Why Renaming Giorgione Turns a True Sentence False

Same Person, Different Name — Why Does the Sentence Turn False?

There was a painter in 16th-century Venice named Giorgione. His actual name was Giorgio Barbarelli. Giorgione wasn’t his name — it was a nickname, meaning “Big Giorgio.” He got it because of his large build.

So the following sentences hold:

• Giorgione was called that because of his build. → True
• Barbarelli was called that because of his build. → False

It’s the same person. Yet swap out one name, and true becomes false. Something clearly feels off here, but it’s not easy to pin down exactly where the problem lies.

Giorgione’s Castelfranco Madonna, painted in 1505

In 1971, MIT philosopher Richard Cartwright supplied the answer. Identity — the fact that two things are the same — and substitutability — the license to swap one for the other in a sentence — are different principles, and the first doesn’t guarantee the second. I think this 55-year-old argument describes, with striking precision, the structure of the substitution debates we’re all having every day now.

We Thought It Was One Principle. It Was Actually Two.

There’s something usually called Leibniz’s Law¹. But under that single name, two different statements have been tangled together.

First, the principle of identity. If x and y are the same thing, then whatever property x has, y has too. This is a claim about the world.

Second, the principle of substitutivity. If x and y are identical, then substituting one for the other in any sentence should never change that sentence’s truth value. This is a claim about language.

Most people treat these as the same claim, because they assume that language is simply a transcription of the world. But the Giorgione sentence shows that the second principle is just wrong. A single counterexample is enough to break a principle, and we just saw one.

So does the first principle fall too? For that, there would need to be some property that Giorgione actually has and Barbarelli actually lacks. What would that be?

Is it the property of “being called that because of his size”? This is where it gets stuck. What “that” refers to isn’t fixed outside the sentence. Whether it means “called Giorgione” or “called Barbarelli” depends entirely on which name has been plugged into the sentence. With nothing supplied beforehand, “that” points at nothing at all.

This is exactly the point Cartwright identified. The problem was never the person — it was the sentence. That sentence wasn’t only talking about a single individual. It was simultaneously talking about the name used to refer to that person. A name wasn’t a label you could freely attach or detach from an object; it was a component that determined whether the sentence was true or false.

So when substitution breaks down, it doesn’t mean the two people become different people. It means that the sentence’s slot wasn’t simply a slot for a person — it was a slot that used the name for that person as well.

The signature line on a contract works the same way.

• The contract was signed by the CEO.
• The contract was signed by Kim Min-su.

Even if Kim Min-su is that company’s CEO, the two sentences aren’t the same sentence. The first signs in the capacity of company representative; the second signs in a personal capacity. This is exactly why legal teams are so sensitive about how a title is written on a signature line. Even when the person is identical, which name occupies that slot changes what the signature legally does.

Pluto’s reclassification confirmed the “number of planets” argument all over again

Cartwright’s second example is even more interesting.

• 9 is necessarily greater than 7. → True
• The number of planets is necessarily greater than 7. → False

As of 1971, when the paper was written, there were 9 planets. 9 and the number of planets were the same thing. And yet only one of these statements is true.

The reason lies in the word “necessarily”². That 9 is greater than 7 is true no matter how the world happened to turn out. But how many planets there are is just a contingent fact. If the solar system had formed a little differently, there might have been 6.

Cartwright’s diagnosis is this: the second sentence actually admits two readings.

On the first reading, it means that the sentence “the number of planets is greater than 7” is itself necessarily true. That’s false. On the second reading, it means that the number of planets is fixed at some particular value, and that specific value is necessarily greater than 7. Since that value is 9, this reading is true.

The sentence was ambiguous from the start. No principle was broken — the sentence simply never made clear which of the two meanings it carried. In philosophy, this distinction is called de re and de dicto³.

What I found fascinating rereading the paper this time is this: Cartwright mentions, almost in passing, that 9 being the number of planets is simply a fact astronomy happens to supply. She uses it only as background for the argument.

Then, 35 years later, in 2006, the International Astronomical Union reclassified Pluto as a dwarf planet, and the count dropped to 8. The premise the paper had rested on turned out to be false — outside the paper itself.

But this doesn’t undermine the argument. It completes it. Cartwright’s whole point was that the phrase “the number of planets” was never meant to point to a specific value — it was meant to point to a slot, and whatever value fills that slot can always change. What looked like a philosophical argument was, in effect, confirmed a second time by an actual change in astronomy.

