This study constructs a dual-cycle framework integrating financial fraud identification and ethical governance for artificial intelligence-assisted audit. It uses financial indicators, text features, and corporate governance variables to build multi-dimensional detection models. Traditional models and machine learning models including tree ensemble, explainable, and review-adjusted models are compared. Empirical results show that machine learning models outperform traditional models in fraud identification accuracy and stability. The explainable module and manual review can effectively reduce ethical risks such as opacity, bias amplification, over-reliance, and responsibility ambiguity. This work balances technical performance and ethical compliance, reveals ethical challenges of intelligent audit applications, and provides a governance path for the standardized application of artificial intelligence in auditing.
The rapid advancement of artificial intelligence (AI) has endowed society with unprecedented decision-making capabilities while also raising profound ethical challenges. In the context of increasingly autonomous AI systems, balancing fairness, explainability, and accountability has become a frontier issue in technology ethics. Grounded in Kant's moral philosophy, this study introduced the universalization principle and ends-in-themselves principle into AI mechanism design and evaluation. By synthesizing existing scholarship and analyzing typical scenarios, an ethical constraint model was constructed for universalization testing, rights protection, and adaptive feedback. Using public datasets and real-world scenarios, the model was evaluated according to the criteria of fairness, explainability, and risk control. Findings show that internalizing Kantian ethics enhances system autonomy under moral norms and provides a robust pathway for algorithmic governance in complex social contexts. Integrating theory with practice, these innovative technical solutions offer guidance for optimizing future ethical algorithms.
Existing human resources training programs in most e-commerce companies fail to adequately prepare employees to use emerging technologies effectively against the growing sophistication of card-not-present (CNP; a type of fraud in which a stolen credit card or debit card is used for transactions conducted without the physical card) fraud. To fill the gap, this study proposes an artificial intelligence-enabled gamified training framework (AI-GTF; a system combining AI techniques with gamified learning methods to improve user engagement) to enhance the capabilities of e-commerce workforces for CNP fraud detection. A mixed-method research design is applied to achieve the research aim. Questionnaire data collected from 295 e-commerce workforces across Asia were analyzed using exploratory factor analysis. As a result, key components of AI-GTF were established to effectively detect CNP fraud in e-commerce. Afterwards, an interview was conducted with 10 e-commerce human resources personnel, ascertaining relevance, usability, and perceived usefulness of the proposed AI-GTF in terms of enhancing the capabilities of e-commerce workforces for CNP fraud detection. This study advances human resources practices that enhance the ability to detect sophisticated CNP fraud activities in timely manner, addressing critical organizational risks in e-commerce.
To address ethical and safety risks including misleading marketing, minor exposure, and algorithmic opacity in artificial intelligence (AI) applications of the tobacco industry, this study proposed a technology-ethics dual-driven safety framework. It adopted a three-layer architecture: (1) a risk perception layer using multimodal sensing and federated learning for real-time detection and privacy preservation; (2) an ethical decision-making layer with a dynamic rule engine, explainable AI, and adversarial training to mitigate algorithmic bias; and (3) an audit traceability layer based on blockchain for immutable whole-process logging and accountability. Validated on the TobaccoComply-21K dataset and real-world industry data, the framework reached 98.7% compliance detection accuracy, a 0.8% weighted false positive rate, and a 95-millisecond audit delay. As the first work integrating federated learning, explainable AI, and blockchain audit into a tobacco-tailored full-chain ethical governance system, it promoted responsible AI application in highly regulated sectors.
Drones offer great potential as intelligent mobility solutions, yet challenges remain to their successful integration in society. This study analyzed experts' perspectives regarding public acceptance of urban drones in the Swiss context, with the aim of improving the understanding of the underlying values affecting acceptance. Trust emerged as the key to public perceptions, and critical to establishing trust is connecting more deeply with the public through integrating trust-building mechanisms, aligning with societal values resonating with the public, and involving a wider range of experts in shaping governance strategies. As conclusion, a responsible innovation framework and more effective implementation approaches can provide additional insights into how to build trust around an accepted set of values that serve the public good. This points towards an important next step-developing a process enabling consensus to be reached and ethical standards to be established.
