Open Access

Article

Article ID: 3893

Verifying artificial intelligence-generated images: Socio-technical approaches to authenticity

by Michael Mncedisi Willie

Computing and Artificial Intelligence, Vol.3, No.4, 2025;

The rapid proliferation of artificial intelligence (AI) has transformed visual media, enabled highly realistic AI-generated images, and raised ethical, social, and security concerns. Generative artificial intelligence (Generative AI) architectures, including Generative Adversarial Networks (GANs), Variational Autoencoders (VAEs), and diffusion models, allow content creation that is increasingly indistinguishable from human-made visuals, facilitating creativity, education, and communication. However, these capabilities also introduce risks of manipulation, identity fraud, misinformation, and deepfake attacks across social, political, corporate, academic, and humanitarian domains. This study investigates AI image verification as a socio-technical response to synthetic visuals, focusing on social media, artistic, and forensic contexts. It employed a qualitative design combining thematic literature review and case study analysis. Thematic analysis identified patterns in verification approaches, including pixel-level analysis, metadata forensics, machine learning classifiers, watermarking, and blockchain-enabled methods. Case studies explored real-world applications, highlighting perceptual biases, strategic use of synthetic content, and governance and digital literacy challenges. Findings reveal that human perception alone is insufficient for reliably discerning authenticity, with individuals frequently misclassifying AI-generated images as real. Integrating machine learning, metadata analysis, and blockchain verification, hybrid technical approaches significantly improve detection accuracy. Socio-technical factors, including platform policies, ethical norms, organisational governance, and user literacy, shape the effectiveness of verification methods. The study presents a conceptual framework linking technological, organisational, and societal dimensions, emphasising the need for coordinated strategies that combine algorithmic innovation, regulatory oversight, and public engagement. Practical implications include deploying hybrid verification systems, strengthening governance and ethical standards, enhancing digital literacy, and fostering cross-disciplinary collaboration to safeguard trust, authenticity, and integrity in digital media.

Open Access

Article

Article ID: 3781

Triadic integration of Artificial Intelligence: Bridging strategy, research, and operational systems

by Michael Mncedisi Willie

Computing and Artificial Intelligence, Vol.3, No.3, 2025;

Artificial Intelligence (AI) has emerged as a transformative enabler across strategic management, qualitative research, and crowdsourced operational systems. However, adoption is shaped by human judgement, organisational processes, and socio-technical factors. Existing literature often examines AI applications in isolation, overlooking integrative approaches that balance technical capability with human and ethical oversight. This study systematically synthesises evidence to examine AI’s impact across multiple domains, identifying patterns, limitations, and opportunities, and proposes a human-centred framework for responsible deployment. A systematic integrative review was conducted, encompassing peer-reviewed journals, technical reports, and policy documents. Data extraction focused on AI capabilities, human-AI interaction, governance, methodological rigour, and socio-technical integration. Thematic analysis identified recurring patterns and gaps across domains. This study reveals that AI-driven decision-support systems enhance predictive analytics, scenario planning, and resource allocation, yet require managerial expertise, governance, and interpretive oversight to translate insights into actionable strategy. Furthermore, AI-assisted tools improve thematic analysis, coding, and data synthesis efficiency, but human interpretation remains critical to maintain contextual depth, methodological rigour, and ethical integrity. Lastly, Platforms such as Waze and Google Maps demonstrate real-time operational value, yet outcomes are contingent on data quality, user engagement, and trust, highlighting the socio-technical dependencies of AI deployment. The Triadic AI Integration Framework (TAIF) operationalises these insights by linking AI capabilities, human interpretation, and organisational processes within a human-centred, ethically governed structure. Effective AI adoption requires interpretive oversight, socio-technical alignment, and cross-domain integration to maximise strategic, research, and operational impact. Future research should empirically test TAIF, explore socio-technical adaptation, and examine long-term organisational and societal outcomes.

Open Access

Article

Article ID: 2342

Joint weight adversarial attack based on human skeleton action recognition

by Haopeng Mu, Ling Guo, Xiaozhou Zhang

Computing and Artificial Intelligence, Vol.3, No.3, 2025;

In recent years, due to the excellent spatial information correlation, small data volume and high computational efficiency of bone data, it has been widely applied in action recognition fields such as autonomous driving and intelligent security. However, in practical applications, attackers only need to apply a small perturbation to the input bone data to cause the attacked model to make incorrect recognition of the corresponding action, thereby resulting in a significant drop in recognition accuracy and even potentially causing serious consequences in high-risk scenarios such as autonomous driving. To solve this problem, many attack methods have been proposed, such as attacks that limit the angle changes between bones or attacks that alter the length of bones. These methods can, to a certain extent, increase the attack success rate of action recognition models, but most of these methods attack the bone data by simply disregarding the influence of each joint bone node on the overall action. In this paper, we propose a new adversarial attack method, that is, to attack through interfering with the coordinate data of the entire skeletal joint nodes. In our method, the concept of joint weights is proposed, and a time cropping translation attack is designed based on joint weights to improve the attack success rate. We conducted experiments on our method. The experimental results show that our attack success rate is stable at over 60%.

