The Australian Government's introduction to artificial intelligence, published by the National AI Centre on ai.gov.au, opens with this definition [1]:
"Artificial intelligence (AI) is technology that learns from data and recognises patterns. It uses those patterns to generate predictions, recommendations or content."
That is a fair description of one family of AI: machine learning, and the generative AI built on it. It is not a description of artificial intelligence. The international definitions that Australia's own policy work relies on are far broader, and so is the National AI Centre's own guidance. On a page whose purpose is to explain AI to Australian businesses, that gap matters.
What the page says
- AI "learns from data and recognises patterns".
- AI systems "rely on two core elements": "data to learn from", and "a model or algorithm that can analyse the data and produce a result or recommendation".
- "Unlike traditional software that follows fixed rules, AI learns patterns from data."
- The video on the page, which draws on the Human Technology Institute's research, says that when AI began in the 1950s, AI systems "became known as expert systems", and that "the last decade has seen the rise of machine learning".
- The same video says that "Machine learning is underpinned by statistics, linear algebra, probability theory, and calculus", that "AI systems tend not to be explicitly programmed", and that they "tend to be unpredictable and opaque".
- Its "common types of AI" are generative AI, machine learning, automation with AI features, agentic AI, computer vision and natural language processing.
What the international definitions say
The OECD. Its definition, revised in November 2023, is the one the National AI Centre itself uses in its Guidance for AI adoption, published in October 2025 on the same website [2]:
"A machine-based system that, for explicit or implicit objectives, infers, from the input it receives, how to generate outputs such as predictions, content, recommendations, or decisions that can influence physical or virtual environments. Different AI systems vary in their levels of autonomy and adaptiveness after deployment."
The European Union's AI Act uses the same core wording in its legal definition [3]. It then says what the inference can rest on: "machine learning approaches that learn from data how to achieve certain objectives, and logic- and knowledge-based approaches that infer from encoded knowledge or symbolic representation of the task to be solved" [4]. The European Commission lists those second approaches as "knowledge representation, inductive (logic) programming, knowledge bases, inference and deductive engines, (symbolic) reasoning, expert systems and search and optimisation methods" [5].
The international standard for AI terminology, ISO/IEC 22989:2022, defines an AI system as an "engineered system that generates outputs such as content, forecasts, recommendations or decisions for a given set of human-defined objectives" [6]. It is free to read.
None of these definitions requires learning from data, none requires a trained network, and none mentions linear algebra. What they share is inference: a system that works out, from what it receives, what to produce. Learning patterns from data is one way to do that. There are many others.
What the page gets wrong
- It defines one family of AI as all of it. "Learns from data and recognises patterns" describes machine learning. It leaves out every system that reasons from encoded knowledge, searches for solutions, plans, or reasons about cause and effect. The EU explicitly names systems that reason from encoded knowledge and that search for solutions, and the OECD definition the National AI Centre itself uses excludes none of them.
- Data to learn from is not a core element of all AI. A knowledge based system infers from what experts encoded. A planner infers from a description of the problem. The EU's own legal text names them as AI [4]. It excludes only systems that execute rules "defined solely by natural persons".
- The history is out by decades.
- The mathematics of one family is presented as the mathematics of AI. Linear algebra and calculus underpin neural networks. Other families rest on logic, on search over graphs, on probability over networks of beliefs, on evolution, and on information theory.
- "Unpredictable and opaque" is a property of some architectures, not of AI. A knowledge based system can show the rule behind every conclusion. A system built on causal states can be drawn: you can count its states and read what moves it from one to the next. In such systems, as we argued in an earlier post, bias becomes something you can point at [9].
- The "types of AI" mix kinds of things. Machine learning is a technique. Computer vision and natural language processing are applications. Generative AI chat tools are products, and automation features are features. The techniques that are not machine learning get no category at all.
AI is a family of approaches
The standard textbook of the field, Russell and Norvig's Artificial Intelligence: A Modern Approach, works through search, logic, planning and reasoning under uncertainty before it reaches machine learning [10]. Here are ten families, each with the work it is usually traced to:
| family | what it does | where it is usually traced to |
|---|---|---|
| Search and planning | Finds a sequence of actions that reaches a goal | Hart, Nilsson and Raphael, 1968 [11] |
| Logic and knowledge representation | Reasons from facts and rules that people encode | The EU's list of AI approaches [5] |
| Expert systems | Applies a specialist's knowledge to a specialist problem | DENDRAL, Stanford, 1960s [8] |
| Probabilistic reasoning | Reasons under uncertainty, over networks of beliefs | Pearl, 1988 [12] |
| Causal inference | Asks what happens if we intervene, and what would have happened if we had not | Pearl, 2009 [13] |
| Evolutionary computation | Improves solutions by variation and selection | Holland, 1975 [14] |
| Fuzzy logic | Reasons with degrees of truth rather than true or false | Zadeh, 1965 [15] |
| Reinforcement learning | Learns what to do from reward | Sutton and Barto [16] |
| Neural networks, including deep learning and large language models | Learns patterns from large amounts of data | Rosenblatt, 1958 [7] |
| Computational mechanics | Learns the causal states of a process from the stream of its observations | Crutchfield and Young, 1989 [17] |
The ai.gov.au page describes one row of that table.
