Tag Archives: independence friendly logic

Punny Logic

Update 12 Feb: This post had been expanded upon and, after submission, accepted for publication in Analysis published by Oxford University Press. View the final version here.


It is hard to explain puns to kleptomaniacs because they take things literally.

On the surface, this statement is a statement of logic, with a premise and conclusion.

Given the premise:

Kleptomaniacs take things literally.

We may deduce the conclusion:

It is hard to explain puns to kleptomaniacs.

Now, whether the conclusion strictly follows from the premise is beside the point: it is a pun, and meant to be funny. However, as a pun, it still has to make some logical sense. If it didn’t make any sense, it wouldn’t, and couldn’t, be funny either. While nonsense can be amusing, it isn’t punny.

What is the sense in which the conclusion logically follows from the premise then, and how does this relate to the pun?

Puns play off ambiguity in the meaning of a word or phrase. In this case the ambiguity has to do with the meaning of to take things literally. It can mean to steal, or it can mean to only use the simplest, most common definitions of terms.

In the first meaning, by definition, kleptomaniacs steal, i.e. they literally take things.

So then “take things literally” is true.

In the second meaning, by deduction, since puns play off multiple meanings of things, it is hard to explain a pun to someone who only uses the single, most common definition of a term. That is, if they take things literally, they won’t recognize the multiple meanings required to understand a pun.

So if someone “takes things literally” it is true that it is hard to explain puns to them.

Therefore, between the two meanings, we can informally derive the statement: it is hard to explain puns to kleptomaniacs because they take things literally.

However, if we wanted to write this out in a formal logical language, then we would need a formal way to represent the two meanings of the single phrase.

Classically, there is no way to give a proposition multiple meanings. Whatever a proposition is defined as, it stays that way. A can’t be defined as B and then not defined as B: (A=B & A≠B) is a contradiction and to be avoided classically. But let’s start with a classical formulation:


TTL1 mean to Take Things Literally, in the 1st sense: to steal

TTL2 mean to Take Things Literally, in the 2nd sense: to use the most common definitions of terms.


  1. ∀x [ Kx → TTL1x ]
    For anyone who is a Kleptomaniac, Then they take things literally (steal)
  2. ∀y[ TTL2y → Py ]
    For anyone who takes things literally (definitionally), Then it is hard to explain puns to them

What we want, however, is closer to:

  1. ∀z [[ Kz → TTLz] → Pz ]
    For anyone who is a Kleptomaniac, Then they take things literally, Then it is hard to explain puns to them

with only one sense of TTL, but two meanings.

Since TTL1 ≠ TTL2, we can’t derive (3) from (1) and (2), as is. And if TTL1 = TTL2, then we would have (1) A→B, and (2) B→C, while trying to prove (3) A→B→C, which logically follows. However, there would no longer be a pun if there was only one meaning of TTL.

What is needed is to be able to recompose our understanding of ‘to take things literally’ in a situation aware way. We need to be able to have the right meaning of TTL apply at the right time, specifically Meaning 1 in the first part, and the Meaning 2 in the latter.

Intuitively, we want something like this, with the scope corresponding to the situation:

  1. ∀z [ Kz → { TTLz ]1 → Pz }2

In this formula, let the square brackets [] have the first meaning of TTL apply, while the curly braces {} use the second meaning. Only the middle — TTL — does double duty with both meanings.

Achieving this customized scope can be done by using Independence Friendly logic. IF logic allows for fine-grained scope allocation.

So let:

S mean to steal.

D mean to take things definitionally.


  1. ∀x ∀y ∃u/∀x ∃v/∀y [ Kx → ( x=u & y=v & Su & Dv → TTLvu ) → Py ]
    If anyone is a kleptomaniac then there is someone who is identical to them who steals… and if there is someone who takes things definitionally then there is someone identical to them for whom it is hard to explain puns to… and the person who steals and the person who takes things definitionally then both Take Things Literally.

The scope gymnastics are being performed by the slash operators at the start and the equality symbols in the middle part of the equation. What they are doing is specifying the correct meanings — the correct dependencies — to go with the correct senses: Stealing pairs with Kleptomania and taking things Definitionally pairs with being bad at Puns, while both pairs also meaning Taking Things Literally. With both pairs meaning TTL, and each pair being composed independently, Equation (5) therefore provides a formalization of the original pun.


Finding new applications for existing logical systems provides a foundation for further research. As we expand the range of topics subject to logical analysis, cross-pollination between these subjects becomes possible.

