Human intelligence redefined

Authors: Dr. Aleksei Minin and Kirill Tsyganov

We also would like to acknowledge Prof. Alois Knoll (TUM) for his support and fruitful thoughts and comments during the paper preparation.

Abstract.

Recent LLMs success opens a new chapter in Artificial General Intelligence (AGI) discussions. Intelligence, as a hardware-agnostic entity, has been defined by multiple philosophical schools. The philosophical ideas were later confirmed by physiological studies. And today the narrowing gap between LLMs’ and humans’ ability to build logical chains forces us to think about intelligence once again.

“Knowing yourself is the beginning of all wisdom.”― Aristotle

We will consider intelligence from philosophical and physiological perspectives and define it as a knowledge retrieval ability. Then we will draw analogies of human intelligence with LLMs and propose potential benefits of utilising multi-dimensional sensorial vocabularies natural for living organisms in LLMs architecture. While going through the logic of different types of intelligence and knowledge we will derive the role of Language. Finally, we will formulate the differences between human intelligence and LLM one.

Hypothesis 0: Intelligence is the ability to generate knowledge, use it, make rational reasoning and act accordingly.

What is intelligence?

Human intelligence is a complex and multifaceted construct that has been the subject of extensive scientific inquiry. It encompasses various cognitive abilities, including reasoning, problem-solving, memory, language comprehension, and learning. Understanding the nature of human intelligence is crucial for advancing fields such as artificial intelligence (AI) and cognitive neuroscience.

Philosophical outlook

The concept of intelligence has a rich history in ancient Greek philosophy. Greek philosophers made significant contributions to the understanding of intellect and its development over time.

One of the earliest philosophical perspectives on intellect can be traced back to the Pre-Socratic philosopher Anaxagoras (5th century BCE). Anaxagoras proposed that the mind, or nous, is the ultimate source of order in the universe. According to him, the mind is responsible for organising and governing all things.

Plato, a renowned philosopher from ancient Greece, further developed the notion of intellect in his dialogues. In Plato’s philosophy, intelligence is not derived from sensory experience but rather from rational contemplation and understanding of abstract forms or ideas. Plato argued that individuals possess innate knowledge that can be recollected through philosophical inquiry.

Aristotle, a student of Plato, expanded upon these ideas and developed his own theory of intellect. Aristotle distinguished between two types of intellect: passive intelligence and active intelligence. According to Aristotle, passive intelligence has the capacity to receive information from sensory experiences, while active intelligence is responsible for abstract reasoning and understanding universal concepts.

Later on, Stoic philosophers such as Zeno of Citium and Epictetus emphasised the role of reason and rationality in human intelligence. They believed that individuals should cultivate their rational faculties to achieve “wisdom and live virtuous lives”.

The ideas put forth by these Greek philosophers laid the foundation for subsequent discussions on intellect in Western philosophy. Their notions influenced later thinkers such as Descartes, Kant, and modern cognitive scientists who continue to explore the nature and development of human intelligence.

As to the more recent advancements in understanding intelligence, Howard Gardner’s theory of multiple intelligences (Gardner, 1983) posits that intelligence is not a single entity but rather comprises distinct modalities or “intelligences”, including linguistic, logical-mathematical, spatial, musical, bodily-kinesthetic, interpersonal, intrapersonal, and naturalistic intelligences. This theory emphasises the diversity of human cognitive abilities and highlights the importance of recognizing and nurturing individual strengths.

Physiological outlook

A cognitive neuroscience research has shed light on the neural basis of human intelligence. Studies employing neuroimaging techniques such as functional magnetic resonance imaging (fMRI) have identified brain regions associated with specific cognitive functions. For instance, the prefrontal cortex has been implicated in executive functions like working memory and decision-making (Miller & Cohen, 2001), while the temporal lobes are involved in language processing (Hickok & Poeppel, 2007).

