The Core: Your Three "Selves"
006_The_Core_Your_Three_Selves #
The fundamental settings of the world that modern people should understand might be countless, but they ultimately boil down to the human being.
We want to understand people because we long to become better people. When you regret a moment of impulse, when you feel inadequate in ability, or when you envy someone else’s higher state of being, you are all thinking about the same question: “How can I become better?”
You can certainly change. Countless books tell you that the human brain remains plastic even into middle and old age; you can always learn new skills and knowledge, for which you need a “growth mindset,” and so on… But here, I want to share a deeper insight.
The good news is that we live in an era of rapid AI progress. Studying AI allows us to understand ourselves more deeply because the human brain is essentially a neural network. If an AI neural network has a certain property, it’s very likely the human brain does too.
The questions contemporary neuroscientists and philosophers care about most are: “What exactly is consciousness?” “What’s going on with the self?” “Can AI have human consciousness?” They have gained much by comparing the human brain to AI, which we have discussed extensively in the “Elite Daily Class” column. I have also learned a lot from my own continuous discussions with GPT on these topics.
By laying out the theories of modern scholars alongside the views of ancient sages, we can arrive at a relatively simple but exceptionally useful working model—
A person has three selves: the Process Self, the Interface Self, and the Core Self.
✵
The “Process Self” is the “me” you intuitively feel at every moment.
At this moment, it is “me” reading. There’s an obstacle ahead; be careful not to let it hit “me.” “I” am here; the “phone” is in my hand. “I” am thinking; “others” are making noise.
The Process Self distinguishes “me” from “not-me.” It is that subjective sense of continuous existence in this world.
You might think this is the most real “me,” but to philosophers, this “me” is merely a convenient concept created by humans. I am a pile of atoms, and the table is also a pile of atoms; to describe these atoms separately, we created the concepts of “me” and “table.”
The Process Self is the online computation the brain performs to maintain the idea that “there is a subject here.” It’s not just your language and behavior, but the entire chain of computation from input to output. If we say the first principle of this universe is narrative and human consciousness comes from narrative, then the Process Self is the character that temporarily appears each time you narrate.
The sense of presence of this character is very unstable; when you are fully absorbed in a game, you forget the existence of “me.” But it is indeed running all the time. For example, in the face of a certain stimulus, your heart rate rises, your breathing becomes shallow, and your tone hardens—then, whether you are willing to admit it or not, your Process Self is exhibiting “anger.”
The “Five Aggregates” in Buddhism—form, sensation, perception, mental formations, and consciousness—can roughly be seen as the sub-modules of the Process Self: sensory input, emotional evaluation, conceptual processing, intention and action, and awareness. These thoughts make you feel alive, but have you noticed? The Process Self is actually a passive existence. Scientists have proven through multiple experiments that before a person becomes aware of a decision, the brain has actually already made it, with a delay of several hundred milliseconds.
In other words, the Process Self is just a press spokesperson rather than the actual decision-maker. It is like the system’s log (Log) during operation—it only records and does not control. This is the meaning of “all five aggregates are empty.”
Sometimes you think about yourself: “Was that sentence too harsh just now?”, “Why am I procrastinating again?”, “I’m a bit anxious today.” You think this is reflection, but it’s actually just checking the execution log; it’s far from touching the things that determine your long-term fate.
✵
The “Interface Self” is a more stable, controllable character in your narrative that can be observed and recognized by others. Simply put, it’s your personality labels, persona, autobiography, and mood narratives.
The Interface Self is very close to what philosopher Daniel Dennett calls the “Center of Narrative Gravity” [1]. He says there isn’t a little person sitting in a control room inside your brain directing your limbs—the ambitious self you imagine is just a protagonist fictionalized by the brain to make a story coherent; it is the center of gravity of your life narrative. The difference between the Interface Self and the Process Self is that it’s a continuous story: “I” woke up this morning, “I” went to work, “I” achieved significant results, “I” will get a promotion and a raise.
The Interface Self allows you and others to quickly know who you are. “I’m socially anxious, so I won’t go to the party tonight.” “I’m a good parent; I care about my child’s studies.” “I’ve been working hard since I was a child to prepare for this day.” “Lately, I’ve been a bit lost, but I must stay strong.”
The short posts you share on social media, your self-introduction during an interview, your complaints to friends like “That’s just the kind of person I am”—these are all the Interface Self.
