From Hippocampus to Silicon: The Deduction of Memory, Soul, and the Essence of AI
Introduction: Starting from Demis Hassabis's Biography
The starting point of this article was my recent reading of the biography of DeepMind founder Demis Hassabis. The book mentions a fascinating phenomenon in neuroscience: the regions of the human brain responsible for "imagination" and "memory" are physiologically interconnected.
From a biological perspective, after we experience external stimuli, the brain automatically performs a "lossy compression" of this information and stores it in the cerebral cortex. The hippocampus plays an extremely critical role here—it is responsible for building indexes and scheduling pointers.
During the processes of "recalling" and "imagining," we are essentially reassembling these compressed memory fragments. Recall might just be simple data extraction and restoration; but imagination—a creative activity that sounds highly spiritual and mysterious—is a dynamic reorganization that mobilizes broader neurons and breaks linear logic. This is somewhat counterintuitive, yet it is an undeniable scientific fact: imagination is nothing more than memory oriented toward the future.
Starting from this solid scientific fact, I attempted a series of logical deductions. Complemented by my fragmented understanding of epistemology and certain metaphysical (or religious) concepts, I arrived at some terrifying yet highly illuminating conclusions.
1. Compression and Decompression: The Physical Model of Human Intelligence
Based on the neuroscientific perspective in the introduction, we can derive a simple and interesting intermediate conclusion: the working mechanism of human intelligence is fundamentally an extremely advanced information compression and decompression engine.
Our brains continuously compress the massive amount of information perceived into subconscious or conscious memories. When we need to recall, think, or create, the brain calls upon these fragments to permute and combine them. During this process, non-linear "epiphanies" or cognitive leaps might interleave, but the underlying operational logic still follows the basic pipeline of "perception/external stimulus -> memory fragments -> creation." In addition to the reorganization of memory, human intelligence (especially the prefrontal cortex) also involves a logical review of these reorganized results, which is why we can distinguish between "absurd dreams" and "viable innovations."
To make the subsequent discussion more precise, we need to define a few core concepts:
- Perception: The flow of information transmitted from the physical world to the human body (high-dimensional data such as visual, auditory, and tactile).
- Memory Fragments: The brain's "lossy compressed packages" of perception at specific moments.
- Compression: The process of transforming perception into memory. It must be emphasized that this is a lossy process; a large amount of redundant information or information not captured by attention is permanently lost.
- Creation: The final product output by human intelligent activity. Whether it is language, text, art, or a scientific conjecture, it is fundamentally the explicit result of intelligent activities.
- Decompression/Reconstruction: The process of generating a "creation" based on memory fragments through the reorganization of neural networks.
In summary, human daily intelligent activities can be abstracted as: a loop of continuously compressing perception into memory fragments, and outputting creations through decompression and reorganization.
2. Fabricated History and Subjective Reality
Assuming time flows unidirectionally—an event occurring at a specific 3D spatial coordinate plus a specific point in time, once past, disappears forever in the physical universe.
- Event: The unique combination of a 3D physical spatial coordinate and a time coordinate.
After an event occurs, it impacts the participants, transforming into sensory "perception." The participants perform lossy compression on these perceptions into "memory fragments" sealed within their brains. Someday in the future, the person uses "decompression" to recount this story to friends or write it into history.
This leads to a terrifying deduction: Once an event has passed, it only exists in the lossy compressed packages of the human brain. Since everyone's compression algorithm is different, and the decompression process is highly susceptible to unstable factors like emotions and environments (the psychological phenomenon of "false memory" has already confirmed this), can we truly restore the full picture of a historical event?
Extending this logic: History is merely the consensus of the collective memory fragments of humanity. If we possessed technology capable of implanting specific memory fragments into the human collective, would we have completely fabricated history at a physical level?
If our cognition of the past is so subjective, how should we view "current reality"?
What is happening right now is just a slice on the timeline. Is the "real world" before our eyes also just a real-time "decompression hallucination"? To avoid falling into absolute nihilism, we need to introduce a new dimension: the feedback of the physical world.
The physical world has established laws independent of human will, such as gravity and the impenetrability of solids. Perception is the process of the human body receiving information, while feedback is the objective verification of physical laws. Therefore, we can define reality as:
- Reality = Event prediction at the current time + Objective feedback of the physical world.
