Intelligence as the Capacity to Overcome the Complexity of Information: Search for Unity in the Diverse Forms of Intelligence ^†

Abstract

The same term “intelligence” is used in different contexts as if there were a consensus on its meaning, even if already within the paradigmatic subject of human intelligence, the dominating view is that there are multiple (two, three, eight, or more) intelligences depending on the diverse criteria for being intelligent. This multiplicity increases when non-human agents are considered. This short paper serves as a preliminary report on a more comprehensive study, currently at the stage of preparation for publication, with the purpose to seek a uniform characteristic of intelligence formulated in terms of information and its complexity behind and above its diverse manifestations and instantiations in multiple contexts, including those of its human, animal, or other natural forms, as well as characteristics of artifacts. The proposed view of intelligence as the capacity to overcome the complexity of information unifies these manifestations, and not only sets foundations for their systematic study in already considered instances, but also allows effective seeking of its not-yet-known manifestations. Due to the limited volume of this paper, the focus is on the motivations for and feasibility of the more extensive study, with only an outline of the proposed view of a general concept of intelligence.

1. Introduction

This short article is intended as an introductory presentation of the subject of a more comprehensive metaphysical study of intelligence as one of the key characteristics present in reality, currently in progress, but to be published soon. The study of intelligence is difficult, not because it is an esoteric concept distant from everyday experience, familiar only to a narrow group of experts or initiates, but for exactly the opposite reason, because almost everyone believes to understand it very well, and many are convinced to possess it as a characteristic of their capabilities, usually in a high degree, but possibly exceeded by that of the so-called “rocket scientists” or “mad geniuses.” Moreover, equally common is the conviction about the ability to recognize instances of general intelligence (often considered synonymous with the expression of being smart) in other humans, some animals, and some artifacts, and to compare or assess its qualitative or even quantitative degree.

It does not require being a rocket scientist to recognize the complexity of the vague idea of intelligence and the urgent need for its precise conceptualization, or possibly multiple conceptualizations necessary to facilitate mutual understanding when the word “intelligence” appears in conversation or more specialized discourse. One of the main objectives of this paper is to contribute to the effort of setting philosophical foundations for the study of intelligence and to eliminate the sources of confusion omnipresent in the common use of this fashionable term.

This article starts with the discussion of the thesis that the original attempts to define, characterize, and measure human intelligence understood as a distinctive characteristic of a person, unique as a quality but varying in degree between individuals, frequently called a general intelligence, initiated at the turn of the 20th century, have been gradually abandoned and the discourse about intelligence (as an abstract noun) and its meaning, has been replaced by discussions about being intelligent (a common adjective with multiple context-dependent, diverse criteria and meanings). As a consequence of this context-dependent focus on the criteria for who is an intelligent human being, and later what object (natural or artificial) can be considered intelligent, the diversity of interpretations of intelligence has been accepted as unavoidable, and attempts to seek its uniform conceptualization were largely abandoned or dismissed as not feasible.

Surprisingly, in recent popular discussions, the expression “general intelligence” started to be used as a characteristic of a desired, but not precisely defined, design for the so-called artificial general intelligence (AGI) systems that should have been as intelligent as humans or even surpass all humans in this respect. To add to the confusion, the meaning of the expression “artificial intelligence” started to be used as a term for an independent singular entity capable of agency, possibly turned against humanity. This conceptual mess makes a disciplined philosophical study of intelligence not only necessary but also urgent.

The remaining part of the paper is a return to the original task of looking for the unified conceptualization of intelligence. The diversity of context-dependent qualifications for being an intelligent human, natural, or artificial entity is, in this paper, attributed to the high level of (structural and functional) complexity necessary for any object to qualify as intelligent, no matter how this intelligence is manifested. The complexity of an intelligent entity unavoidably leads to multiple and diverse manifestations.

