Overview
ISO/IEC TR 24372:2021 - Information technology - Artificial intelligence (AI) - Overview of computational approaches for AI systems - is a technical report that summarizes the state of the art in computational approaches used by AI systems. Rather than prescribing requirements, it describes main computational characteristics of AI systems and surveys algorithms and approaches (both machine-learning and non-machine-learning), referencing use cases in ISO/IEC TR 24030. The report is intended as a practical reference for understanding technical options when designing, assessing or regulating AI solutions.
Key topics covered
- Main characteristics of AI systems
Typical attributes such as adaptable, constructive, coordinated, dynamic, explainable, discriminative vs. generative, introspective, trained/trainable, and support for various data types.
- Computational characteristics
Taxonomy by computation: data-based vs. knowledge-based, infrastructure-based, algorithm-dependent, and multi-step vs. end-to-end learning.
- Types of computational approaches
High-level classification into knowledge-driven and data-driven approaches, plus hybrid designs.
- Selected algorithms and techniques
Concise overviews of practical methods used in AI systems:
- Knowledge engineering: ontologies, knowledge graphs, semantic web.
- Logic & reasoning: inductive, deductive, hypothetical, Bayesian inference.
- Machine learning: decision trees, random forests, linear/logistic regression, k-NN, Naïve Bayes, and neural network families - feedforward, RNN, LSTM, CNN, GAN - plus transfer learning, BERT-style transformers (bidirectional encoder representations), XLNet.
- Metaheuristics: genetic algorithms, heuristics and search techniques.
Practical applications and who uses it
- AI system architects and engineers - to choose suitable computational approaches (knowledge-driven vs. data-driven) and algorithms for specific tasks.
- Procurement, risk & compliance teams - to understand capabilities and limitations when evaluating AI vendors and solutions.
- Researchers and educators - as a curated, standards-based survey of contemporary AI methods.
- Standards developers and regulators - to align policy and assurance frameworks with common computational patterns and use cases.
- Auditors and QA teams - to support review criteria around explainability, training regimes and algorithm selection.
Related standards
- ISO/IEC 22989 - AI concepts and terminology (normative reference).
- ISO/IEC 23053 - Framework for AI systems using machine learning (normative reference).
- ISO/IEC TR 24030 - Use cases referenced by TR 24372.
ISO/IEC TR 24372 is a practical, standards-aligned primer for anyone needing a compact, authoritative overview of AI computational approaches, algorithms, and system characteristics for design, assessment and governance of AI systems.