Open Access

Article

Article ID: 2601

Resources management and execution mechanisms for thinking operating system

by Ping Zhu, Pohua Lv, Weiming Zou, Xuetao Jiang, Jin Shi, Yang Zhang, Yirong Ma

Computing and Artificial Intelligence, Vol.3, No.2, 2025;

To achieve interpretable machine intelligence surpassing human cognitive levels and realize the ultimate objective of co-evolutionary human-computer interactions, this article analyzed various related aspects such as the human-computer interaction process, knowledge base construction, visual programming tool development, and thinking operating system design. This article proposed a method for simulating human thinking processes by computer: Firstly, it clarified the route by starting from the “teaching and learning” mode, which was the human-computer interaction computing mode, enabling the gradual accumulation of knowledge and data, and established the thinking knowledge base. Secondly, it established human thinking simulation mechanisms on the thinking operation system, including state perception, common sense judgment, error rollback, static logic structure analysis for the programs, and dynamic execution path analysis. Thirdly, it discussed the computer implementation methods of the thinking operation system and applications in detail, using mechanisms such as autonomous enumeration and rule induction of input data features, common sense judgment rollback, automatic error self-healing, online self-programming, and system adaptation (generalized pattern matching); all the above mechanisms were commonly used in human thinking. Finally, it summarized the whole article, and the future research directions were proposed by the authors.

Open Access

Article

Article ID: 2300

Semantic backpropagation: Extending symbolic network effects to achieve non-linear scaling in semantic systems

by Andy E. Williams

Computing and Artificial Intelligence, Vol.3, No.2, 2025;

Addressing humanity’s most complex challenges—such as poverty, climate change, and systemic inequality—requires solutions that scale non-linearly with their key variables. Traditional symbolic-level backpropagation algorithms, which power neural networks, achieve non-linear scaling through hierarchical feature extraction. However, these algorithms are constrained by their reliance on symbolic representations and numeric optimization, limiting their applicability to context-rich, real-world systems. This paper introduces semantic backpropagation, a novel extension of symbolic backpropagation, designed to operate on semantic representations that encode richer contextual and relational information. We hypothesize that (1) symbolic-level network effects can be generalized and replicated at the semantic level through semantic backpropagation algorithms, and (2) the non-linear scaling observed in symbolic backpropagation can also be achieved in semantic systems. To test these hypotheses, we developed a simulation framework that dynamically constructs, evaluates, and optimizes networks of interventions, such as value chains, using semantic query loops and iterative fitness optimization. The results demonstrate that semantic backpropagation demonstrates the potential to replicate symbolic-level network effects and achieve non-linear scaling through cooperative semantic interactions. Collaborative idea generation within this framework produced an exponential increase in the number and impact of business ideas compared to independent idea generation, providing initial evidence for the potential of semantic backpropagation to address multi-dimensional challenges. This work bridges the paradigms of symbolic precision and semantic richness, offering a powerful new tool for designing decentralized collective intelligence systems and solving global problems at scale. Semantic backpropagation provides a theoretical and practical foundation for leveraging semantic-level network effects to exponentially enhance the impact of human and AI collaboration. This work does not claim to present final empirical validation. Rather, it defines and tests a generative framework whose full implementation lies beyond current infrastructure. It proposes a theory of recursive semantic coherence whose feasibility must be evaluated not by external metrics alone, but by its ability to generate conceptual resolution and future testable models across domains.

Open Access

Article

Article ID: 3104

Lightweight weighted average ensemble model for pneumonia detection in chest X-ray images

by Suresh Babu Nettur, Shanthi Karpurapu, Unnati Nettur, Likhit Sagar Gajja, Sravanthy Myneni, Akhil Dusi, Lalithya Posham

Computing and Artificial Intelligence, Vol.3, No.2, 2025;

Pneumonia is a leading cause of illness and death in children, underscoring the need for early and accurate detection. In this study, we propose a novel lightweight ensemble model for detecting pneumonia in children using chest X-ray images. Our main contribution lies in the development of a novel, particularly weighted average ensemble model that combines two lightweight pre-trained convolutional neural networks (CNNs), MobileNetV2 and NASNetMobile, an ensemble combination that has not been previously explored in the field of deep learning for image classification tasks. These models were selected for their balance of computational efficiency and accuracy, fine-tuned on a pediatric chest X-ray dataset, and combined to enhance classification performance. The proposed ensemble model achieved a classification accuracy of 98.63%, significantly outperforming individual models such as MobileNetV2 (97.10%) and NASNetMobile (96.25%) in terms of accuracy, precision, recall, and F1 score. Moreover, the ensemble model outperformed state-of-the-art architectures, including ResNet50, InceptionV3, and DenseNet201, while maintaining computational efficiency. The proposed lightweight weighted average ensemble model presents a highly effective and resource-efficient solution for pneumonia detection, making it particularly suitable for deployment in resource-constrained settings.