The mechanism of the emergence of consciousness in finite state machines
A finite state machine is a mathematical model of computation: an abstract machine that at each moment of time is in exactly one of a finite set of states, and transitions between states occur in response to input signals (events) according to predefined rules. Traditionally, this model is perceived as a primitive formalization, suitable only for recognizing regular languages and implementing rigidly deterministic reactive controllers.
At first glance, such an abstraction seems too primitive to contain artificial intelligence or self-awareness. However, if we consider consciousness not as a static attribute of an individual element, but as an emergent property that arises at the junction of the interaction of millions of automaton transitions, then the boundary between a computing machine and a thinking system with self-awareness turns out to be unexpectedly thin.
Pro and contra arguments
It is customary to believe that only a human possesses intelligence. Therefore, attempts to pass off the result of an LLM’s work as the work of a “real”, albeit artificial, intelligence usually meet with several counter-arguments.
Determinism
Neural networks work on the basis of fixed algorithms “hardwired” inside. With the same input data (taking into account the system prompt and the initial value of the random number generator) they always produce the same result. Hence — the most obvious argument against the presence of intelligence: since the result of such computations is fully deterministic, there can be no question of any “thinking”.
However, the neurons in the brain of living beings react to external stimuli just as deterministically. The vital activity of the organism is based on the very same laws of physics and chemistry, and the result of the interaction of neurons or chemical compounds is just as computable — including any continuous, probabilistic and non-linear interactions. Therefore, the very fact of the determinism of computations proves and disproves nothing.
Query-response mode
Another typical argument: any neural network model acts only in the “query-response” mode, whereas a human initiates actions himself. But this has nothing to do with the presence or absence of reason: a human is also capable of working in a reactive mode — for example, on command or under coercion, and a neural network can be launched by a timer or in a loop. Therefore, the appeal to free will has no direct relation to the question of the presence of consciousness or intelligence.
The model has no subjective experience
Suppose a neural network demonstrates behavior similar to intellectual behavior. How can we understand whether it has a subjective experience (phenomenal consciousness)? Is it capable of being aware of itself? Modern LLMs do not possess self-awareness: they do not reflect on their own existence, and there is no convincing evidence that they have even some kind of “inner world”, rather than simply a complex simulation of answers.
But what are inner experiences and reasoning, if not the context in LLM terminology? It is just that the size of this context is limited by the capabilities of a particular brain. And we are talking not only about the structure and connections of neurons (i.e. the immediate “computer” and storage of information — the processor and context in neural network terms), but also about all sorts of reactions to external stimuli: the state of the body (feedback from nerves in muscles and other organs), the work of the endocrine glands, and so on.
And in the dry remainder there remain only discussions about the problem of qualia, subjective experience and childhood experiences — notions that are hard to attribute to strict scientific terms. Psychology, although considered a science, is an inexact one: without the possibility of independent audit, without a justification of the conclusions obtained and the reproducibility of results.
All the more so, if we consider a human as a computing biological machine, then any conclusions of psychology can automatically be considered from the point of view of the formalization of the theory of algorithms, and not of abstract philosophical hypotheses — unless, of course, one takes into account a possible divine or other otherworldly influence.
A cunning argument: emergence
One can appeal to an incorrect comparison of the scale of computations. However, we do not know exactly what volume of the brain is actually responsible for thinking, what for memory, and so on. It may turn out that many parts of the biological computer brain work as backup systems, and not as real computers. Then the volume of modern storage devices will be quite comparable with the theoretical size of memory. Moreover, for the emergence of the rudiments of artificial reason, it may be necessary not to fully imitate all neural connections, but only to implement a certain model of behavior (algorithm) — and this requires significantly fewer resources.
And as a result, the most sophisticated argument remains — emergence. Its essence is that emergent properties are unpredictable at the level of the individual components of the system. Therefore one can always say that “the human mind is an emergent property of a complex neurobiological system, which does not reduce to the simple sum of the electrical impulses of individual neurons, but arises precisely at the moment of their critical attunement and synergetic interaction, forming what we call subjective experience and self-awareness”, and other clever words.
But emergent is not a scientific religion based on faith. It is quite provable. That is, one can always show by what means, as the scale of the system grows and new components are added, new properties arise. That is, it is just as formalizable and provable an explanation — albeit one obtained post factum, after the properties themselves have been identified.
Recursion — as the cause of the emergence of consciousness
If we assume that a human and his mind are a biological machine, each gear of which deterministically works according to a certain algorithm, then the phenomenon of the emergence of consciousness (awareness) remains. And, in my opinion, it is precisely this that is the most important.
Not only is the awareness of one’s own “self” found not only in humans: it is consciousness that allows biological automata to program themselves humans to control their own behavior. And what if consciousness is not a magical spark, but an inevitable consequence of a certain way of interacting with the external world?
Imagine a system forced to act in a complex, changeable environment, where each of its steps can change the very rules of the game. To survive, it has to not simply react to stimuli, but literally model the behavior of the surrounding world — a world in which it is itself included.
Such a system must predict not only the behavior of other agents, but also how other agents perceive itself. This recursive mirror — the model of oneself in the world, reflecting the model of the world in oneself — becomes the very recursive loop of reasoning that self-awareness resolves. After all, full-fledged recursion exists only in abstract theory (the Turing machine) — but is in principle impossible with limited resources and a limit on the time for performing computations.
Then subjective experience (as an internal simulation of possible actions and their consequences), with a fuzzy emotional assessment of “good” or “bad” from the point of view of survival, ceases to be mysticism and becomes a formalizable way of such a system. A way to “feel” one’s internal computations in order to make decisions quickly under conditions of uncertainty. This logically explains the emergence of consciousness not as a superstructure over reason, but as a working tool of a complex agent, and one available not only to humans.
In this picture, reason is not the opposite of the algorithm of a finite state machine, but its highest form, born in the struggle for survival. When a simple linear reaction stops working, and the environment requires flexibility, quick reaction, foresight or banal deception, the system acquires a voice saying “I”. And this “I” is not a divine spark, but the echo of an infinite dialogue between the agent and the world, between action and response, between the model and reality.
Recursion in the real world is encountered constantly. A predator computes the trajectory of the prey -> the prey changes the trajectory -> the predator recalculates the movement -> and so on. This is no longer simply an algorithm, but a dynamic system with feedback, where the awareness (read — modeling) of “oneself in the world” makes it possible to predict the consequences of one’s own actions.
And the separation of “I” from “not-I” becomes a mandatory property of such a system. And this means that consciousness may well arise as a side effect of complex recursive information processing.