Editorial: Explainability and transparency of artificial intelligence: emergence of AI forensics
Year: 2026
Type: editorial
Abstract: Explainability and transparency of artificial intelligence: emergence of AI forensicsTo engender trust in artificial intelligence (AI) systems, there is a growing need for AI forensics so that we can systematically examine, validate and explain how decisions and outputs are generated by AI models.Organizations are increasingly adopting AI models for automation of routine tasks, enhancing decision-making, improving operational efficiency and uncovering insights from complex datasets.Generative AI and large language models are now enabling organizations to perform knowledge work and augment employee productivity.In a lot of highly regulated industries such as healthcare, finance, cybersecurity, auditing and legal decision-making, stakeholders require mechanisms to verify model behavior, identify sources of bias, assess reliability and investigate failures such as hallucinations, erroneous recommendations or unexpected outcomes (Doshi-Velez and Kim, 2017; European Commission High-Level Expert Group on AI, 2019).AI models are often highly complex and opaque, making it difficult to reconstruct decision pathways or determine the factors influencing a specific output (Adadi and Berrada, 2018;Rudin, 2019).The lack of transparency in AI models, limited access to training data, model drift over time and the probabilistic nature of generative AI further complicate investigations (Mitchell et al., 2019;Bommasani et al., 2021).Establishing accountability, preserving audit trails, ensuring reproducibility and developing standardized methodologies for collecting and evaluating evidence remain ongoing challenges (Brundage et al., 2020; NIST, 2023).Addressing these issues is essential for improving trust, regulatory compliance and the responsible deployment of AI technologies and falls under the general rubric of AI forensics.AI forensics is an emerging discipline that focuses on examining, validating and explaining the behavior and outputs of AI systems.It involves understanding how AI models arrive at their decisions, identifying sources of error or bias, detecting faults in model design or deployment and investigating hallucinations in the models (Doshi-Velez and Kim, 2017;Linardatos et al., 2021;Ji et al., 2023).AI forensics emphasizes transparency, explainability and accountability by reconstructing decision pathways, analyzing training data, training algorithms and trained models and ascertaining the reliability and accuracy of AI models (Ribeiro et al., 2016;Lundberg and Lee, 2017;Brundage et al., 2020).AI forensics is critical to assurance, governance and auditability, which seek to establish mechanisms for monitoring, verifying and explaining AI behavior throughout the system lifecycle (M€ okander et al., 2023; OECD, 2019).As AI systems become increasingly integrated into high-stakes domains such as healthcare, finance, cybersecurity and auditing, AI forensics provides a critical framework for ensuring trust, regulatory compliance and the responsible use of intelligent systems by enabling the systematic investigation of model decisions, failures and unintended consequences (Floridi and Cowls, 2019; European Commission High-Level Expert Group on AI, 2019).Often, explainability and transparency are used in the same breath in the context of AI; they are related but distinct concepts in AI.Transparency refers to the degree to which the internal workings of an AI system, such as architecture, training data, parameters, decision rules and development processes, are open and accessible for inspection.A transparent model allows stakeholders to understand how the system is constructed and operates.Explainability, on the other hand, focuses on providing understandable reasons for specific decisions, predictions, or outputs generated by the model, regardless of whether its internal mechanisms are fully visible.Thus, a system can be explainable without being fully transparent, as post hoc techniques such as SHapley Additive exPlanations (SHAP), Organizational
Authors: Sanjay Goel, Gurvirender Tejay
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: diamond
OpenAlex ID: W7171567717