Intelligence is among the most widely discussed topics in philosophy, psychology, biology, and, especially today, in artificial intelligence. Yet despite the ubiquity of the term, intelligence remains poorly and inconsistently defined. Some definitions emphasize problem-solving, others focus on learning, and still others stress adaptation or prediction. What is needed is a unifying definition that spans humans, other animals, and artificial systems alike.
This paper proposes such a definition:
Intelligence is the ability to construct and refine models of reality from observation, teaching, or inherited adaptation, and to use these models to reconstruct past events or predict possible futures for the purpose of guiding decisions or actions.
This definition is both concise and comprehensive. It emphasizes the processes of model-building and application, and it situates intelligence within an iterative cycle that bridges perception, memory, prediction, and action. The following sections expand on each element of this definition.
Constructing and Refining Models of Reality
At its core, intelligence involves the creation of internal models that represent aspects of external reality. These models are abstractions rather than literal copies: they simplify complexity into usable patterns. A physicist’s equations describing quantum mechanics, a child’s expectation that unsupported objects will fall, and a cat crouching before a pounce - anticipating where prey will move next-are all examples of such models.
Refinement is essential. Models must be updated when expectations fail, outcomes deviate, or new data become available. This continual feedback process-constructing, testing, correcting-constitutes the engine of adaptive thought.
Sources of Models: Observation, Teaching, and Inherited Adaptation
1. Learning from Observation
The most fundamental way of building models is through direct engagement with the world. Organisms detect regularities in their environment-gravity causes objects to fall, predators tend to stalk from cover, fire burns-and abstract these patterns into models that guide future action.
2. Learning from Teaching
Another path is teaching, in which models are transferred from one agent to another.
Explaining: The teacher conveys a model, using words, demonstrations, or symbolic representations.
Understanding: The student acknowledges that they have, or believe they have, acquired a working model.
This process enables knowledge to accumulate culturally rather than being rediscovered individually. Language, education systems, and scientific traditions exemplify the exponential power of teaching.
3. Inherited Adaptation
Not all models must be constructed within a lifetime. Evolution encodes species-level strategies into instincts and reflexes, which function as inherited models of reality. The spider’s web-building behavior, the migratory routes of birds, and mammalian fight-or-flight responses are examples. These prepackaged models provide a foundation upon which observational and taught learning can build.
Training and Inference
The distinction between training and inference - first made explicit in artificial intelligence research - offers a useful framework for understanding how models are built and applied across domains.
Training is the process of constructing or refining a model. In AI systems, training occurs when large datasets are used to adjust the parameters of a network, often at great computational cost. In humans and animals, training corresponds to learning through observation and teaching. At the evolutionary scale, instincts represent the outcome of training performed over aeons, with natural selection as the trainer.
Inference is the process of applying an already-trained model to new inputs. In AI, this is the rapid phase where a system recognizes an image or generates text. In biology, inference occurs when a bird uses its inherited migratory map or when a human applies learned knowledge to solve a problem.
What makes this distinction powerful is its universality. Current AI systems show training and inference as clearly separable processes, but the same logic applies across life. Instincts are “pre-trained” through evolution and then applied almost exclusively in inference. Human cognition blends both: slow, effortful training through experience and teaching, followed by rapid inference when making decisions. Intelligence, in this sense, always involves both phases - though in differing proportions depending on the system.
Applying Models: Reconstructing the Past and Predicting the Future
Once models are trained, they can be applied through inference. Two complementary applications are central:
Reconstructing the Past: Causal reasoning allows agents to infer how events unfolded, even without direct perception. Historians, detectives, and ordinary individuals reconstructing the causes of a present situation all engage in this backward simulation.
Predicting the Future: The more celebrated aspect of intelligence is foresight. A predator anticipating the movements of prey, a scientist modeling climate change, or a child predicting the trajectory of a rolling ball all demonstrate forward simulation. A crouching cat, for example, does not simply react to the mouse’s current position - it models where the mouse will be a moment later, and times its pounce accordingly.
Both reconstruction and prediction transform models into tools for understanding temporal dynamics and guiding action.
Guiding Decisions and Actions
The ultimate purpose of intelligence is pragmatic: to guide decisions and actions that improve outcomes relative to an agent’s goals. A bird choosing whether to forage or flee, a pilot steering through turbulence, or a policymaker weighing the risks of intervention all illustrate how intelligence translates models into behavior. It is at this point - where thought meets the real world - that the adequacy of intelligence is revealed.
Applicability Across Domains
One of the most persistent challenges in defining intelligence is the apparent diversity of its forms. Psychologists and educators often describe “multiple intelligences” as if they were fundamentally distinct: analytical, emotional, social, creative, bodily-kinesthetic, and so on. This view highlights the richness of human and animal capacities, but risks fragmenting intelligence into unrelated categories.