Why it matters
A government explainer is not just a web page. It shapes how businesses think about what is possible, how agencies write procurement and risk rules, and what counts as "AI" when money is allocated.
If AI means a trained network that learns from large data and is opaque by nature, then every rule written for it will assume all three. Architectures that work from little data, run on ordinary computers and can be inspected end to end get no credit for any of it. They are measured against assumptions that do not apply to them. For a country that wants its own AI capability, narrowing the definition to the one architecture dominated by a few overseas companies is a strange place to start.
Where Senua AI sits
Senua AI is built on the last row of that table. Our engine learns the causal states of each stream it reads, from the stream itself, as we set out in our last post [18].
We do not use models. No LLMs. No SLMs. Of any kind.
No GPU. No data centre. No AI token costs, ever. Senua AI runs on ordinary, low end CPUs.
Senua AI does learn from data, so the page's first sentence would just about cover us. Little else on the page would. Our engine is not a network of trained weights, it can be inspected state by state, and it needs no data centre. Under the OECD, EU and ISO definitions, there is no question at all.
A suggested correction
The National AI Centre does not need a new definition. It already uses the right one in its own guidance. Our suggestion for the opening of the page, built on that definition:
"Artificial intelligence is a family of technologies that infer, from the input they receive, how to produce outputs such as predictions, recommendations, content or decisions. Some learn patterns from data, as machine learning and generative AI do. Others reason from encoded knowledge, search for solutions, plan, or learn the causal structure of a process. AI systems vary in how independently they act and how much they adapt after they are deployed."
Australian businesses deserve an introduction to AI that describes the whole field, not just the part currently in the news.
References
The page, and the definitions
- National AI Centre, Artificial intelligence explained, ai.gov.au, read 19 October 2026, including the transcript of its video.
- National Artificial Intelligence Centre, Guidance for AI adoption: implementation guidance, October 2025, definitions table, "AI system", citing the OECD.AI Policy Observatory.
- European Union, Regulation (EU) 2024/1689 (the AI Act), Article 3(1).
- European Union, the AI Act, Recital 12.
- European Commission, AI Act Service Desk, Frequently asked questions, read 19 October 2026.
- ISO/IEC 22989:2022, Information technology: Artificial intelligence: Artificial intelligence concepts and terminology, clause 3.1.4. Listed by Standards Australia; made publicly available by ISO and IEC (Holistic AI).
History and the families of AI
- F. Rosenblatt, The perceptron: a probabilistic model for information storage and organization in the brain, Psychological Review 65(6) (1958) 386 to 408. Peer reviewed.
- R. K. Lindsay, B. G. Buchanan, E. A. Feigenbaum and J. Lederberg, DENDRAL: a case study of the first expert system for scientific hypothesis formation, Artificial Intelligence 61(2) (1993) 209 to 261. Peer reviewed.
- Senuamedia Blog, Bias you can point at.
- S. Russell and P. Norvig, Artificial Intelligence: A Modern Approach, 4th edition, Pearson (2020).
- P. E. Hart, N. J. Nilsson and B. Raphael, A formal basis for the heuristic determination of minimum cost paths, IEEE Transactions on Systems Science and Cybernetics 4(2) (1968) 100 to 107. Peer reviewed.
- J. Pearl, Probabilistic Reasoning in Intelligent Systems: Networks of Plausible Inference, Morgan Kaufmann (1988).
- J. Pearl, Causality: Models, Reasoning, and Inference, 2nd edition, Cambridge University Press (2009).
- J. H. Holland, Adaptation in Natural and Artificial Systems, University of Michigan Press (1975).
- L. A. Zadeh, Fuzzy sets, Information and Control 8(3) (1965) 338 to 353. Peer reviewed.
- R. S. Sutton and A. G. Barto, Reinforcement Learning: An Introduction, 2nd edition, MIT Press (2018).
- J. P. Crutchfield and K. Young, Inferring statistical complexity, Physical Review Letters 63 (1989) 105 to 108. Peer reviewed.
- Senuamedia Blog, The Senua AI Inversion, 19 October 2026.