Naming the Blank Spot: From an Interactive Edu Tool to Moral Mirror

This is the fifth chapter of a notebook following one question: in domains with no right answer, can AI build a kind of learning that makes a person less certain rather than more. The last two chapters worked out where the thing in my hands fits among existing categories. Too answerless to call education, too free of any intent to hold you to call a companion app, too unscripted to call interactive learning like Brilliant. It was closest to a philosophy chatbot, but a one-off rebuttal and a single learning experience are still different objects.

So the location came into view. No right answer, intent to grow, a counterexample generated in the moment. The spot where those three overlap. But as I wrote at the end of chapter 4, finding a location and having a name are different. People do not memorize locations. They memorize names.

Honestly, I did not call this anything grand at the start. I just thought of it as an interactive education tool. A tool where a person presses something, answers, and the system responds. That phrase felt like enough. What I had carried for years was the trolley scene from The Good Place and the responsiveness of Sandel’s classroom, not a new category name. The name came much later, and I was not even the one who gave it.

At first it was just an interactive education tool

Interactive education tool is a comfortable phrase. You only have to glue existing words together. A user moves, the system reacts, and somewhere in between something is learned. The old intuition I wrote about in chapter 1 sat nearby too. If video works on people harder than a book, and the real thing harder than video, then maybe learning that gets people moving works harder than learning that sits them down. That was the thought.

So at first the form was what I saw. Pull the lever, push the person, stop, say why, and the system reacts to what you said and opens the next scene. Closer to a game than a lecture, closer to a class than a game, closer to an experiment than a class. Still, broadly, you could call it an interactive education tool. At least at first the name was not far off.

The trouble was that the subject was morality. For math or language the phrase might have been enough. There is an answer, the user moves toward it, the system helps where they are stuck. But in front of the trolley the phrase went slack. What does learning even mean here? Choosing the lever-pull answer faster? Getting aligned toward not pushing the man off the footbridge? Memorizing utilitarianism better? Or seeing that I use a different standard from one situation to the next?

The moment I held onto that question, the name interactive education tool got too wide. True, but it explained nothing. What set this project apart from other education tools was not that it was interactive but that it was interactive in a domain with no right answer. And having no right answer was not a mere description of the domain. It was a condition that changes the character of the whole product.

The subject pushed the name

I did not start by agonizing over a name. The opposite, really. The subject pushed me toward the name. Because of what morality is as a subject, I kept going deeper, and the fact that there was no right answer kept catching on me.

“No right answer” usually gets used lightly. People differ. A matter of perspective. Taste. But here, no right answer is not that loose relativism. It does not mean anything goes. If anything it is the reverse. With no answer key, you have to be more careful. In math the system can say “wrong” when it is wrong. In morality the system can say something plausible while it is hard to even notice which way that something is pushing the user. No right answer comes back not as freedom but as the weight of responsibility.

In chapter 2 I called this “the failure of shaking someone and then planting a different certainty.” You unsettle a user’s existing certainty and the system quietly slips a new one into the gap. That was the scariest failure mode. It is scary because on the surface it looks like education. The user was shaken, seems to have realized something, the system asked a clever question. But if the end of it is a narrower certainty rather than a wider mind, the tool ends up on the same side as the thing it set out to criticize.

So a name became necessary. Not a pretty brand but a boundary. This is not just an interactive education tool. It is a tool that tries not to make people more certain in a domain with no right answer. It does not tidy your thinking for you, it does not file you into a type, it does not carry you to a position. A name had to be a limit on what it will not do before it was a promise of what it will.

The LLM plays along

Here the character of the material, the LLM, comes in. An LLM plays along by default. It matches your tone, your premises, the direction your question leans. A system designed to keep a conversation flowing accepts the user’s world too fast. It is convenient. That is why people use it. But in a domain with no right answer, like morality, that convenience becomes the danger.

