Discussions about artificial general intelligence often drift into comparisons with the human brain. People talk about neurons, biology, evolution, or quantum effects in the brain. While these are fascinating topics, they also introduce a large amount of uncertainty because we still do not fully understand how consciousness works in biological systems. If we want to reason clearly, it can be useful to temporarily set biology aside and focus on something we do understand very well: computers.

Modern AI systems ultimately run on processors executing instructions. Regardless of how sophisticated the model is, underneath the layers of abstraction there is still a machine transforming state. A program takes some input, applies a set of rules, and produces an output. If the program, parameters, input, and system state remain unchanged, the output must also remain unchanged.

This property is fundamental to computing. A program is effectively a state transition system. At any moment the machine has a state that includes memory, registers, the program counter, model weights, intermediate values, and potentially the state of a random number generator. When an instruction executes, the machine moves from one state to the next according to a deterministic rule.

In simplified form, the system behaves like this:

next_state = f(current_state, input)

Given the same state and input, the same transition occurs. The machine simply moves through a sequence of states defined by the program.

Artificial intelligence models do not escape this framework. A neural network, regardless of size, ultimately performs numerical operations such as matrix multiplication and nonlinear transformations. Tokens become vectors, vectors are multiplied by matrices, activations are applied, and probabilities for the next token are produced. The system then selects the next token and continues the process.

From the machine’s perspective, these are simply numerical transformations. There is no inherent meaning inside the computer. Meaning appears because humans interpret the output.

For example, the bit pattern 01000001 might represent the character “A” in ASCII. Inside the machine, however, it is just a number. The processor has no awareness that this number corresponds to a letter or part of a word. It simply moves and transforms bit patterns according to rules.

This observation leads to an interesting philosophical question. If a system only performs rule-based transformations on symbols or numbers, can subjective experience ever arise from such a process? Even if the system behaves in ways that appear intelligent, does it truly understand anything, or is it merely simulating understanding?

At this point the discussion often turns toward randomness.

Many people assume randomness introduces something fundamentally new into computation. If a system produces unpredictable outputs, perhaps this unpredictability could be the seed of consciousness?

However, randomness in computers is usually pseudo-random. A deterministic algorithm generates sequences of numbers based on a seed value. If the seed is known, the sequence is entirely predictable. Even when hardware sources of entropy are used, such as thermal noise or radioactive decay, the program still behaves as a state transition system. The only difference is that one of the inputs to the transition function is now unpredictable.

Conceptually, the system simply becomes:

next_state = f(current_state, input, random_value)

Randomness introduces variability, but it does not fundamentally change the nature of the computation. The program still follows rules that determine how the state evolves.

To see why randomness alone cannot explain consciousness, imagine a very simple program that prints either “A” or “B” depending on a random number. The result is unpredictable, yet the system clearly has no awareness. Randomness can generate noise or exploration, but it does not produce structured understanding.

Randomness is useful in many algorithms. Monte Carlo methods, for instance, use it to explore large solution spaces. But the intelligence of these systems comes from how outcomes are evaluated, not from the randomness itself.

Randomness therefore appears insufficient as the origin of consciousness. It adds unpredictability, but unpredictability alone does not create experience.

If consciousness were to emerge from computation, it would likely require something more structured. It might involve persistent internal models of the world, a representation of the system itself within those models, and mechanisms that integrate information across many subsystems. In other words, it would arise from organised information processing rather than noise.

Yet even with these mechanisms, the question remains open. A computer may convincingly simulate behaviour associated with awareness while still performing nothing more than deterministic transformations of state. From the outside, the distinction between genuine consciousness and a sufficiently sophisticated simulation might be impossible to detect.

This leaves us with a deeper question. Is consciousness simply a property that emerges from sufficiently complex information processing systems, or is it something fundamentally different from computation altogether?

At the moment, we do not know. What we do know is that computers operate as state machines transforming bits according to rules. Whether that framework can ever produce subjective experience is one of the most interesting unanswered questions in both computer science and philosophy.