), which endowed them with new cognitive abilities for survival through simulation.
neocortex (
Neurologist Mountcastle made several important discoveries about the neocortex:
: The response to sensory stimuli is the same within the vertical columns of the neocortex. 
: There are numerous vertical connections within the columnar structures, but relatively fewer lateral connections between them. : Different regions of the neocortex (such as auditory and visual cortices) have essentially the same arrangement.
Based on these findings, the conclusion is that cortical columns are tightly packed structural units, and different cortical columns execute algorithms with consistency.
Helmholtz machine
Helmholtz proposed that human perception is largely an inferential process; we do not directly perceive all of reality, but instead simulate it through inference. Based on this theory, computer experts Hinton and Dayan developed the Helmholtz machine. Unlike other neural networks that only possess input functions (used for pattern recognition), the Helmholtz machine also adds reverse links (for generating patterns).
This design reflects a feature of the human brain: perception and imagination cannot occur simultaneously, nor can the generation and recognition in neural networks happen at the same time.
The neocortex is actually a "world prediction machine" that can reconstruct the three-dimensional world and predict what will happen next. However, current artificial intelligence systems generally lack the ability to build such "world models".
Three abilities brought by simulation
: For example, when rats perform maze tasks, they explore through mental simulation and plan future actions in advance.
: The brain learns from unchosen alternative paths by constructing causal relationships, understanding the possible different outcomes if different choices had been made at the time.
: It allows recalling specific past experiences or fragments, which differs from procedural memory. Procedural memory refers to birds learning to fly or children learning to ride bicycles, while episodic memory involves recalling specific events.
Model-based reinforcement learning
. Among these, the aPFC excels at simulating the behavior of other animals, inferring the intentions behind their actions, and using these intentions to predict the animals' next moves.
thinking).
with deep learning techniques, allowing it to predict potential outcomes of different choices based on the current state (i.e., building a model) and continuously optimize strategies during exploration. This process is similar to how mammals pause to think in complex situations, predicting future possibilities and making goal-directed decisions.
Dishwashing robot
The role of the motor cortex is not to directly produce movement commands, but to generate movement predictions and convert these predictions into actual actions by adjusting neural circuits. It continuously observes the state of body movements, interprets these actions, and predicts subsequent behaviors. The motor cortex's capabilities in sensation and motor planning enable early mammals to learn and perform precise actions.
In this process, the aPFC (agranular prefrontal cortex) does not need to focus on the specific motor details required to achieve goals; it is only responsible for high-level navigation paths. Similarly, the motor cortex does not need to care about the higher-level objectives of behavior, but only focuses on achieving specific low-level motor goals, such as picking up a cup or playing a particular chord.
By reverse-engineering the mammalian motor system, this ultimate application can be extended to dishwashing robots. Dishwashing robots can rely on a similar mechanism for task division, where the motor cortex handles the fine movements required to execute tasks, while the aPFC focuses on planning higher-level task goals.