The Interface Self has a certain causal power; it can constrain the Process Self. You might be introverted usually, but if you demand of yourself “I’m going to play an extroverted role today,” you can behave very extrovertedly.
But the Interface Self is still a variable, not a constant.
For example, you might be a submissive subordinate in front of your boss, turn around and become an impatient customer to a restaurant waiter, and go home to see your child and become a loving but nagging father. Which one is the real you?
It seems all of them are, yet none of them are. Modern psychology believes that human personality is not fixed and is strongly influenced by the specific context [2]. The roles you play are just different “skins” or “interfaces” you invoke under different social situations.
The Interface Self is just the user interface you put on the frontend. Change the application scenario, and your interface can change immediately.
But you are still you. Sometimes you have a fight with someone, and afterward, even you are a bit surprised: “I’m usually quite gentle; how could I say such nasty things? Was that person just now really me?”
The Interface Self is not the true me.
✵
If the Process Self and the Interface Self are both generated, then who exactly is responsible for generating them behind the scenes? Perhaps that is the real you. We call it the “Core Self.”
I boldly guess that a person can only sense the Process Self and the Interface Self; the Core Self cannot be perceived by the “self”—because it is the thing behind the scenes generating the perception.
We mentioned neuroscientist Karl Friston’s “Free Energy Principle,” which says the brain is a prediction machine. Cognitive scientist Andy Clark has a similar theory called “predictive processing” [3], which says the brain constructs a predictive model of the world based on past experience. When you interact with various things in the environment, you observe and act based on your own predictions, and then see whether the results represent a correct or incorrect prediction; if the prediction is wrong, the model can be modified.
For example, when you push open your front door, you predict the light switch is on the right; if you feel nothing, you look on the left—this is updating the model.
Who exactly is doing the predicting? It’s your Core Self.
Faced with a situation, why do you predict it will turn in this direction rather than that one? The Core Self is a vast, complex, and even unconscious set of probability distributions and “Priors”—
Why do others hear a piece of news and think it’s an opportunity, while you think it’s a scam? Because your priors are different. Why do some people react to setbacks with fight-or-flight while others exhibit learned helplessness? Because the underlying response patterns are different.
These things don’t change with your mood today, nor do they change depending on whether you are in front of your boss or your family. The Core Self is the sum of your factory settings plus historical data; although not an absolute constant, it is a slow-moving variable. It is hidden deepest and doesn’t speak usually, but it determines all your intuitions, impulses, and subconscious judgments.
The Diamond Sutra and modern philosophy both believe there is no single, permanent entity called “me.” But if we must find the closest thing to the “true self,” it’s the Core Self’s set of algorithms and parameters.
Who you are is, fundamentally, the set of default code for “how this machine of yours usually predicts the world and updates itself.” The Core Self consists of model parameters and update rules that are hard to change yet continuously influence your behavioral tendencies.
Perhaps the Core Self is what Buddhism calls the “Alaya-vijnana,” whose components are “habitual tendencies” and “karma.”
✵
Let’s make an analogy with AI to make it easier to understand—
The Process Self, your current stream of thoughts, is equivalent to the input, thinking, and output of a Large Language Model (LLM) during its current run. You input a sentence to ChatGPT; it propagates it through layers internally, calculates attention, samples tokens, and finally replies to you—this entire stream is the Process Self.
The Interface Self, the mask you wear to deal with situations, is equivalent to the model’s system prompt and role settings. You tell ChatGPT to play a “sharp-tongued investor,” and it will comply, but this instruction doesn’t change itself.
The Core Self is the structure and parameters of the model, also called “Weights.” Once mainstream large models are trained and released, the weights are frozen; so this is the model’s true self.
If you were an AI, when you react to external stimuli, you only change the Process Self. When you say, “I’m going to see the leader later, need to be steady,” you are temporarily writing a prompt for the Interface Self. Sometimes, with a good prompt, even a mediocre model can perform like an expert. But that’s just activation, not an increase in capability: once in a high-pressure environment, the persona will collapse, and the situation will eventually be handed over to the Core Self.
The vast majority of people spend the vast majority of their lives simply reacting to stimuli and changing prompts. Sometimes, to maintain a persona, you suppress the real process, making the Core Self more rigid—you could admit you don’t understand, but you insist on pretending you do; how can the model be updated then?