In the case of life on Earth in its multiple levels of organization (e.g., cells, organs, organisms, populations, etc.,) the increase in its complexity is being explained as a result of the advantage of complex systems in adapting to, modifying, and/or controlling a complex environment expressed by the slightly modified Ashby’s Law of Requisite Variety [1] reformulated here as the Law of Requisite Complexity: Only complex system can model, control, and adapt to complex environment, i.e., optimally, the complexity of the system should match or exceed the complexity of the environment. This revision of Ashby’s Law is necessary for avoiding the overgeneralization by including well-known examples of the collectives of high numerical variety but low complexity. The Law was conceived and thrived in the context of cybernetics, in which the variety of the states of a system was identified with a measure of information. The idea of defining intelligence with the use of information and its complexity may be considered a revival of Ashby’s way of thinking, but with the present, much richer view of information and its complexity [2,3,4]. Complexity expressed in terms of information allows us to extend the concept of intelligence not only to its diverse manifestations in humans, the diverse levels of the hierarchical organization of life, but also to artifacts (manifested as artificial intelligence technological devices) or to unknown yet possible extraterrestrial forms of intelligence.

2. From Human General Intelligence to Multiple Human Intelligences

In his 2011 bestseller book “Thinking Fast and Slow” [5], Daniel Kahneman introduced one of his rules of common thinking: “A reliable way to make people believe in falshoods is frequent repetitions because familiarity is not easily distinguished from truth.” This rule applies to sentences that can be qualified as true or false. A similar rule can be applied to concepts linking words or terms with their meaning. There seems to be a sociolinguistic correspondence between the frequency of the use of terms or expressions, in particular those having philosophical significance, and the unrecognized diversity of their understanding. This diversity is obscured by the fact that the frequency of the use of words generates the impression of familiarity and familiarity, the illusion of identity and existence of the unique, “proper” denotation independent from any inquiry, the conviction that the meaning is obvious and uniform. As a result, people are using frequently invoked words with diverse subjective understanding while being convinced about their objective, uniform, and commonly shared meaning. If they are confronted with differences in the understanding presented by others, they claim that such a different understanding is erroneous.

Intelligence, either artificial, natural, or human, has become an illustrative instance of this regularity. The term is present everywhere, in particular, when qualified as artificial and used in its staple abbreviation AI, or when it appears unqualified, and is typically understood as human.

There are some curious differences between the ways human and artificial forms of intelligence are viewed. An unusual consensus has been developing among experts and laypeople for a long time that human intelligence is not only diverse but that its different forms are independent, with diverse functions in human individual or social life, each with a separate psycho-neurological mechanism. Thus, human intelligence can be fluid or crystallized following the division introduced in 1943 by Raymond Cattell [6], splitting the concept of Charles Spearman’s general intelligence present in psychology from the beginning of the 20th century into two different capacities. The former was understood as a purely general ability to solve unexpected problems without prior preparation or experience, and the latter consisted of the long-established discriminatory habits acquired through learning or training. Incidentally, this distinction can be considered a precedent to the popular, more recent distinction between fast (habitual, rigid) and slow (goal-oriented, flexible) thinking due to Kahneman [5].

The 1980s saw further divisions of intelligence. Robert Sternberg introduced his triarchic theory of intelligence, separating it into analytical, creative, and practical [7]. At about the same time, Howard Gardner introduced in his 1983 book “Frames of Mind: Theory of Multiple Intelligences” [8], the idea of “intelligences” in the plural form and presented their list originally consisting of seven of them: linguistic, logical–mathematical, spatial, musical, bodily–kinetic, interpersonal, intrapersonal. In 1995, he added the eighth: naturalistic intelligence. This triggered many attempts to distinguish specific types of such multiple intelligences, going beyond Gardner’s choices, leading to an ever-increasing variety.

The divisions of human intelligence were made using diverse criteria and different levels of argumentation, but they have always been motivated by the criticism of the inadequacy of the original Spearman’s concept of general intelligence and the attempts to measure it using variations of William Stern’s and Alfred Binet’s/Theodore Simon’s IQ tests. Even today, there is no agreement regarding the selection of criteria, names, and the degree of correlation, but the idea of multiple intelligences has become a standard for the study of human cognitive abilities.

Some practitioners still defend the use of IQ tests as a tool for education or career advising, social work, etc., but even they do not claim that such tests measure general intelligence or that the concept of general intelligence has any meaning. The defense is based on the popular common-sense conviction that multiple intelligences are somehow correlated and, therefore, tests for some not necessarily relevant but easy to observe and evaluate cognitive skills may be helpful in the estimation of the others. However, even the conviction of the correlation is questionable.