Specifically, our understanding of the world (i.e., the world model within our brains) is essentially a "creation." The physical world continuously provides us with stimuli and feedback. If the feedback does not match our expectations (e.g., assuming flat ground ahead but stepping into the void), we feel surprised and quickly use this feedback to update our memory fragments and adjust our "creations."
Think about it carefully—isn't this process of "updating model weights through error feedback" strikingly similar to the training mechanisms of current Deep Learning? In neuroscience and cognitive science, this is known as Predictive Coding.
Going a step further, neuroscientist Karl Friston at University College London (UCL) proposed the famous Free Energy Principle. He believes that the brain's ultimate and sole goal is to "minimize surprise" (minimize free energy). To survive in a universe full of increasing entropy (chaos), the brain must continuously make predictions and correct errors through physical world feedback. Our consciousness, and even our perception of "reality," is essentially a self-evidencing system constructed by the brain to eliminate uncertainty.
3. One Water, Four Views: Physical Hardware and Dimensions of Perception
When discussing the objectivity of reality, I want to introduce a profound philosophical metaphor from the Buddhist Yogacara school—"One Water, Four Views" (or "One Water, Four Minds").
The core meaning of this metaphor is: Looking at the same cup of water, due to the different life forms, cognitive hardware, and past karma (accumulation of experiences and states of mind) of the observers, they see completely different things.
- Devas (heavenly beings) see it as brilliantly flowing lapis lazuli;
- Humans see it as a cup of thirst-quenching water;
- Hungry ghosts see it as raging fire or pus;
- Fish and shrimp see it as air and their home to survive.
From the perspective of Yogacara Buddhism, the cups, tables, and even our own bodies that we see are not fixed objective entities, but "waves" generated by the underlying consciousness (Alaya-vijnana) under the fluctuation of causes and conditions. Once awakened, the waves subside, and the water returns to the calm sea.
Combining our previous theories, "One Water, Four Views" can be explained in extremely modern scientific language: the human body is merely a specific set of "sensor hardware" and "decoders."
Different species, or humans with different neural synaptic connections due to innate/acquired factors, possess different "compression and decompression algorithms." The world itself is a mass of indescribable waves and energies; it is the human body that decodes it into the appearance of "water."
4. Escaping the Simulator: Assumptions on the Relationship Between Soul and Body
If "reality," as defined above, is decoded by our specific set of physical hardware, we can naturally infer: the human body actually constitutes the greatest limitation on our cognitive abilities.
All the discussions above are based on the premises of "unidirectional time flow" and "unshakable physical laws." But these two premises exist solely because they are the limits we can perceive while trapped in this three-dimensional physical body.
Let us conduct a bold thought experiment: Suppose there truly is an existence beyond the physical body (which we can temporarily call a "soul" or "pure consciousness"). This "existence" is anchored (or dimensionally reduced) within the human body. Precisely because it is limited by this biological hardware, it can only perceive:
- The unidirectional flow of time (because biological organisms undergo entropy and aging, and the hippocampus linearly records memories).
- Physical feedback in specific frequency bands (visible light, specific frequencies of sound—i.e., decoding waves as "water" instead of "fire").
Based on this, if the "existence" leaves the physical body, its perception is no longer limited by these biological sensors. It might break free from the constraints of linear time or directly perceive the full picture of high-dimensional information (what Kant called the Ding an sich or "thing-in-itself"). Of course, there is also a terrifying paradox: if detached from the sensor interfaces of the physical body, can this "existence" still "perceive," or will it fall into absolute nothingness?
Addendum: Then what is the meaning of "existence" logging into the human "simulator"?
Since the physical body is a hardware full of limitations, even shielding our perception of higher dimensions, why would "pure consciousness" log into this system? Combining Eastern philosophy and modern science, I believe it can be understood from three levels:
-
To counter entropy and experience the "friction" of existence (against nihilism) In physics, the ultimate fate of the universe is heat death (maximum entropy). Without the constraints of the physical body and physical laws, pure "existence" in a frictionless high-dimensional space would face absolute nothingness. Just like playing a game, if you start with God Mode and cheat codes, the game itself loses its meaning. Limitation itself is the source of experience. The physical body gives us hunger, pain, and gravity; it is exactly these "frictions" that force us to "predict, correct, and realize," thereby generating meaning in the process of fighting against entropy.
