Can You Prove That Ice Cream Is Better Than Broccoli?
A Kid, a Freezer, and a Tough Choice
You’re at the ice cream shop with your friends. Everyone shouts orders — double fudge, mint chip, mango sorbet — but you freeze up. You like vanilla, but chocolate is amazing too. And what if the strawberry swirl suddenly tastes better today? You finally mumble, “I guess I prefer vanilla right now.” What just happened inside your head? Can anyone — even a mathematician — write down a rule that explains why one thing feels better than another?
That’s the question at the heart of preference logic, a branch of philosophy and computer science that tries to turn gut feelings like “I like this more” into a careful language with its own grammar. It started in the 1950s with philosophers who wanted to decode the word “better.” Today it lives inside the apps that recommend your next movie and the video games that predict who you’ll pick as an ally. And it all begins with a very simple idea: whenever you compare two things, you’re using a preference relation — a hidden ranking that can be studied, tested, and, sometimes, even proved to be wrong.
Is It Just Because You Like It, or Is There a Reason?
In 1963, the Finnish philosopher Georg Henrik von Wright (1916–2003) made a distinction that still shapes the whole field. He noticed that sometimes you prefer something simply because you like it — that’s an intrinsic preference. You love the taste of mango, so mango sorbet wins, no explanation needed. But other times your preference comes with a receipt: you pick the mango sorbet because it’s half the price of the chocolate fudge, or because you read that mangoes are healthier. That’s an extrinsic preference — a preference backed by an outside reason.
Von Wright’s split matters because it changes how you can argue about what’s “better.” If your friend prefers chocolate purely for the taste, you can’t really debate that — there’s no external fact to poke at. But if she claims chocolate is better because it’s more filling, you could bring up evidence that mango sorbet has more fiber. Once reasons enter the game, preferences become like little arguments, and that means they can be supported, attacked, or even proven inconsistent.
The Mathematician Who Wrote Rules for “Better Than”
Before von Wright sorted his types of preference, a Swedish logician named Sören Halldén (1916–1999) published a book called On the Logic of “Better” in 1957. He treated the word “better” the way mathematicians treat the symbol “>” — by laying down a handful of sharp rules that any sensible ranking should obey.
The first rule is asymmetry: if you claim that A is better than B, you can’t also claim that B is better than A in the exact same situation. It’s like saying in one breath “soccer is more fun than basketball” and “basketball is more fun than soccer” — you’re stuck in a loop. The second rule is transitivity: if A is better than B and B is better than C, then A must be better than C. If you like vanilla over chocolate and chocolate over strawberry, logic forces you to rank vanilla above strawberry. If you don’t, your preferences contain a hidden contradiction, like a circle in a board game that never ends.
Halldén also introduced a subtle “expansion” principle. Imagine you can only win one prize at the carnival. Would you rather win a giant panda (keep panda, lose unicorn) or a giant unicorn (keep unicorn, lose panda)? If you pick panda, the formula says you truly prefer panda over unicorn, regardless of the other stuffed animals on the shelf. This trick lets logicians strip away all the extra noise and compare two options side by side, as if they were the only things in the universe.
Imagining All the Worlds That Could Be
Once you have rules for “better than,” you need a way to picture what those rules refer to. That’s where possible world semantics comes in. Think of every imaginable situation — winning the lottery, eating a soggy sandwich, acing a test — as a separate little world. A preference order arranges these worlds from least to most desirable. When you say “I prefer playing video games to doing homework,” you’re really declaring that the world where you play games sits higher on your ranking than the world where you crack open a textbook.
This picture gives the logic its teeth. Suppose you discover that your ranking of worlds contains a loop: you prefer the game-world to the homework-world, the homework-world to a bored-world, but then the bored-world to the game-world. The rules of preference logic call that out as a cyclic order — a sure sign that something has gone wrong. It’s like a video game character who, if they walk far enough in one direction, ends up back where they started without ever turning around.
But there’s a catch that von Wright already noticed. When you claim “video games beat homework,” you almost never mean “video games beat homework in every possible circumstance.” You mean all else being equal. You’d probably prefer homework if the homework was a fun puzzle and the only video game was a broken demo. Logicians call this condition ceteris paribus — Latin for “other things being equal.” To compare fairly, you have to freeze every other fact — same amount of time, same weather, same snack — and change only the one thing you care about. This tiny phrase turned out to be the key that unlocked machines that can predict what you want.
Teaching Computers to Guess What You Like
When you scroll through a streaming service and it offers “Because you liked Spider-Ham,” you’re seeing a cousin of ceteris paribus in action. Computer scientists in the 1990s, including Craig Boutilier (1963–2018) and his colleagues, invented something called a conditional preference network, or CP-net for short. A CP-net is a bicycle wheel of simple rules: “I prefer a light laptop to a heavy one, all else equal,” “I prefer a fast laptop to a slow one, if the weight is the same,” and “I prefer a long battery life, all else equal.”
The machine stores each rule in a little table and links them with arrows that show which features depend on which others. If you then feed it two laptops — one light but slow, the other heavy but fast — the CP-net uses ceteris paribus to keep everything else fixed and calculates which one you’d pick. It doesn’t need to compare every possible combination; it only needs the handful of “if this, then that” preferences you gave it. That means recommendation systems can be both fast and surprisingly accurate, even when they don’t know every detail about you.
The trick works because the same logic that Halldén and von Wright wrote down with pen and paper can be carried out by a computer, silently checking for loops and asking the fewest possible questions. Next time a game suggests a character or a website recommends a song, there’s a fair chance it’s running a miniature preference logic under the hood, built out of the simple ideas of “better than,” “all else equal,” and conditional rules.
Why This Matters When You’re Staring at a Menu
You don’t need a supercomputer to use this logic. The next time you’re frozen in front of an ice cream counter, a jumble of reasons and raw tastes churning inside you, try thinking like a CP-net. Ask yourself: is there a flavor I’d prefer no matter what the other flavors are? That’s your top priority. Then imagine changing just one thing — say, switching from cone to cup — and ask if that flips the whole ranking. If it does, you’ve found a condition that matters, and you can label it and make it explicit.
Von Wright’s old insight helps too. If you can name an outside reason for your pick (“I want the cheapest scoop”), it’s an extrinsic preference, and you can bargain with your friends — maybe they’ll chip in for the chocolate fudge. If no reason exists except “I just like it,” you’re in the land of intrinsic preference, and that’s fine too. No one owes a proof for loving mango.
Preference logic doesn’t tell you what you should want. It gives you a language to clean up your rankings and spot the knots. And it reminds you that comparing — whether you’re choosing a dessert, a movie, or a friend — is never just a random flip of a coin. It’s a structure, and once you see the structure, you can start to understand why you pick what you pick.
Think about it
- Imagine you’re designing a robot that buys groceries for your family. What three simple rules would you program into its CP-net to make sure it picks food that everyone will actually eat? Would those rules ever conflict?
- Can you think of a situation where your “gut” preference and your “reasoned” preference pull in opposite directions? How would you decide which one to follow — and could someone else ever prove you made the wrong choice?
- If a new app could predict all your shopping choices perfectly based on past preferences, would you still feel like you were making a free decision? Why or why not?