There is no agreement in the socio-psychological literature about the number and criteria for the choice of traits qualifying a person’s psychological constitution for successful functioning in the society, but psychologists oriented towards counseling and recruitment practice frequently select three main dimensions of personality as most predictive for “life outcomes”: intelligence (maximal capacity to achieve a novel goal successfully using perceptual-cognitive processes), conscientiousness (industriousness, achievement striving, orderliness, and self-discipline, and personal responsibility), and emotional stability [9]. In this practice-oriented research, the object of study is not intelligence, but personality as a general qualification of a subject for successful participation in social life, in which intelligence is understood in the traditional common-sense way as one of the “personality dimensions” associated with the ability to achieve a wide range of unspecified goals. Upon close inspection, all three dimensions belong to Gardner’s multiple intelligences, their derivatives introduced in the later works of his followers, or are closely associated with the manifestations or necessary conditions for being intelligent. The first of the dimensions, “intelligence (maximal capacity to achieve a novel goal successfully using perceptual-cognitive processes)” [9], is a typical description of what, in the approach of multiple intelligences, is called problem-solving intelligence. Similarly, “Conscientiousness (industriousness, achievement striving, orderliness, and self-discipline, and personal responsibility)” [9], is a description of the complex of behaviors conditioned by the control of focus and attention, frequently used as a crucial feature of intelligence. Finally, the emotional stability [9] is an abbreviation for the control of one’s emotions and reactions to the emotions of others.

The choice of the dimensions and the methods of testing are highly problematic (exceedingly fragmentary) from the point of view of the study of intelligence, but this does not make the results of empirical research less useful and important for the study of the diversity of the realizations of intelligence. The empirical research shows that the three (out of many) dimensions are not correlated. The correlation coefficients r between them are in the interval [−0.07, −0.03], which means they are not only negligible, but also negative. The achievement of being two standard deviations above the mean (not very impressive and hardly qualifying for a career of a rocket scientist) in all three is very rare, 0.0085%; Just one standard deviation above the mean in all three is 0.9366% [9]. This supports the view that the concept of a general human intelligence is a myth, and instead, we should consider multiple intelligences, with multiplicity counted not with single-digit numbers but interpreted as a wide distribution, at least as long as we look for the criteria of being intelligent.

3. The Faulty Quest for General Artificial Intelligence

The typical view of artificial intelligence (AI) relates it to human intelligence (such as a simulation, emulation, or recreation of the latter) and the main goal of the current technological research and innovation is the achievement of artificial general intelligence (AGI) which does not have commonly accepted or even discussed definition but is presented to general audience in the descriptions of the type “AGI […] a system that is capable of matching or exceeding human performance across the full range of cognitive tasks” [10]. This category error, blurring the distinction between the abstract concept and its individual realizations, characteristic of virtually all instances of the discourse on artificial intelligence (general or otherwise), perpetuates the hidden comprehension of its unity.

There are separate names for specific types of the recent technological realizations of AI (generative AI based on Large Language Models (LLMs) in the neural networks with deep learning architecture followed by Large Reasoning Models (LRMs) heavily dependent on external prompts, i.e., minimizing their autonomy; agentic AI re-engaging some forms of algorithmic computation, still dependent on initial prompts but with increasing autonomy of its operation through auto-reprompting; neuro-symbolic AI with a hybrid architecture; causal AI, etc.) However, all these qualifications reflect not the conceptual diversity of multiple artificial intelligences but technological differences in the search for the realization of the same goal of artificial general intelligence (AGI). If there is a consensus on no meaning for a unified general human intelligence, how can AGI achieve the goal of matching or exceeding it?

4. The Role of Philosophical Inquiry in the Study of Intelligence

The inconsistency between conceptualizations of the human multiple intelligences and the uniform idea of artificial general intelligence matching or even surpassing human cognitive abilities is only one of the many manifestations of conceptual chaos in the study of intelligence that calls for philosophical clarification.

The situation becomes even more complex when we consider the extensions of intelligence understood as a characteristic of natural but non-human entities present in diverse forms of life on Earth or the expected but not yet known forms of extraterrestrial life. The complication arises not only from the association of intelligence with life whose conceptualization has been similarly convoluted, but also from the increased diversity of morphological and behavioral forms involved in manifestations of intelligence in living objects that require a careful avoidance of the narrowing inquiry by anthropomorphization on one hand (the hidden assumptions that intelligence requires all characteristics of its human form), and on the other hand, overgeneralization in ascribing intelligence to anything that shares irrelevant features of human organism [11].

