioners in optimizing personalization strategies and informs policy discussions surrounding digital market regulation, data privacy, and technological competition.

Recent military use of machine learning tools for target identification has raised concerns about how these systems affect compliance with International Humanitarian Law (IHL). Decision support systems such as Israel's Lavender have shown that operators may rely too heavily on algorithmic recommendations, especially during time pressure. This reliance can speed up strike decisions while making it harder to trace the source of errors. IHL rules on distinction, proportionality, necessity and precaution were developed for human decision makers who can understand and question the basis of an attack. However, contemporary targeting software produces recommendations that are difficult to audit and may reflect hidden biases in training data. This article highlights four gaps in current scholarship: limited analysis of non-autonomous decision support tools, the lack of behavioural research on automation bias in military contexts, weak transparency in national Article 36 weapons reviews and insufficient attention to technical safeguards that could prevent errors before they occur. The study proposes a combined legal and socio-technical approach to examine how these systems function across their entire life cycle. It evaluates their practical effects on IHL compliance and outlines reforms in doctrine, review processes and system design. The goal is to ensure clearer accountability and stronger protection for civilians as algorithmic tools become a routine part of modern warfare.

(1) Background: This study tracked the reporting of obesity in the Australian news media over three decades and how changing representations over time were linked to obesity-related public health policy developments. (2) Methods: Machine learning and computational language analysis techniques (word embedding, dichotomous bias mapping) were used to identify language biases associated with obesity in 157,237 relevant articles drawn from the Australian Dow Jones digital database of print news media articles from 1990 to 2019. (3) Results: Obesity-related terms were stigmatised on four key dimensions (gender, health, socioeconomic status, stereotypes), with language biased towards femininity and lower socioeconomic status in particular. Biases remained relatively steady from 2005 to 2019, despite recent policy initiatives directly seeking to address obesity stigma. To some degree, for each of the four dimensions, cosine values moved toward 0 over time (i.e., no association with one dimension poll or the other), but remained around 0.20. There was a strong relationship between news media and public health policy discourse over the 30-year study period. (4) Conclusions: With increasing recognition of the health consequences of weight stigma, policymakers and the media must work together to ensure public weight management narratives avoid discourse that may stigmatise heavier individuals, particularly women, and/or reinforce negative obesity stereotypes.

ABSTRACTAccurately predicting individual antidepressant treatment response could expedite the lengthy trial‐and‐error process of finding an effective treatment for major depressive disorder (MDD). We tested and compared machine learning‐based methods that predict individual‐level pharmacotherapeutic treatment response using cortical morphometry from multisite longitudinal cohorts. We conducted an international analysis of pooled data from six sites of the ENIGMA‐MDD consortium (n = 262 MDD patients; age = 36.5 ± 15.3 years; 154 (59%) female; mean response rate = 57%). Treatment response was defined as a ≥ 50% reduction in symptom severity score after 4–12 weeks post‐initiation of antidepressant treatment. Structural MRI was acquired before, or < 14 days after, treatment initiation. The cortex was parcellated using FreeSurfer, from which cortical thickness and surface area were measured. We tested several machine learning pipeline configurations, which varied in (i) the way we presented the cortical data (i.e., average values per region of interest, as a vector containing voxel‐wise cortical thickness and surface area measures, and as cortical thickness and surface area projections), (ii) whether we included clinical data, and the (iii) machine learning model (i.e., gradient boosting, support vector machine, and neural network classifiers) and (iv) cross‐validation methods (i.e., k‐fold and leave‐one‐site‐out) we used. First, we tested if the overall predictive performance of the pipelines was better than chance, with a corrected 10‐fold cross‐validation permutation test. Second, we compared if some machine learning pipeline configurations outperformed others. In an exploratory analysis, we repeated our first analysis in three subpopulations, namely patients (i) from a single site, (ii) with comparable response rates, and (iii) showing the least (first quartile) and the most (fourth quartile) treatment response, which we call the extreme (non‐)responders subpopulation. Finally, we explored the effect of including subcortical volumetric data on model performance. Overall, performance predicting antidepressant treatment response was not significantly better than chance (balanced accuracy = 50.5%; p = 0.66) and did not vary with alternative pipeline configurations. Exploratory analyses revealed that performance across models was only significantly better than chance in the extreme (non‐)responders subpopulation (balanced accuracy = 63.9%, p = 0.001). Including subcortical data did not alter the observed model performance. Cortical structural MRI alone could not reliably predict individual pharmacotherapeutic treatment response in MDD. None of the used machine learning pipeline configurations outperformed the others. In exploratory analyses, we found that predicting response in the extreme (non‐)responders subpopulation was feasible on both cortical data alone and combined with subcortical data, which suggests that specific MDD subpopulations may exhibit response‐related patterns in structural data. Future work may use multimodal data to predict treatment response in MDD.

Cost overruns remain one of the most persistent challenges in construction and infrastructure project management, often undermining efficiency, sustainability, and stakeholder trust. With the rise of digital transformation, artificial intelligence (AI) and machine learning (ML) provide new opportunities to enhance predictive decision-making and strengthen project control. This study introduces a digital project management framework that integrates the Pelican Optimization Algorithm (POA) with Light Gradient Boosting Machine (LGBM) to deliver reliable and interpretable cost overrun forecasting. The proposed POA-LightGBM model leverages metaheuristic-driven hyperparameter optimization to improve predictive performance and generalization. A comprehensive evaluation using multiple error metrics Coefficient of Determination (R2), Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), and Mean Absolute Percentage Error (MAPE) demonstrates that POA-LGBM significantly outperformed baseline LGBM and alternative metaheuristic configurations, achieving an average R2 of 0.9786. To support transparency in digital project environments, SHapley Additive exPlanations (SHAPs) were employed to identify dominant drivers of cost overruns, including actual project cost, energy consumption, schedule deviation, and material usage. By embedding AI-enabled predictive analytics into digital project management practices, this study contributes to advancing digital transformation in project delivery, offering actionable insights for cost control, risk management, and sustainable infrastructure development.

Owing to the global lockdowns that resulted from the COVID-19 pandemic, fuel demand plummeted and the price of oil futures went negative in April 2020. Robust fuel demand projections are crucial to economic and energy planning and policy discussions. Here we incorporate pandemic projections and people’s resulting travel and trip activities and fuel usage in a machine-learning-based model to project the US medium-term gasoline demand and study the impact of government intervention. We found that under the reference infection scenario, the US gasoline demand grows slowly after a quick rebound in May, and is unlikely to fully recover prior to October 2020. Under the reference and pessimistic scenario, continual lockdown (no reopening) could worsen the motor gasoline demand temporarily, but it helps the demand recover to a normal level quicker. Under the optimistic infection scenario, gasoline demand will recover close to the non-pandemic level by October 2020.