Furthermore, genetic studies have provided evidence for the heritability of intelligence. Twin studies (Plomin & Deary, 2015) have shown that genetic factors contribute significantly to individual differences in intelligence. However, environmental factors also play an important role in intellectual development.

Defining intelligence

Based on the previous studies, we define intelligence as a mechanism that allows us to generate knowledge (build it over time via training) and use it rationally (be able to access knowledge acquired at different points of time and make reasoning based on the current situation and experience from the past).

Defining knowledge

Before we proceed with intelligence, we need to identify the term Knowledge.

Knowledge is a complex concept that has been explored from various perspectives in philosophy and cognitive science. Different philosophical schools and thinkers have offered diverse viewpoints on the nature, acquisition, and justification of knowledge. Here are three prominent perspectives on knowledge:

• Empiricists argue that knowledge is derived primarily from sensory experience and observation of the external world. According to this perspective, knowledge is acquired through direct contact with reality and the accumulation of empirical evidence. Prominent empiricist philosophers include John Locke (1690), David Hume (1748), and John Stuart Mill (1843). One can consider Ants following this idea of Knowledge. In this case Knowledge is not accumulated in a single Ant, but in a statistical combination of pheromones ants produce. The knowledge about the shortests path towards the food source comes from the most “smelling path”. Ants feel it with their sensorics and thus can use the Common knowledge about the shortest path towards food or home. Following Aristotle and Stoic’s, Ant has only passive intellect, that maximises its own utility for the colony. Meaning, this species are not intellectual, from Aristotle point of view, but are able to generate Common Knowledge, having just passive intellect. Passive intellect does not generate Individual knowledge, but communities of species, having passive intellect, can generate Common Knowledge about the real world. Genetic algorithms (or Evolution) are also a good example of a passive intellect. Most of the bio inspired global optimisation algorithms would fit into this notation.
• Rationalists emphasise the role of reason and innate ideas in acquiring knowledge. They argue that certain truths can be known independently of sensory experience through rational reflection and intuition. Rationalist philosophers such as René Descartes (1641), Gottfried Leibniz (1714), and Immanuel Kant (1781) have explored this perspective. Rationalists put more attention to the active intellect in Aristotle notations. Meaning, that Active intellect — is a collection of abstract ideas about the world that can collaborate within this active intellect, that is why it is called active. Important is that active intellect is “empty” from the very beginning and is able to acquire the knowledge via some process, that is called learning or training, while passive intellect is built from the very beginning.
• Constructivists propose that knowledge is actively constructed by individuals based on their experiences, interactions with others, and mental frameworks or schemas. According to this view, knowledge is not simply passively received but actively created through cognitive processes such as assimilation and accommodation. Jean Piaget (1952) and Lev Vygotsky (1978) are influential figures in constructivist theories of knowledge. Constructivists introduced one more level of complication for the learning process making it a result of a collaboration of individuals, known as a society. Here this is interesting, since we know the role of society in passive intellect by Aristotle, but here we first see that society plays an important role in active intellect as well, creating Common Knowledge of abstract ideas.

The mechanics of different knowledge generation is as follows: conflict of empirical evidence (passive intelligence) with Common Knowledge triggers active intelligence. The outcome of active intelligence generates new abstract ideas which contribute to Common Knowledge after interaction with other individuals, that makes a role of a society relatively high.

These perspectives provide different lenses through which knowledge can be understood:

• Passive knowledge (Empiricists) obtained by Passive intelligence operating with sensorics-based experiences,
• Active knowledge (Rationalists) obtained by Active intelligence of an individual operating with abstract ideas and concepts,
• Common knowledge (Constructivists), being a product of society voting mechanism on individuals’ active knowledge

The role of Language

Many species have developed language as a means of communication, enabling them to describe objects within their passive knowledge using speech sounds, visual symbols like pictures, or other forms. For example, when a bird spots a food source, it emits a distinct sound that signifies “food” to other birds. Similar communication patterns exist for identifying enemies, water sources, and other reflexive behaviours. Through the exchange of passive knowledge about the real world through language, species can begin to generate Common knowledge.