The model-based definition resolves this by showing that all forms of intelligence share the same underlying cycle. The differences lie not in the essence of intelligence, but in the domain and content of the models being trained and applied.
Analytical intelligence: Logical reasoning, mathematics, and technical problem-solving exemplify the construction of abstract, formal models. Equations, algorithms, and logical proofs are symbolic representations of reality, enabling precise predictions and reconstructions.
Emotional intelligence: Emotions may seem less “rational,” but emotional intelligence operates on the same cycle. It requires building models of one’s own emotional states and those of others, predicting likely responses, and guiding interpersonal action. Anticipating that a colleague might feel defensive after criticism is an act of emotional inference no less than solving an equation is an act of analytical inference.
Social intelligence: At the group level, intelligence involves modeling not just individuals but the interactions among them. Social intelligence includes understanding norms, roles, and group dynamics, as well as predicting outcomes of collective action. A skilled negotiator, a community leader, or even a child navigating playground politics demonstrates intelligence by simulating social futures and acting accordingly.
Artificial intelligence: In computational systems, the principle is the same. An AI agent constructs models - through data, programming, or learning algorithms - and refines them to make predictions or decisions. Whether in natural language processing, autonomous driving, or game-playing, AI follows the same recursive cycle of training, inference, and action.
Instinctual intelligence: At the evolutionary level, intelligence is encoded genetically. Instincts are preloaded models of reality, shaped by natural selection. Migration routes, courtship displays, and predator-avoidance behaviors exemplify inherited models that allow organisms to act effectively without individual learning. These models are less flexible than learned ones, but they still fit the same definition of intelligence.
What emerges from this unified perspective is a continuity rather than a fragmentation of intelligence. Analytical, emotional, social, artificial, and instinctual intelligences are not separate faculties, but diverse expressions of the same fundamental process. They differ in content, domain, and medium, yet all instantiate the cycle: Construct or inherit → Train (through experience, teaching, or evolution) → Infer past/future → Act.
Thus, the apparent “many intelligences” of popular theory are best understood not as separate kinds of intelligence, but as distinct applications of a single, general capacity. This reframing preserves the richness of human and non-human abilities while grounding them in one coherent definition.
Consciousness in Relation to Intelligence
The role of consciousness is more elusive. While intelligence can be defined functionally, consciousness concerns the experience of those functions. Two major perspectives are possible:
Consciousness as the executor of intelligence: the “something” that runs the model and initiates action. On this view, every intelligent agent that fits the definition possesses some level of consciousness, from insects to artificial agents.
Consciousness as self-modeling: the capacity of an agent to include itself within its own model of reality, predicting not only the external world but its own behavior within it. This more restrictive view associates consciousness with systems capable of self-reflection: humans, some animals, and possibly advanced AI.
Regardless of interpretation, consciousness is distinct from intelligence. Intelligence provides the capacity to build and apply models; consciousness determines how those models are experienced and integrated. A spider’s intelligence may be largely instinctual with minimal consciousness, while the human brain, capable of abstract reasoning and self-reflection, leverages consciousness to far greater effect.
Thus, intelligence can be defined clearly, while consciousness remains an open and variable phenomenon, possibly universal in minimal form, yet vastly different in depth and influence across systems.
Conclusion: Intelligence as a Dialogue with Reality
This model-based definition highlights intelligence as an iterative process rather than a static trait. It is not merely a storehouse of facts but a cycle of perceiving, modeling, training, inferring, refining, and acting. This cycle operates in evolution, in individual cognition, and in artificial systems, providing a unifying framework that transcends disciplinary boundaries.
Intelligence, then, is the art of navigating uncertainty with models that are provisional yet powerful. Consciousness may or may not be involved, but intelligence itself remains the functional bridge between reality and effective action.
Cosma Pabouctsidis – August 20, 2025
cosmapa@gmail.com
Intelligence is among the most widely discussed topics in philosophy, psychology, biology, and, especially today, in artificial intelligence. Yet despite the ubiquity of the term, intelligence remains poorly and inconsistently defined. Some definitions emphasize problem-solving, others focus on learning, and still others stress adaptation or prediction. What is needed is a unifying definition that spans humans, other animals, and artificial systems alike.
This paper proposes such a definition:
Intelligence is the ability to construct and refine models of reality from observation, teaching, or inherited adaptation, and to use these models to reconstruct past events or predict possible futures for the purpose of guiding decisions or actions.
This definition is both concise and comprehensive. It emphasizes the processes of model-building and application, and it situates intelligence within an iterative cycle that bridges perception, memory, prediction, and action. The following sections expand on each element of this definition.
