Progress within artificial intelligence has accelerated so rapidly during the past decade that terminology has struggled to keep pace with technological development. Terms such as Artificial Intelligence, Machine Intelligence, General Intelligence, Artificial General Intelligence and Machine General Intelligence are frequently used interchangeably within public discourse despite representing distinct conceptual frameworks. This lack of precision creates confusion not only within popular media but also within technical, regulatory and commercial discussions.
A rigorous understanding of Machine General Intelligence requires these concepts to be clearly differentiated. Machine intelligence describes the capability of computational systems to perform intelligent behaviour. General intelligence describes the underlying cognitive property of adaptive reasoning across domains. Machine General Intelligence represents the convergence of these concepts within engineered computational systems capable of broad, transferable and autonomous cognition.
This chapter establishes these distinctions before examining why generality, rather than raw computational performance, represents the defining characteristic of truly intelligent systems.
Intelligence as Adaptive Information Processing
Before discussing artificial or machine intelligence, it is necessary to consider intelligence itself.
Historically, intelligence has resisted simple definition because it encompasses numerous interacting cognitive processes rather than a single measurable ability. Psychologists have proposed theories centred upon general cognitive ability, multiple intelligences, working memory, executive control and adaptive behaviour. Neuroscientists investigate intelligence through biological mechanisms, while philosophers examine its epistemological and phenomenological dimensions.
Across these disciplines, however, several common characteristics consistently emerge.
An intelligent system is capable of:
- acquiring knowledge from experience;
- representing information internally;
- reasoning under uncertainty;
- adapting behaviour in response to changing environments;
- transferring knowledge between contexts;
- solving previously unseen problems; and
- improving future performance through learning.
Viewed in this manner, intelligence is fundamentally an adaptive information-processing process rather than a property unique to biological organisms.
This perspective represents one of the most significant conceptual developments in contemporary computational science. Intelligence need not depend upon neurons, DNA or biological evolution. Rather, it emerges whenever sufficiently sophisticated systems continuously acquire information, construct models of their environment, evaluate predictions and update those models through interaction with reality.
Human brains implement these processes through biological neural networks. Digital computers implement analogous processes through mathematical computation executed upon semiconductor hardware. The physical substrates differ dramatically, yet both systems transform information into increasingly effective models of the world.
Understanding intelligence as a computational process provides the theoretical foundation upon which Machine General Intelligence research is built.
Machine Intelligence
Machine Intelligence (MI) refers broadly to computational systems capable of exhibiting behaviours that would traditionally require human cognitive ability.
Unlike conventional software, which follows explicitly programmed instructions, machine intelligence acquires behaviour through learning, adaptation and inference. Rather than specifying every possible rule manually, developers construct algorithms capable of discovering statistical regularities within data, enabling systems to improve performance as additional experience becomes available.
Machine intelligence therefore encompasses numerous technologies, including:
- machine learning;
- deep learning;
- reinforcement learning;
- probabilistic reasoning;
- computer vision;
- natural language processing;
- autonomous robotics;
- planning systems; and
- knowledge representation.
These technologies differ considerably in implementation yet share the common objective of enabling machines to perform cognitive rather than purely mechanical tasks.
Importantly, machine intelligence should not be viewed as synonymous with human-like intelligence. A modern chess engine demonstrates extraordinary strategic capability yet possesses no understanding of biology, economics or language. Similarly, image-recognition systems may outperform humans at identifying specific medical conditions while remaining incapable of explaining their reasoning or adapting beyond narrowly defined domains.
Machine intelligence therefore describes capability without necessarily implying breadth.
A machine may be extraordinarily intelligent within one context while remaining incapable of functioning outside it.
Narrow Artificial Intelligence
Nearly every successful AI application deployed today belongs within the category commonly known as Narrow Artificial Intelligence (ANI).
Narrow AI systems are optimised for particular objectives under relatively constrained operating conditions.
Examples include:
- medical image classification;
- speech recognition;
- autonomous vehicle perception;
- fraud detection;
- recommendation systems;
- industrial process optimisation;
- machine translation;
- large language models specialised for text generation.
These systems frequently achieve or exceed human performance within their intended domains.
However, their competence does not generalise naturally.
Knowledge acquired while mastering protein folding does not enable the same system to understand constitutional law.
An autonomous driving system cannot immediately begin designing aircraft.
A language model trained on billions of documents still lacks grounded understanding of physical reality unless explicitly integrated with perception and interaction.
Consequently, narrow intelligence should be understood as specialised expertise rather than general cognition.
This distinction mirrors human expertise.
An accomplished neurosurgeon does not automatically become an accomplished architect.
Specialisation remains powerful but inherently limited.
General Intelligence
General Intelligence differs fundamentally from narrow intelligence.
Rather than measuring performance within isolated domains, general intelligence describes the capacity to acquire new competencies, integrate knowledge from multiple disciplines and adapt successfully to unfamiliar environments.
Several characteristics distinguish genuinely general intelligence.
Breadth
Competence extends across numerous domains without requiring complete redesign.
Learning mathematics assists scientific reasoning.
Language supports planning.
Social understanding informs negotiation.
Knowledge forms an interconnected cognitive network rather than isolated modules.
Transferability
Perhaps the defining feature of general intelligence is transfer learning.