In this process, the mechanism becomes more complex and aligns closely with principles found in natural language processing (NLP) and large language models (LLM). A vocabulary of words is established, which corresponds to specific objects. Over time, through evolutionary principles, the number of neurons in the brain of certain species has increased (hypothesis on that is that Stoic’s argument on rationalism went further and evolved in more definitions of what is “rational”, driving the necessity for more reactions and thus higher capacity of the brain). This expansion allows for greater prompts, leading to higher levels of inference and a more sophisticated range of inputs and outputs. Consequently, a self-assurance mechanism is formed, enabling the generation of active knowledge — classes derived from passive knowledge — which results in predictable consequences for objects and corresponding reactions.

By taking into account these notations and factors, one can conceptualize passive knowledge as an autonomous lexicon of objects, while active knowledge can be likened to a textual representation elucidating the interactions among these objects within the tangible realm. The capacity for cognitive thought is accompanied by the ability to generate active knowledge independently, without necessitating direct engagement with the external environment. This generation can occur autonomously or through communication with other individuals. Here, the principle of rationalism once again emerges.

Communication with the real world is a protracted and perilous process. Conversely, communication with fellow members of one’s society presents a less hazardous and more expedient means of acquiring new active knowledge. It is crucial to note that passive knowledge can only be acquired through collaboration with the real world, whereas active knowledge can be transmitted via language. To illustrate this distinction, one may consider that comprehending how to perform a task differs from actually executing it, representing two distinct levels of understanding.

Despite the last fact, societies managed to learn how to create the Common knowledge that became a foundation for civilisation development. Common knowledge — a combination of active knowledge (texts, videos, music etc.) and passive knowledge (vocabularies) stored on external data storages.

That is what defines human intelligence from the beginning till the very end— ability to learn from the real world and generate Passive knowledge about the world, learn Language to be able to generate Active knowledge, use Common knowledge to foster the active knowledge and generate more active knowledge being rational about the individual it belongs to.

Hypothesis 1: Passive intelligence is merely a property of a multi-dimensional environment.

Based on the concept discussed above, we have generated the following hypothesis: intelligence comes from the large dimensionality of the real world.

What does this mean? For the time being we will consider only Passive knowledge, that generates a vocabulary of the brain of any species that have intelligence.

Each object in this case is represented with a multi dimensional embedding mechanism. Consider human sensorics.

The human body has several sensory systems that allow us to perceive and interact with the world around us. The five main senses are sight (vision), hearing (audition), touch (gustation), smell (olfaction), and touch (somatosensation). Each sense relies on specific sensory organs and physiological processes, which can be explained by various principles of physics.

1. Sight (Vision): The sense of sight is enabled by the eyes, which contain specialized cells called photoreceptors. These photoreceptors, known as rods and cones, convert light energy into electrical signals that are transmitted to the brain via the optic nerve. The physics of vision involves the properties of light, such as its wavelength and intensity, which determine how it interacts with objects, reflects off surfaces, and enters the eye to form an image.
2. Hearing (Audition): Hearing is made possible by the ears, which consist of the outer ear, middle ear, and inner ear. Sound waves in the environment are collected by the outer ear and channelled into the ear canal. These sound waves then vibrate the eardrum and tiny bones in the middle ear, transmitting the vibrations to the cochlea in the inner ear. Inside the cochlea, hair cells convert these mechanical vibrations into electrical signals that are sent to the brain for interpretation. The physics of hearing involves understanding how sound waves propagate through different mediums and how they are transformed into mechanical vibrations within the ear.
3. Taste (Gustation): The sense of taste is mediated by taste buds located on the tongue and other parts of the mouth. Taste buds contain specialised receptor cells that detect different chemical compounds present in food or drink. When these compounds dissolve in saliva and come into contact with taste receptors, they trigger electrical signals that are transmitted to the brain for perception. The physics involved in taste perception primarily relates to molecular interactions between chemicals and taste receptors.
4. Smell (Olfaction): Smell is facilitated by olfactory receptors located in the nasal cavity. These receptors detect and respond to various odour molecules present in the air. When we inhale, these odour molecules bind to olfactory receptor cells, initiating electrical signals that are transmitted to the brain for interpretation. The physics of smell involves understanding how odour molecules interact with olfactory receptors and how these interactions generate neural signals.
5. Touch (Somatosensation): The sense of touch involves various types of sensory receptors distributed throughout the skin and other tissues. These receptors detect mechanical stimuli such as pressure, temperature, pain, and vibration. When stimulated, these receptors generate electrical signals that are transmitted through nerve fibers to the brain for processing. The physics of touch encompasses understanding how mechanical forces are detected by different types of sensory receptors and how they are converted into electrical signals.