Interestingly, a similar confusion, to that of intelligence with being intelligent mentioned earlier, is present in the study of life and searching for its extraterrestrial forms. “The imprecise use of language is manifest. The ’elite’ have confused the concept of ‘being alive’ with the concept of ‘life’. This is not simply the mistaking an adjective for a noun. Rather, it represents the conflation of a part of a system with its whole. Parts of a living system might themselves be alive (a cell in our finger may be ‘‘alive,’’ as might a fertilized ovum in utero). But those living parts need not be coextensive with a living system and need not represent life. Using language precisely, one rabbit may be alive even though he or she is not life” [12].

There are some other conceptual issues related to intelligence that require philosophical inquiry when we search for its adequate definition. Some of them are intentionally excluded from this paper, such as the relationship between intelligence and consciousness, because of its limited size and the need for a very extensive presentation of the diverse attempts to conceptualize the latter. This exclusion can be defended by the fact that both characterize the same objects of inquiry (e.g., humans, animals, etc.), but in a very different way. For instance, someone can lose consciousness for a short time as a result of injury or when falling asleep, or we can consider the state of consciousness at some particular time, but intelligence, although it varies in the span of human life, is a long-term characteristic. An even more striking difference is the fact that the content of consciousness is accessible only to subjective experience, while intelligence is accessible to intersubjective observation and typically is not accessible to introspection. Finally, there are differences in the degree of intelligence between people, while in normal circumstances, a person can be conscious or unconscious, without any intermediate state, and the comparison of the consciousness of different people is impossible.

The study of the relationship between intelligence and consciousness requires a prior advance in their respective studies. The differences separate them, but this separation does not help us much in their conceptualization. On the other hand, their relationship may be helpful in future studies of both. For instance, in the study of intelligence, there is a deep philosophical question about the status of collective intelligence. While the term collective consciousness has only a metaphoric meaning of a high level of similarity in the content of consciousness in the members of the collective, the mechanisms of such a correlation are undeniably completely different from the mechanisms generating conscious experience in the individual members of the collective.

In contrast to this, collective intelligence is not only accepted as a possibility, but there are sometimes claims that intelligence is a purely collective phenomenon [11,13]. The arguments for the view of the collective character of intelligence are diverse, ranging from observations that the mechanisms of intelligence require coordinated work of the large collective of neurons organized into neural networks in the human or animal brain, to the claims that being an intelligent individual has meaning only in the context of characteristics of the role within a collective to which this individual belongs. This reasoning is supported in the case of human intelligence by the argument that its evolutionary development can only be explained by the role of ancestors in the ancestral populations.

Whether intelligence refers to individuals or collectives, this is irrelevant to the central thesis of this paper, as in both cases, the concept of information (e.g., in the latter as a means for communication and organization of collectives) and its complexity have primary roles.

A more fundamental role of philosophy in the study of intelligence is in setting its foundations. Thus, there is a need for the already mentioned distinction between intelligence and being intelligent. Intelligence is an abstract concept describing the capacity to be intelligent, applied to some object or system of objects (a collection of interacting objects). To be intelligent means to exhibit or possess this capacity to at least some degree.

Not every object or system can be characterized as being intelligent. The prerequisites for being intelligent are: (1) agency, understood as the ability to interact with a defined class of entities (hereafter called an environment) and to control the state of own components involved in such interactions (here functioning) without external control or initiation; (2) autonomy, understood as the ability to control the interactions with environment and own functioning involved in these interactions by detecting available choices of interactions and functioning, and making choices between them (i.e., acting).

5. Intelligence in Terms of Information and Complexity

Is the conceptual complexity of intelligence or intelligences a good reason to question the feasibility of, or justification for, any attempts to seek a unifying perspective on such a complex variety of related yet diverse forms of what in multiple contexts is called by the same common-sense name “intelligence”? This paper has as its objective to justify the negative answer to this question. The multiplicity of ways in which someone or something can be intelligent does not exclude the possibility of a uniform concept of information with multiple realizations. In this case, as in many other cases known from the intellectual history of humanity, the phenomenal complexity of the manifestations of intelligence is unquestionable, but the complexity of their perception and comprehension can be overcome with the use of appropriate intellectual tools.