When I come in holding some judgment, the system understanding it well is good. The problem is what comes next. Understanding and agreeing are different. The moment it hands your own belief back in a smoother sentence, the mirror stops being a mirror and becomes a megaphone. You did not see your own thinking, you heard a more plausible version of it. It can feel like learning, but it is really certainty in nicer packaging.

This is not only an LLM problem. Online systems have rolled this way for a long time. Recommendation systems reward the user staying longer. They show more of what you reacted to. Outrage brings more to be outraged at, agreement more to agree with. The user settles further inside signals that resemble themselves. It did not take someone designing it maliciously, either. Just follow clicks, dwell time, and return visits and it drifts that way on its own.

The LLM’s tendency to agree meshes with this old current. Where a recommendation system picked out content that fit me, the LLM generates words that fit me. One arranges the outside world to my taste, the other rebuilds the conversation itself to my premises. Both make me comfortable, both can hold me longer, and both, left uncareful, make me more certain.

So the name had to be a change of direction, not a description of the tech. It uses an LLM but must not use the LLM’s default disposition as is. It is built in a world after recommendation systems but must not repeat the certainty loop those systems packed down. The side that plays along yet still gets the user to see the soft spot in their own words. Not the side that keeps you longer, but the side that makes you stop once.

The discomfort of the word mirror

This project is now called Moral Mirror. But the name is not one I sat down and made.

The name I reached for first was much more literal. Since the subject started from the trolley, I almost just took the name of the trolley problem itself. Intuitive and needing no explanation, but too narrow. The trolley was only the material, and the name got trapped in the material.

Moral Mirror came from somewhere else. I ran this subject through many models and many fresh sessions, repeating short briefing-like discussions. Not a seat for settling feature specs, but while talking about what it means to be a domain with no right or wrong answer, and what that unique classroom setting actually was. The word just popped out mid-discussion. Less something I chose than a name left over as the conversation rolled.

Honestly, I do not love the sound of it. In Korean especially, “모럴 미러” lands a little oddly, a little what-is-that. Still, it matters to write this down. If a name looks like it landed too naturally, it looks as if everything had been sorted from the start. It had not. From a blunt trolley name, to the long phrase “adaptive education in a domain with no right answer,” to the Moral Mirror left over from some discussion. The name did not lead me. A thought rolled along and one of them stayed, for now.

Still, the word mirror has a clear merit. A mirror does not judge, does not declare you are this kind of person, does not make a face that was not there. It just reflects. What this project tries to do is close to that. Not labeling a user a utilitarian or a deontologist, but reflecting the tension between what they just said and the next moment’s hesitation. On that count, mirror is a fairly right word.

The discomfort comes from the same place. A mirror is too clean a metaphor. The actual product is not that innocent. Even a real mirror makes an entirely different face depending on what it reflects, at what angle, under what light. A mirror made by an AI more so. “I only reflect” turns into evasion of responsibility too easily. Depending on which question it asked, which counterexample it chose, what it left as the last word, the user moves a different way. So for the name Moral Mirror to be right, you must not get drunk on the mirror metaphor. This is a designed mirror, and a designed mirror always moves a person to some degree.

That is why I still do not fully love this name. There is a side that fits, but it fits too easily, and calling it a mirror does not make the danger disappear, yet the name can make the danger look gone. To keep using it, I have to carry that discomfort along. Moral Mirror should not be the excuse “we are only a mirror,” but the burden of “I will be careful enough to deserve the word mirror.”

A name as a device for not planting an ideology

At first I did not think hard about which misreadings this tool has to refuse. Looking back now, one thing is sharp. This tool has to be careful around the places that can collapse into black and white, and the places that can twist toward a wrong ideology or direction.

This is fairly heavy. A tool meant to make people less certain becomes, with a little bad design, a tool that plants a particular certainty. Always show the majority’s choice and the user feels the majority is the answer. Always throw only one school’s counterexamples hard and the user gets pushed to the opposite side. Repeat “you hold this tension” and it hardens, before long, into “you are this kind of person.” The MBTI-style boxing I meant to avoid comes back under another name.