Knowledge acquired in one context improves performance within another.
Humans continually demonstrate this ability.
Children learning musical rhythm often improve language acquisition.
Engineers apply mathematical reasoning across multiple disciplines.
Scientists adapt experimental methods between unrelated fields.
General intelligence therefore depends upon recognising structural similarities rather than memorising isolated solutions.
Adaptability
Novel situations represent the true test of intelligence.
General intelligence enables reasoning under unfamiliar circumstances where no explicit prior training exists.
Rather than retrieving stored answers, intelligent systems construct new explanations from existing knowledge.
Continuous Learning
Human cognition never truly stops learning.
Every experience modifies existing internal models.
General intelligence therefore requires lifelong learning rather than isolated training followed by fixed deployment.
Knowledge evolves continuously without catastrophic forgetting.
These properties distinguish general cognition from specialised optimisation.
They also define the principal objective of Machine General Intelligence research.
Machine General Intelligence
Machine General Intelligence may therefore be defined as:
A computational system capable of autonomous, adaptive and transferable reasoning across diverse domains, acquiring new knowledge through continual learning while demonstrating cognitive flexibility comparable to that of an ordinarily educated human adult.
Several elements of this definition deserve careful examination.
Autonomous
The system determines intermediate objectives independently.
Rather than requiring detailed human supervision, it plans, evaluates and revises strategies according to changing circumstances.
Adaptive
Performance improves continuously through experience.
Unexpected situations become learning opportunities rather than failure conditions.
Transferable
Knowledge generalises across problems.
Experience solving one class of challenge informs reasoning within structurally related domains.
Context-Sensitive
Behaviour depends upon environmental context rather than rigid programmed rules.
Identical information may produce different decisions depending upon social, temporal or physical circumstances.
Broadly Competent
Competence extends beyond individual tasks.
The system combines language, perception, reasoning, planning, abstraction and memory within an integrated architecture.
Collectively, these characteristics differentiate Machine General Intelligence from contemporary artificial intelligence systems.
Machine Intelligence versus Machine General Intelligence
The distinction between MI and Machine General Intelligence is analogous to the difference between expertise and understanding.
Machine Intelligence answers questions efficiently within predefined domains.
Machine General Intelligence understands problems sufficiently to formulate new questions.
Machine Intelligence recognises patterns.
Machine General Intelligence constructs explanations.
Machine Intelligence optimises.
Machine General Intelligence reasons.
Machine Intelligence executes.
Machine General Intelligence learns continuously.
This distinction becomes increasingly important as modern foundation models demonstrate impressive emergent capabilities.
Large language models often appear generally intelligent because they perform numerous tasks within a single architecture.
Yet broad capability alone does not necessarily constitute genuine general intelligence.
Current systems remain constrained by limitations including:
- incomplete causal reasoning;
- limited grounded understanding;
- restricted long-term planning;
- imperfect continual learning;
- vulnerability to hallucination;
- inconsistent abstraction;
- limited self-directed learning.
Consequently, today's most advanced AI systems represent important milestones towards Machine General Intelligence rather than its completion.
Machine General Intelligence versus Machine Superintelligence
Machine General Intelligence should also be distinguished from Machine Superintelligence.
Although frequently conflated, these concepts describe different stages of capability.
Machine General Intelligence seeks parity with broad human cognitive flexibility.
Machine Superintelligence describes systems exceeding humanity across virtually every intellectual dimension simultaneously.
Such superiority may include:
- scientific reasoning;
- mathematical discovery;
- engineering;
- creativity;
- strategic planning;
- memory;
- communication;
- social modelling;
- technological innovation.
Whereas Machine General Intelligence represents an achievable research objective grounded in contemporary cognitive science, superintelligence remains speculative.
Importantly, the transition from Machine General Intelligence to Machine Superintelligence may not be gradual.
Once systems become capable of autonomously improving their own learning algorithms, architectures and scientific understanding, recursive improvement could theoretically accelerate capability beyond human levels.
Whether such scenarios prove realistic remains actively debated.
For this reason, the present white paper focuses primarily upon Machine General Intelligence rather than speculative superintelligence.
Intelligence as Collaboration Rather Than Competition
Perhaps the greatest misconception surrounding Machine General Intelligence is the assumption that intelligent machines and intelligent humans necessarily compete.
Historical evidence suggests precisely the opposite.
Technology repeatedly increases human capability by extending cognitive reach.
Writing extended memory.
Libraries extended collective knowledge.
Computers extended calculation.
Networks extended communication.
Machine intelligence extends reasoning itself.
Viewed from this perspective, Machine General Intelligence represents another cognitive technology within humanity's continuing intellectual evolution.
Its greatest contribution may therefore lie not in replacing human intelligence but in expanding the space of problems humanity is capable of solving.
Scientists may investigate theories previously beyond computational reach.
Physicians may integrate millions of patient histories into personalised treatments.
Governments may simulate policy consequences before implementation.
Educators may provide genuinely individualised instruction to every learner.
Artists may explore creative spaces previously unimaginable.
In each case, intelligence remains fundamentally human in purpose while becoming increasingly augmented in capability.
Machine General Intelligence therefore represents not the endpoint of human cognition but the beginning of a new form of collaborative intelligence in which biological and computational reasoning complement one another.