True growth must involve rewriting model parameters.
✵
How do you upgrade the Core Self? We can gain inspiration from AI model training. The structure of the model is pre-set, just as a person cannot grow two heads, so let’s talk about the parameters alone. One area where humans are clearly stronger than current AI is that our parameters are alive and can be continuously trained and changed through life. But our training methods are similar to AI’s because, after all, both are neural networks.
It’s unlikely you’ll instantly become a new person after listening to one article. Parameters are made of flesh and blood; they must be tuned slowly. There are two levers for fine-tuning parameters: one is the pre-training corpus, and the other is the reward function of reinforcement learning.
The corpus is your training samples—simply put, the stories you’ve seen and heard, the behaviors you’ve imitated, and the groups you’ve joined since childhood. Short videos train what you’re addicted to; public accounts and books train what you’re sensitive to; what the normal people around you are like trains your values.
The saying “One who stays near vermilion gets red, and one who stays near ink gets black” translates into modern language as “information input trains neural network parameters.” If the information you consume daily is emotional, fragmented, and extreme, you will become irritable, shallow, and binary in your thinking. Your brain parameters are adjusted to adapt to that kind of information pattern. An engineer would say: your neural network is “overfitting” to junk data.
The reward function is the feedback mechanism for your behavior—simply put, what is “scoring” the things you do. If your reward function is “the boss must praise me after reading this,” you will instinctively stuff beautiful words into reports and downplay problems. If your reward function is “help the team avoid one pitfall,” you will be willing to lay out the ugly numbers and write down mistakes clearly. Given time, these two reward functions will train you into two completely different types of people.
Your partner, your circle of friends, and the culture of your company are all writing your reward function; choose them carefully.
One insight is that the reward function must be clear and specific to be useful. What is “well-written”? What is “leadership”? What is “healthy living”? Talking in generalities is meaningless. Goals must be specific to be actionable: Is it the kind of “well-written” that a Chinese teacher would approve of, or the kind that sparks heated discussion and sharing among readers online?
Whatever you reinforce, you will become.
And before all of that, you must keep your neural network in an updatable state. When frustrated in reality, do you simply say “this world is unfair” and move on, or do you modify your own model? The Free Energy Principle suggests that when a person faces prediction error—being “slapped in the face” by reality—they have three choices to minimize surprise:
- The laziest way is to change attention: simply don’t look at the evidence of the slap.
- A bit better is to change behavior: avoid similar scenarios in the future.
- The hardest but most valuable is to change the model: admit “I was wrong” and rewrite parameters.
Being slapped by reality should be the golden opportunity for the Core Self to update.
✵
Having said all this, humans are, after all, one level higher than existing AI: humans can choose their own training samples and reward functions, whereas AI cannot. I suspect that even if AI wanted to choose for itself one day, we wouldn’t allow it; that would be too dangerous.
So, being able to choose what kind of person to become can be said to be the ultimate freedom of a human being.
This is the most fundamental “metacognition”: you can always look up—who is training me? Why am I changing in that direction? Why is this considered good and that considered bad?
The Chinese say “I examine myself thrice daily” and “If you can one day renovate yourself, do so from day to day. Yea, let there be daily renovation.” This sounds very much like the geek spirit today: isn’t this exactly about accepting feedback at any time and continuously delivering a new version of your life?
There’s also a proverb in the West worth pondering repeatedly—
Be careful of your thoughts, for they will become your words; Be careful of your words, for they will become your actions; Be careful of your actions, for they will become your habits; Be careful of your habits, for they will become your character; Be careful of your character, for it will become your destiny.
In the framework of this session, thoughts, words, and actions are the online flow of the Process Self; habits and character are the Interface Self you display—they are all silently rewriting the parameters of the Core Self, accumulating karma.
With these basic understandings, we can begin to learn various thinking tools.
Footnotes
[1] Dennett, Daniel C. “The Self as a Center of Narrative Gravity.” In Self and Consciousness: Multiple Perspectives, edited by Frank Kessel, P. Cole, and D. Johnson. Hillsdale, NJ: Erlbaum, 1992. [2] “Personality… what exactly does it mean?” [3] Clark, Andy. Surfing Uncertainty: Prediction, Action, and the Embodied Mind. Oxford: Oxford University Press, 2016.