Example

When considering the human body’s five senses in relation to perceiving an object like a tree, here’s how each sense might experience and interpret it:

1. Sight: The visual perception of a tree involves observing its physical characteristics. A person would see the tree’s shape, size, colour, texture, and details such as leaves, branches, and bark. They may notice the tree’s height, the patterns formed by its branches, and any unique features it possesses.
2. Hearing: Although trees are not typically associated with distinct sounds, there are auditory elements related to them. For example, a person near a tree might hear the rustling of leaves as the wind blows through them or the creaking of branches in response to movement. Additionally, they may hear birds chirping or insects buzzing around the tree.
3. Touch: When touching a tree, one can feel its various textures and surfaces. Running their hands over the roughness of the bark provides tactile feedback. They may also feel the smoothness of leaves or the prickliness of thorns if applicable. The temperature of the tree’s surface can also be sensed — it might feel cool in shade or warm when exposed to sunlight.
4. Smell: Trees can emit distinctive scents that vary depending on their species or surrounding environment. For instance, being near a pine tree could evoke a fresh and resinous aroma, while flowers blooming on certain trees might produce fragrant smells. Additionally, the earthy scent of soil and vegetation around the tree contributes to the overall olfactory experience.
5. Taste: While taste is not commonly associated with perceiving trees directly, some parts of certain trees can be consumed as food or used for flavouring purposes. For example, fruits from fruit-bearing trees offer tastes ranging from sweet to sour or bitter. However, caution should be exercised as not all parts of trees are edible and some may even be toxic.

It’s important to note that the sensory experience of a tree can vary depending on individual perception, environmental factors, and cultural influences. Additionally, some senses may be more dominant than others in perceiving certain aspects of a tree.

The consolidation of all sensory signals processed by functional zones (passive intelligence) becomes a passive knowledge. Basically, passive intelligence serves as a mechanic that generates a vocabulary of objects (embedding vectors) in a multidimensional space. And we believe that multidimensionality of vocabularies of a brain might bring potential benefits:

• better classification accuracy in multidimensional space, assuming that the curse of dimensionality is addressed by parallelizing deep neural network
• fuzzy principles of an “acceptable quality” of restoration of different memories under a reduced prompt dimensionality (e.g. dejavu effects or minor music restoring a chain of sad moments in life from the past).

Hypothesis 2: Active intelligence is the language.

Ideas and concepts represent generalisations of empirical-driven passive knowledge.

• By definition, generalisations describe causal and temporal object interactions.
• Language, specifically its grammar, is focused on conveying these causal and temporal object (artefact of passive knowledge) interactions.
• So, specific concepts are artefact of corresponding language, and they do not exist outside the language.

Using these concepts and artefacts of passive knowledge using grammar one can generate active knowledge. The hypothesis is that this process, Active intelligence, is completely defined by the Language.

Hypothesis 3: Common knowledge is the main mechanism of human civilization development.