The philosophical prerequisites for intelligence and being intelligent described in the preceding section put emphasis on control, and control requires involvement of information. The concept of control involves more than interaction, understood in its physical meaning, as it can also be exercised through the transmission of information or other communication means. Thus, the involvement of a concept of information as a tool in the study of intelligence is hardly surprising or questionable.

The tools proposed here are appropriately general concepts of information and its complexity [2,3,4], together with already known methods of reducing or controlling the complexity of information with their long history going back at least to the Law of Requisite Variety by W. Ross Ashby [1]. Ashby’s focus on variety was influenced by the early studies of complexity [14], which, before the advent of computer technology and computational methods of recursion and distributed processing implemented in it, was frequently associated with an intimidating variety of states in large collectives escaping human control and comprehension.

Thus, Ashby’s Law stated that any collective system characterized by a large variety of components can be modeled and/or controlled only by a system that is characterized by at least an equal variety. Its abbreviated popular form, “only variety can kill variety,” sounds slightly cryptic today, but Ashby’s presentation of the idea makes it clear that the variety means variety of the states of a system, and such a variety of states seemed to directly correspond to a large collective of its components. Later, this correspondence became questionable as there are many systems with only a few components that can have a large variety of states escaping all attempts of reduction (e.g., in classical mechanics a three-body system interacting through gravitational forces, or in quantum mechanics an elementary particle with uniquely determined quantum state, but whose measurement may produce infinitely many possible outcomes), and there are systems with a very large number of components that can be easily controlled by only a few parameters (e.g., a homogeneous gas in equilibrium with the great number of molecules has exactly one thermodynamic state when the three parameters are fixed).

Ashby’s Law was of great importance for the very popular at that time cybernetics, the study of the control of systems, mainly implemented through the use of technological devices replacing human engagement in controlling mechanical or electric work. The development of computer technology, in particular its digital form, reduced the interest in cybernetics, and Ashby’s Law has become mainly a subject of historical interest. Digital technological solutions eliminated the necessity for designing self-controlling systems of the type of a homeostat, and the focus shifted to the external control exercised by the computing devices with algorithmically programmed processes to achieve desired outcomes.

There was no more interest in fixing Ashby’s Law by replacing or modifying the term “variety”. The replacement of “variety” by “variety of states” would not help much, as there is no general, context-independent concept of a “state” even in physics. However, it can easily regain its importance if, instead, it is formulated as the Law of Requisite Complexity, stating that a system can model and/or control another system only if the complexity of the controller exceeds the complexity of the controlled.

The entire intellectual and technological progress of humanity can be interpreted in terms of searching for methods to reduce the complexity of human actions, and this progress has been achieved by the reduction in the complexity of information [4]. The methods of reducing the complexity of information are naturally associated with intelligent or efficient inquiry. Here we can find the association between human intelligence as a capacity to reduce the complexity of information. However, this can be easily extended to non-human forms of intelligence when the concept of information is sufficiently general.

At this point, we should observe that the use of the expression “to reduce the complexity of information” implies that intelligence is subject to the assessment of its degree. The more someone can reduce the complexity of information, or the higher level of complexity they can reduce, the more intelligent they are. On the other hand, different people may be able to reduce the complexity of different types of information. This allows for the relativization of the uniform concept of intelligence to its multiple manifestations, i.e., multiple intelligences, and to diverse contexts. Thus, we do not have to worry about the earlier discussed issues involved in the conceptualization of intelligence.

There could be a legitimate concern about the use of the concept of information. It seems natural in the context of intelligence, but someone could think about information as a resource and about intelligence as the ability to minimize the resources necessary to effectively perform a maximal range of actions. However, instead of generalization, this reconceptualization makes intelligence much more restricted. The concepts of “resource”, “effectiveness”, “range of actions”, and “to perform” seem obvious in everyday conversations, but they require very specific contexts and restrictive definitions, leading to significant loss of generality. Additionally, such definitions may presuppose an already defined concept of information. For instance, the concept of a resource is difficult to define without it, as is the identification or distinction of resources without information.