I see this same distinction in company documents every day. Take the goal “retain our #1 client.” Whether “#1 client” means the specific company that’s #1 right now, or whoever happens to hold the #1 spot at any given time, completely changes the strategy. In the first case, you should assign a dedicated team to that specific company; in the second, you should reallocate resources every time the ranking shifts. But the goal statement itself doesn’t tell you which one it means. “Retain our core talent” has the exact same problem.

Performance parity alone doesn’t settle the substitution debate

Lately, the substitution debate runs on almost a single question: can AI do it as well as a person can? Benchmark scores, accuracy rates, pass rates, percentage-of-human performance. And beneath that sits an unspoken premise — that once performance is equal, swapping one for the other is fine.

This is exactly the premise Cartwright dismantled 55 years ago. Even if sameness holds true, permission to substitute doesn’t follow from it.

Accepting this changes the diagnosis. When we try to substitute and it doesn’t work out, we usually conclude that performance still isn’t good enough and wait for the next model. But there’s another possibility: that the role was never one where only the output mattered, from the start.

So which roles require a name attached, not just a result? In my own practice, I’ve come to sort this by four criteria.

• Roles where accountability is recorded. Audit opinions, medical judgments, legal counsel, accounting sign-offs — work where who did it has to remain traceable, separate from the outcome itself.
• Roles where the relationship is part of the output. Long-term negotiations or counseling, where the fact that it’s this specific person is part of why the other side engages — that’s baked into the result.
• Roles where a person is meant to grow. This includes the repetitive tasks junior staff used to handle. Remove them today and nothing breaks today — but 5 years from now, no senior will have been made.
• Roles where someone must be able to explain the reasoning later. Work that requires a person who can, after the fact, account for why a judgment was made.

For work that doesn’t meet any of these four conditions, substitution can happen even without flawless performance. Conversely, work that requires these conditions can’t be substituted just by matching performance. This is why, when judging whether substitution is possible, you have to look at a job’s accountability structure and qualification requirements together — not just at performance.

This distinction is already institutionalized across many professions. A diagnosis carries the license number of the physician who made it. An audit report is signed by the accountant in charge. A legal opinion bears the name of the attorney who wrote it. Why? Because someone has to be answerable when the judgment turns out wrong. Here, the name isn’t a label attached to the output — it’s a condition for the output to have any force at all.

That’s why, in these roles, the structure doesn’t easily change no matter how good the tools get. What actually happens isn’t substitution but reassignment: the machine writes the draft, the human signs it. On the surface, it looks like the person is still there — but what that person does has shifted from making to checking. And this shift rarely shows up on an org chart or in a performance metric. I’d argue a good deal of the confusion many organizations are experiencing right now comes from exactly this. While everyone reassures themselves that “substitution hasn’t happened,” the content of the work has quietly changed underneath them.

We should be equally honest about the other side of this. Many of the roles we insist “need a human name attached” may be nothing more than untested inertia. I’ve seen it repeatedly — claims that a meeting must be in person, that it must be handled by someone with seniority, that a human must give final sign-off, repeated with no actual grounding. Cartwright, too, did two things at once: he acknowledged genuine counterexamples to the identity-of-indiscernibles principle, while dismantling the fake ones one by one. That’s precisely the work needed now — sorting out roles that are genuinely irreplaceable from roles that are simply familiar.

Oswarld’s Lens

Helping companies evaluate AI adoption means reading a lot of proposals and internal review documents. There’s one table that shows up in almost every single one of them: a performance comparison table. Accuracy, processing time, cost per case, percentage versus human performance.

And almost every time, there’s one cell left blank. Whose name does this output go out under.

I don’t think that blank cell is a simple documentation oversight. It means the organization understands that spot as a place where only the output emerges. So when the results after adoption fall short of expectations, the first thing said in the meeting room is usually some version of “the model still seems weak.”

That’s half right. The other half is wrong. If that spot was originally a place where a name was attached too, then even if the model gets twice as good, the same problem recurs in the same spot. This isn’t a problem you solve by raising performance — it’s a problem you solve by redesigning where responsibility and sign-off sit.