History demonstrates that human civilization development has been significantly fostered by an ability to store active knowledge. The review of stored active knowledge together with the voting mechanism of society naturally transitioned active knowledge into society (common intelligence) verified common knowledge.

“...We are a scientific civilization. That means a civilization in which knowledge and its integrity are crucial...“ - Jacob Bronowski

Ability to store active knowledge on physical devices (wood, clay, paper, magnetic disks, internet) allows each next generation to continue from the last verified common knowledge, instead of starting from scratch (passive knowledge).

Conclusions

Combined together our hypotheses describe intelligence as a rational process of knowledge generation based on sensorics:

1. Forming passive knowledge from sensorics: sensorics (passive intelligence) allow us to build a passive “prompt” (sensory information induced into the functional zones of the brain) that describes the current situation, happening in the real world. Since the brain receives multidimensional prompts from all functional zones, the self attention mechanism allows it to get this right focus based via the intensity of the sensorics signals.
2. Forming active knowledge from passive knowledge: formed passive knowledge acts as an active “prompt” and triggers active intelligence to generate active knowledge and an output reaction — a response. Comparing with LLMs architecture, the only difference in this case, is that the “prompt” is multidimensional, where tokens reflect objects of passive knowledge and each object has its individual importance (attention) within the “message” (this is done via the amount of neurotransmitters each zone produces for each object of passive knowledge, Purves, 2018). Being rational means in this case, that the output with the highest amount of neurotransmitters is selected as the right one, similar to the softmax function within LLMs. Just since the processes within the brain are not binary, the humans have a certain confidence and can feel doubts about the decisions and the level of these decision rationality, expressing this sometimes openly and sometimes not.
3. Forming common knowledge from active knowledge. Society (common intelligence) acts as an ensemble voting mechanism verifying active knowledge of individuals (Condorcet, 1785). By storing common knowledge one emulates the “warm start” concept in Deep Learning.

From the hypotheses above one can derive the crucial difference between the proposed definition of intelligence and LLMs architecture:

• In (human) intelligence input is a passive knowledge (multidimensional sensory embeddings) obtained by sensorics
• In LLMs input is a common knowledge, i.e., high-level derivative of passive knowledge generated by active intelligence and verified by society.

References:

1. Gardner, H. (1983). Frames of mind: The theory of multiple intelligences. Basic Books.
2. Hickok, G., & Poeppel, D. (2007). The cortical organization of speech processing. Nature Reviews Neuroscience, 8(5), 393–402.
3. Miller, E. K., & Cohen, J. D. (2001). An integrative theory of prefrontal cortex function. Annual Review of Neuroscience, 24(1), 167–202.
4. Plomin, R., & Deary, I. J. (2015). Genetics and intelligence differences: Five special findings. Molecular Psychiatry, 20(1), 98–108.
5. Spearman, C. (1904). General intelligence objectively determined and measured. American Journal of Psychology, 15(2), 201–293.
6. Locke, J. (1690). An Essay Concerning Human Understanding.
7. Hume, D. (1748). An Enquiry Concerning Human Understanding.
8. Mill, J. S. (1843). A System of Logic.
9. Descartes, R. (1641). Meditations on First Philosophy.
10. Leibniz, G. W. (1714). Monadology.
11. Kant, I. (1781). Critique of Pure Reason.
12. Purves, D., Augustine, GJ, Fitzpatrick, D., Hall, WC, LaMantia, AS, McNamara, JO, & White, LE (2018). Neuroscience (6th ed.). Sinauer Associates.
13. Siegel, GJ, Albers, RW, Brady, ST, & Price, DL (Eds.). (2012). Basic Neurochemistry: Principles of Molecular, Cellular and Medical Neurobiology (8th ed.). Academic Press.
14. Marquis de Condorcet (1785). Essay on the Application of Analysis to the Probability of Majority Decisions.