Constructing and Refining Models of Reality
At its core, intelligence involves the creation of internal models that represent aspects of external reality. These models are abstractions rather than literal copies: they simplify complexity into usable patterns. A physicist’s equations describing quantum mechanics, a child’s expectation that unsupported objects will fall, and a cat crouching before a pounce - anticipating where prey will move next-are all examples of such models.
Refinement is essential. Models must be updated when expectations fail, outcomes deviate, or new data become available. This continual feedback process-constructing, testing, correcting-constitutes the engine of adaptive thought.
Sources of Models: Observation, Teaching, and Inherited Adaptation
1. Learning from Observation
The most fundamental way of building models is through direct engagement with the world. Organisms detect regularities in their environment-gravity causes objects to fall, predators tend to stalk from cover, fire burns-and abstract these patterns into models that guide future action.
2. Learning from Teaching
Another path is teaching, in which models are transferred from one agent to another.
This process enables knowledge to accumulate culturally rather than being rediscovered individually. Language, education systems, and scientific traditions exemplify the exponential power of teaching.
3. Inherited Adaptation
Not all models must be constructed within a lifetime. Evolution encodes species-level strategies into instincts and reflexes, which function as inherited models of reality. The spider’s web-building behavior, the migratory routes of birds, and mammalian fight-or-flight responses are examples. These prepackaged models provide a foundation upon which observational and taught learning can build.
Training and Inference
The distinction between training and inference - first made explicit in artificial intelligence research - offers a useful framework for understanding how models are built and applied across domains.
What makes this distinction powerful is its universality. Current AI systems show training and inference as clearly separable processes, but the same logic applies across life. Instincts are “pre-trained” through evolution and then applied almost exclusively in inference. Human cognition blends both: slow, effortful training through experience and teaching, followed by rapid inference when making decisions. Intelligence, in this sense, always involves both phases - though in differing proportions depending on the system.
Applying Models: Reconstructing the Past and Predicting the Future
Once models are trained, they can be applied through inference. Two complementary applications are central:
Both reconstruction and prediction transform models into tools for understanding temporal dynamics and guiding action.
Guiding Decisions and Actions
The ultimate purpose of intelligence is pragmatic: to guide decisions and actions that improve outcomes relative to an agent’s goals. A bird choosing whether to forage or flee, a pilot steering through turbulence, or a policymaker weighing the risks of intervention all illustrate how intelligence translates models into behavior. It is at this point - where thought meets the real world - that the adequacy of intelligence is revealed.
Applicability Across Domains
One of the most persistent challenges in defining intelligence is the apparent diversity of its forms. Psychologists and educators often describe “multiple intelligences” as if they were fundamentally distinct: analytical, emotional, social, creative, bodily-kinesthetic, and so on. This view highlights the richness of human and animal capacities, but risks fragmenting intelligence into unrelated categories.
The model-based definition resolves this by showing that all forms of intelligence share the same underlying cycle. The differences lie not in the essence of intelligence, but in the domain and content of the models being trained and applied.
What emerges from this unified perspective is a continuity rather than a fragmentation of intelligence. Analytical, emotional, social, artificial, and instinctual intelligences are not separate faculties, but diverse expressions of the same fundamental process. They differ in content, domain, and medium, yet all instantiate the cycle: Construct or inherit → Train (through experience, teaching, or evolution) → Infer past/future → Act.
Thus, the apparent “many intelligences” of popular theory are best understood not as separate kinds of intelligence, but as distinct applications of a single, general capacity. This reframing preserves the richness of human and non-human abilities while grounding them in one coherent definition.
Consciousness in Relation to Intelligence
The role of consciousness is more elusive. While intelligence can be defined functionally, consciousness concerns the experience of those functions. Two major perspectives are possible:
Regardless of interpretation, consciousness is distinct from intelligence. Intelligence provides the capacity to build and apply models; consciousness determines how those models are experienced and integrated. A spider’s intelligence may be largely instinctual with minimal consciousness, while the human brain, capable of abstract reasoning and self-reflection, leverages consciousness to far greater effect.
Thus, intelligence can be defined clearly, while consciousness remains an open and variable phenomenon, possibly universal in minimal form, yet vastly different in depth and influence across systems.
Conclusion: Intelligence as a Dialogue with Reality
This model-based definition highlights intelligence as an iterative process rather than a static trait. It is not merely a storehouse of facts but a cycle of perceiving, modeling, training, inferring, refining, and acting. This cycle operates in evolution, in individual cognition, and in artificial systems, providing a unifying framework that transcends disciplinary boundaries.
Intelligence, then, is the art of navigating uncertainty with models that are provisional yet powerful. Consciousness may or may not be involved, but intelligence itself remains the functional bridge between reality and effective action.