Can AI Understand You Better Than You Understand Yourself?

Signup Write New Story Can AI Understand You Better Than You Understand Yourself? by Hanul Lee (이하늘) by Hanul Lee (이하늘) | @ hanulssb Building shared memory for AI agents. Writing Super Second Brain on memory, identity, and human agency.

Building shared memory for AI agents. Writing Super Second Brain on memory, identity, and human agency.

Anyone who has read this far will ask: but is that really you?

It is a fair question. Knowing my birthday and whom I met last week is not enough to be called me. My calendar knows that much.

I approach this question through the word fidelity. I divide the question of how closely the second brain matches me into five layers.

English key to the original diagram. Layer 1, facts: do we know the same facts? Layer 2, importance: do we underline the same things? Layer 3, interpretation: do we read an event through the same values? Layer 4, conclusions: do we reach similar conclusions? Layer 5, revision: when wrong, do we correct ourselves in similar ways? Longer bars indicate greater difficulty in transferring the person intact, not measured performance scores.

Figure 9. Five layers of fidelity. The lower layers can be transferred from records alone. The higher you go, the more you need to know me to transfer them.

Facts can be transferred as long as the records exist. From importance onward, things change. We can read the same meeting minutes and underline different things: someone else marks the revenue figure, while I mark the hesitation in the person’s voice. The position of that underline is me.

The fifth layer is the hardest, and it is the one I watch most closely. What makes a person distinctive emerges more clearly when they are wrong than when they are right: when they notice the mistake, whom they tell first, what they fix first.

This is why fidelity cannot be measured by the quantity of memories. However many records there are, something leaks out in the process of transferring me. I call this the loss rate of the self-model.

When I describe myself, I edit myself. I say, “I tend to keep trusting people to the end.” But the person in the records quietly takes back an assignment after the third mistake. To the extent that I do not know myself, the person I teach the second brain to know differs from the person I actually am.

This is where metacognition—the ability to see oneself accurately—enters. The more accurately people know themselves, the less is lost in the transfer.

There is something I do not want misunderstood. I do not mean that you need strong metacognition to use a second brain. Anyone can use one. It may even be more useful to someone who does not know themselves well.

Recall the gap I described in Chapter 2. Automatically accumulated records do not pass through my editing. That lets the second brain place the person I say I am beside the person the records reveal.

“You said you tend to keep trusting people to the end. But in most cases recorded over the past two years, you took back the assignment after the third mistake. Which of these is your current principle?”

For a long time, I could not answer. That question led me to rewrite my principle. The revised principle went back into the SSOT, and fidelity improved a little.

English key to the original diagram. From the person to the SSOT: I teach it and correct its mistaken understanding. From the SSOT back to the person: the brain reflects me and reveals the gap between my words and actions. At the center: fidelity rises with each circuit of the loop.

Figure 10. I thought I was the one doing the teaching. Half the time, I was the one learning.

A second brain is not a privilege reserved for people with strong metacognition. It is a mechanism through which a person and a digital self correct one another and gradually increase their fidelity.

“Fidelity” and “loss rate of the self-model” are concepts the author uses to describe his experience, rather than validated psychological tests or measured performance metrics. The bar lengths and feedback loop in the diagrams are conceptual.

To decide whether an AI resembles you, what would you look for in the way it corrects a mistake?

This essay is the English translation of Chapter 5 of Super Second Brain, an ongoing Korean whitepaper by Hanul Lee, with an added conceptual clarification. Read the whitepaper series on Substack. Generative AI assisted translation and editing. Both diagrams come from the author’s original whitepaper, with English keys above.

When AI Agents Start Living Your Life for You

How to Build an AI Second Brain That Preserves Evidence and Context