So lately, at the start of any review, I’ve started asking a different question. Instead of asking whether this can be replaced by AI, I first ask whose name is attached to this task’s output, and why that name is necessary. If the answer comes immediately, it’s a spot with fairly high replaceability. If the answer stalls out, or ends with “that’s just how we’ve always done it,” that’s less a signal about replacement and more a signal that it’s time to reexamine the spot itself.

For reference, I put out a weekly newsletter that cross-analyzes technology, economics, and the humanities. Work like today’s — taking an old paper and applying it to a present-day problem — is something I do often.

Closing

Let me sum up.

First, “being identical” and “being interchangeable” are two different principles. Giorgione and Barbarelli — the Renaissance painter Giorgio Barbarelli, better known by his nickname Giorgione — are the same person, yet you can’t swap one name for the other within a sentence.

Second, substitution breaks down not because the referent changes, but because the sentence is doing more than referring to the object — it’s also invoking the name itself.

Third, the real test for replaceability isn’t “what percentage of the performance can it match,” but whether the slot in question only needs the output, or whether it needs the name attached to it too.

I’d like you to try just one thing this week. Pick a task you’re currently responsible for, and write two lines: whose name gets attached to the result, and why that name is necessary. If those two lines come easily, a substantial part of that work can probably already be handed off. If you can’t write them, it’s likely not because the work is irreplaceable — it’s more likely that you’ve simply never sat down to work it out.

Reader, is there a part of your job where, even if the output came out identical, it still has to carry your own name? Tell me in the comments what that task is and why you think that’s the case.

---

💬 If there’s work where the output is the same but it still needs to go out under your own name, tell us in the comments — we’ll factor it into the next issue. 📨 If you have a colleague reviewing AI adoption, please forward this piece to them.

References & Further Reading

Primary sources

• Cartwright, Richard, “Identity and Substitutivity”, in Milton K. Munitz (ed.), Identity and Individuation, New York University Press, 1971, pp. 119-133. Link ··· This is the backbone of today’s piece. The Giorgione argument runs pp. 122-125, and the number-of-planets argument runs pp. 126-131. If the symbolic notation feels like a lot, just reading three pages starting at 122 will give you the core of it.
• Kripke, Saul, “Identity and Necessity”, in the same volume, pp. 135-164. ··· This is the chapter right after Cartwright’s. It tackles the same problem from the opposite direction. Reading the two side by side, you can really feel the temperature of this debate in the early 1970s.
• Quine, W. V. O., From a Logical Point of View, 2nd ed., Harper and Row, 1963, p. 139. ··· The standard statement of the substitution principle that Cartwright takes issue with appears on this page. It was Quine, too, who first pulled the Giorgione example into the center of the debate.

Background

• Frege, Gottlob, “Über Sinn und Bedeutung”, Zeitschrift für Philosophie und philosophische Kritik, vol. 100, 1892, pp. 25-50. ··· This is where the debate starts: why different names for the same object can produce different sentences. The morning-star-and-evening-star example comes from here.
• MIT Department of Philosophy, “Richard Cartwright”. Link ··· This is the author’s complete list of works. If you liked this issue, his 1979 paper “Indiscernibility Principles” is a good next step.

Past issues worth reading alongside this one

• Issue 161: AI models can be copied, but your judgment is yours to make
• Issue 173: The Luddites never hated the machine — they resisted the terms of its introduction

---

The author is Oswarld (Kwangseob Ahn). Current roles: Adjunct Professor at Sejong University, Strategy Consultant at INLEVEL9. Career, research, books, and recent work are kept current on the About page. Latest · July 2026: HEMA-2: A Consolidation-Aware Tri-Memory Architecture with Multi-Channel Scheduling for Lifelong Conversational AI.

📝 Glossary

Footnotes

1. Leibniz’s Law: The principle that identical things are indiscernible. If two objects are truly one and the same, there can be no property that only one of them has — and the crux of today’s piece is that the version of this principle applied to words and the version applied to the world turn out to be different. ↩

2. Necessarily: This means something would have had to be true no matter how the world had turned out — which is different from simply being true right now. It’s true that it’s clear in Seoul today, but that’s not necessary; that 3 plus 4 equals 7, on the other hand, is necessary. ↩

3. De re and de dicto: The distinction between reading a sentence as being about the object itself, versus about the sentence as such. Read “the number of planets is necessarily greater than 7” as a claim about that number’s own property, and it’s true; read it as claiming that the sentence itself must be true, and it’s false. ↩