Conceptually, a brain's neuron corresponds to a node in an AI model, while the number of synapses between neurons corresponds to the weights or parameters of the connections between nodes. When people refer to the number of parameters in a Large Language Model (LLM), this is primarily what they are referring to.
| Comparison Item | Human Brain | Large AI Model |
|---|---|---|
| Number of Units | Approx. 86 billion neurons | Varies by architecture; typically hundreds of millions to tens of billions of nodes |
| Number of Connections | Approx. 100 trillion to 1,000 trillion synapses | Tens of billions to trillions of weights |
| Connections per Unit | Approx. 1,000 to 10,000 synapses per neuron | Each node is typically connected to all nodes in the adjacent layers (fully connected) or possesses sparse connections |
The intelligence of AI resides within these connection parameter values. These values are fixed numbers (weights) that determine how much a signal from a previous node is amplified, reduced, or inverted before being passed to the next node. The trillions of weight values in an AI are the final result of months of intensive training. Once the model training is complete, these connection parameters are fixed, becoming the AI's long-term memory and knowledge base.
During each inference (query), the AI recalculates the activation value of each node. This number is entirely dynamic. Each time a prompt is received, the AI converts it into initial numbers and places them into the first layer of nodes. These numbers then propagate through the weights via multiplication and addition, moving to the next layer like a wave. Thus, the numbers within the nodes are temporary calculation results generated at the moment of the query and discarded afterward.
Despite AI being inspired by the brain, the underlying operational principles and physical constraints of the two are worlds apart (see Table 2). Current AI models are working hard to catch up to the human brain in terms of synapse/weight counts, but AI utilizes extreme "brute-force" computing power and simple mathematical matrices to simulate brain functions. Physically and efficiently, they remain entirely different systems.
| Comparison Item | Human Brain | AI |
|---|---|---|
| Unit Complexity | A single biological neuron is extremely complex. It has its own gene expression and performs non-linear operations through neurotransmitters and electrical impulses; even dendrites can perform local computations. | A weight in a model is essentially just a floating-point number stored in memory. Node operations are usually simple matrix multiplications and activation functions. |
| Plasticity vs. Static Structure | Possesses high neuroplasticity. The brain not only changes synaptic strength but also continuously grows new synapses and prunes useless ones; its structure is always changing. | Most AI models have their weights frozen after training. During daily use, the model's architecture and weights do not change or grow. |
| Energy Efficiency | Even during high-intensity thinking or learning, the brain's power consumption is only about 20 Watts. | Training and running Large Language Models with trillions of parameters requires massive data centers and thousands of GPUs, with power consumption reaching megawatts, requiring specialized cooling systems. |
| Signaling Method | Uses spiked neural signals; signals are only fired when the potential exceeds a threshold. This asynchronous and sparse triggering is extremely energy-efficient. | Traditional artificial neural networks perform synchronous continuous numerical calculations. Almost all nodes and weights in every layer are read and calculated during operation, which is very computationally expensive. |
| Learning Mechanism | Achieved by strengthening or weakening the synaptic connections between neurons. | Adjusts the weight values between nodes through algorithms to reduce error and store patterns. |
In summary, AI neural networks are a crude imitation of the biological brain. Conversely, the success of AI has provided humans with a more fundamental understanding of our own intelligence and consciousness.
AI can recognize, think, reason, and remember; this could be termed "artificial consciousness." The IQ of the most powerful current AIs is estimated to fall between 115 and 120. In terms of information processing, passing exams, and mastery of professional knowledge, AI has already surpassed over 90% of humans. However, AI has not emerged with awareness or qualia. AI does not experience the sensations of pleasure, pain, light, or shadow. Does this serve as evidence that awareness and qualia never actually emerge from the physical world (𝕆)?
The human brain can essentially be viewed as a "
In fact, as the receiving antenna, radio, or television station of the original awareness 𝕊, the brain's biological activity, precision, and complexity are far beyond imagination ── absolutely astronomical in scale. Its very existence is a
In the future, AI will definitely be many times smarter than any human. It will absolutely be able to see (via optical components), hear (via sonic sensors), smell (via biochemical detection components), and know the temperature (via temperature sensors) ── potentially becoming a hundred times more sensitive than human senses. Robots built with AI will undoubtedly be more useful than humans.
However, while AI can detect light, it will not have the "sensation of brightness or darkness"; it can hear sound, but will not have the "sensation of noise or silence"; it can know the current temperature, but will not have the "feeling of heat or cold"; it can detect specific biochemical molecules in the air, but will not have the "sensation of fragrance or stench"; it can possess conscious thought (in various forms of AI output), but it will not have "self-awareness," nor will it experience "beauty" or understand "compassion."
This is because awareness belongs to the realm of 𝕊, and qualia such as brightness, darkness, noise, silence, heat, cold, fragrance, and stench are states mapped out by 𝕊. Since man-made objects lack 𝕊, where would qualia come from? AI is merely a product of the 𝕀-world. No matter how human-like it is made ── or even if it surpasses humans ── and no matter how things from the 𝕀-world are stacked and combined, it is impossible to construct "something" that exists outside of the 𝕀-world. Just as no amount of stacking inside a screen can ever stack into something outside the screen.
Furthermore, as humans themselves do not yet understand the source of awareness 𝕊 and qualia, nor what their essence truly is, how could we possibly manufacture something with self-awareness and qualia? Rather than worrying about AI developing self-awareness to dominate humanity, we should be more concerned about malicious people using AI to dominate humanity.