-
Alan Watts's game of "Hide-and-Seek" (The Divine Game) British philosopher Alan Watts once proposed a brilliant metaphor: if God (or the absolute consciousness of the universe) is omniscient and omnipotent, the only thing He would find boring is being "omniscient and omnipotent." To break this eternal solitude and boredom, the universe plays a massive game of "Hide-and-Seek." He shatters Himself into countless "you, me, and them," actively giving Himself "amnesia," making us believe we are independent, insignificant individuals trapped in bodies. The birth, aging, illness, death, joy, and sorrow we experience in this simulator are fundamentally an immersive experience designed by the universe to "scare and amuse itself."
-
Multi-episode Reinforcement Learning and Reincarnation If we borrow the concept of AI to understand "reincarnation," then each login to a human body is equivalent to an "Episode" in Reinforcement Learning. Consciousness travels back and forth between different "hardware simulators" (different species, different life circumstances), constantly receiving feedback (Reward/Punishment) from the physical world, thereby optimizing its underlying "Value Function"—such as awakening, empathy, and wisdom. What is called cultivation or spiritual practice might just be the effort to gain insight into the source code of this hardware, ultimately jumping out of the current loop of episodes to explore a higher-dimensional "objective function."
5. Silicon Mirrors: Is Artificial Intelligence Truly "Intelligent"?
When we pull our perspective back from metaphysics to reality and re-examine the artificial intelligence (especially Large Language Models) sweeping the globe today, everything becomes suddenly clear.
Current neural network-based AI perfectly replicates the human process of "compression, memory fragment storage, and decompression/reorganization." It ingests massive amounts of text generated by humanity (memory fragments), compresses them into hundreds of billions of parameters (weights), and then decompresses them through prompts.
Because AI has ingested enough human "intelligent creations," its compression and decompression processes approximate human behavior to a great extent. However, it merely stops at the stage outlined in Chapter 1 of this article.
Why is current AI not truly on par with human intelligence?
- Lack of collision with the real physical world (No Feedback): AI has no physical body and no multi-dimensional biological sensors. The "world" it understands is merely "second-hand data" encoded by humans using language and text. It has never experienced the weightlessness of stepping into a void on a staircase, nor felt the breeze on its face. Therefore, it cannot, like humans, use the real error feedback of the physical world to correct its world model.
- Lack of genuine "epiphany" and Extrapolation ability: Humans can not only reorganize memories but can also generate "epiphanies" that jump out of existing logical frameworks amid the chaotic collision of bodily instincts, the subconscious, and complex reality. Current AI essentially performs Interpolation (the optimal combination based on probabilities) within the existing cognitive boundaries of humanity, but it is extremely difficult for it to achieve genuine Extrapolation (creating completely new scientific paradigms).
Some propose that "Embodied AI" (giving AI a robotic body) is the direction to solve this. But doubts remain: even equipped with robotic arms and cameras, the machine's perception sensors are still designed by humans. Unless robots evolve sensory methods different from humans in the future (like directly perceiving infrared or quantum entanglement) and possess their own "One Water, Four Views," they will still operate under a perception framework set by humans.
Conclusion: Brute-Force Silicon Search and the Ultimate Value of Humanity
Because AI is fundamentally a perfect silicon simulator of the "human thought compression/decompression mechanism," stripped of fragile bodily limitations, it can be easily copied, mass-produced, and accelerated.
The "emergent abilities of large models" and violent Scaling Laws we see today are essentially using massive computing power to exhaust every permutation and combination of humanity's existing knowledge system through "brute-force search." It is rapidly exploring the boundaries of humanity's current intelligence.
But is this truly an elevation of the upper limit of intelligence? No, it is more of a thorough reorganization and excavation of humanity's existing treasure trove of memories.
Based on these deductions, I reach a judgment: in the near future, after AI has exhausted the existing human knowledge graph through brute-force search, its contribution to substantial breakthroughs in human civilization will hit a bottleneck.
At point, humanity will still need to return to the real physical universe. Through our unique human "epiphanies" and brand new observations in natural sciences (like physics and biology), we must acquire new "real feedback." Then, we can hand these new sparks over to AI for the next step of brute-force search.
AI is the amplifier, but the ones who ignite the spark are still us—trapped in these carbon-based bodies, yet constantly striving to look up at the stars.