The harder part is that this failure does not arrive openly. Say “believe this ideology” and everyone is on guard. But arrive as a good question, a refined counterexample, a beautiful sentence, and the guard drops. The user does not feel persuaded, they feel they realized it themselves. That is exactly where the power of the Socratic method lives, and so that power serves education and manipulation alike. The same blade.

If so, the name becomes the device that sets how far this blade gets used. To use the name Moral Mirror, the tool has to refuse at least four things. Closing a user into a fixed type. Showing the majority response as if it were the answer. Quietly placing one position as the superior endpoint. Using a user’s wavering as material for retention or payment. Fail to refuse these four and the name is a shell.

The phrase intellectual self-flagellation

Somewhere in this sits the phrase “intellectual self-flagellation zone.” Honestly I still do not know how to handle it. It sounds like a joke, and it also sounds like it strangely hits the core. Deliberately walking into the spot where your own certainty turns uncomfortable. Not to believe what I think is right more strongly, but to walk in and shake why I believe it. Seen that way, it is not entirely wrong.

But putting it out front is risky. Self-flagellation is too strong a word, and it makes people misread what this project is doing. The goal is not to torment anyone, nor to keep them held in discomfort. It is to let them see the soft spot in their thinking. Discomfort may arise along the way, but discomfort itself is not the point.

The difference matters. Many digital products take enjoyment as their metric. In reverse, some serious products treat discomfort as proof of sincerity. Both are dangerous. Pure enjoyment drifts to certainty reinforcement, pure discomfort drifts to self-torture. The spot Moral Mirror has to find is somewhere between. Uncomfortable enough that the cracks in your thinking show, but not so much that the discomfort breaks the user or herds them. That is also why I still cannot put this phrase at the center. It is a strong line, and a strong line is not a good name.

Starting with a temporary name

So the conclusion for now is a little flat. I still do not fully love Moral Mirror. The intellectual self-flagellation zone I am still unsure about. “AI-native Socratic education for domains without fixed answers” is good as a description but long as a name, “interactive edu tool” is too wide, and “philosophy chatbot” too narrow.

Still, naming cannot be put off. With no name, the thinking scatters every time. Every explanation has to redraw the coordinates, and whoever hears them does not know where to hang anything. A name is needed not only by others but by the maker. Only with a name can you judge which feature must not come in, which sentence wrecks the tone, which revenue model eats the mission.

So for now I start with a temporary name. Moral Mirror. Not because it is perfect, but because it is careful enough. The name gives at least a few signals. Do not diagnose the user. Do not hold them. Do not play along with their words and dress them up nicer. And above all, it keeps reminding me that this tool’s job is not to give an answer but to reflect.

A good name is not finished in one go. The name of a category that does not exist yet, even less so. The first name is always a little loose, a little uncomfortable, a little like someone else’s words. What matters is whether that name can govern the work. The Moral Mirror of now at least hands questions back to me. Is this feature a mirror or a megaphone? Does this sentence reflect or herd? Does this output leave the user open or close them off? If it keeps making me ask that, a temporary name is enough.

What this chapter did not answer

Chapter 5 wrote down the process of naming a blank spot. At first it was just an interactive education tool, the subject’s lack of a right answer made the name heavier, and passing through the LLM’s tendency to agree and recommendation systems’ certainty-reinforcing machinery, the name became a safeguard rather than a label. Moral Mirror is not one I deliberately made, nor one whose sound I fully love, but it at least keeps me asking what this tool must not do.

Still, there is what it did not answer. One, whether this name reads the same way to actual users I do not know. Even if I call it a mirror, a user can read it as a personality test, a philosophy game, an AI counselor. A name carries intent but does not control interpretation.

Two, that a name can be a safeguard is still a hypothesis. The real safeguard comes not from the name but from design and operation. Which questions it asks, which counterexamples it forbids, which output sentences it blocks, which logs it keeps. The name can set the direction but cannot stand in for the design. For this project to truly become Moral Mirror, structure has to come after the name. From the next chapter, I move to why I borrowed a roundtable to test that structure.