For instance, using custom dependencies to associate multiple meanings to a single term is not only useful in describing puns. Scientific entities are often the subjects of competing hypotheses. The different hypotheses give different meanings — different properties, relations and dependencies — to the scientific objects under study. Logically parsing how the different hypotheses explain the world using the same terms can help us analyze the contradictions and incommeasureabilities between theories.

On the other hand, while this article may have forever ruined the above pun for you (and me), it does potentially give insight into what humans find funny. Classically, risibility, having the ability to laugh, has been associated with humans and rationality. Analyzing this philosophical tradition with the new logical techniques will hopefully provide existential insight into the human condition.

Posted in independence friendly logic, logic. Tagged with , , , .

Shaking the Tree

Life often results in situations such that no strategy suggests any further moves. We just don’t know what to do next. In a game of perfect information, where each player knows all the previous moves, this can signal stalemate. Take chess: given both sides know everything that has transpired and have no reason to believe that the opponent will make a mistake, there can come a time when both sides will realize that there are no winning strategies for either player. A draw is then agreed upon.

The situation is not as simple in games of incomplete information. Let’s assume some information is private, that some moves in the game are only known to a limited number of players. For instance, imagine you take over a game of chess in the middle of a match. The previous moves would be known to your opponent and the absent player, but not to you. Hence you do not know the strategies used to arrive at that point in the game, and **your opponent knows that you do not know**.

Assume we are in a some such situation where we do not know all the previous moves and have no further strategic moves to make. This is to say we are waiting, idling, or otherwise biding our time until something of significance happens. Formally we are at an equilibrium.

A strategy to get out of this equilibrium is to “shake the tree” to see what “falls out”. This involves making information public that was thought to be private. For instance, say you knew a damaging secret to someone in power and that person thought they had successfully hidden said secret. By making that person believe that the secret was public knowledge, this could cause them to act in a way they would not otherwise, breaking the equilibrium.

How, though, to represent this formally? The move made in shaking the tree is to make information public that was believed to be private. To represent this in logic we need a mechanism that represents public and private information. I will use the forward slash notation of Independence Friendly Logic, /, to mean ‘depends upon’ and the back slash, , to mean ‘independent of.’

To represent private strategy Q, based on secret S, and not public to party Z we can say:

Secret Strategy) If, and only if, no one other than Y depends upon the Secret, then use Strategy Q
(∀YS) (∃z/S) ~(Y = z) ⇔ Q

To initial ‘shaking the tree’ would be to introduce a new dependency:

Tree Shaking) there is someone other than Y that depends on S
(∃zS) ~(Y = z)

Tree Shaking causes party Y’s to change away from Strategy Q since Strategy Q was predicated upon no one other than Y knowing the secret, S. The change in strategy means that the players are no longer idling in equilibrium, which is the goal of shaking the tree.

Posted in game theory, independence friendly logic, logic, philosophy. Tagged with , , .

Яandom Logic

If we try to represent tossing a coin or a die, or picking a card out of a deck at random, in logic, how should we do it?

Tossing a coin might look like:

Toss(coin) → (Heads or Tails)

Tossing a die might be:

Toss(die) → (1 or 2 or 3 or 4 or 5 or 6)

Picking a card:

Pick(52 card deck) → (1♣ or 2♣ or … or k♥)

This begs asking, do these statements make sense? For instance look what happens if we try to abstract:

∀x Toss(x)

such that ‘Toss’ represents a random selection of the given object.

But this is weird because Toss is a randomized function and x is not selected randomly in this formula. Perhaps if we added another variable, we could generate the right sort of function:

∀y ∃x Toss(yx)

Then x would be a function of y: we would select x with respect to y. The problem is still that a Toss involves randomness. So this setup is incorrect because treating x as a function of y is not randomized, because y is not random.

How can we represent randomness in logic?

As noted, functions alone will not work. Variables and interpreted objects cannot invoke randomness. Perhaps we can modify some part of our logic to accommodate randomness. The connectives for negation and conjunction haven’t anything to do with randomness either.

But, if we use the game theoretic interpretation of logic, then we can conceive of each quantifier as representing a player in a game. Players can be thought of as acting irrationally or randomly.

Therefore, let’s introduce a new quantifier: Я. Я is like the other quantifiers in that it instantiates a variable.

  1. Яx T(x)
  2. Tb

However, Я is out of our (or anyone’s) control. It does instantiate variables when it is it’s turn (just like other quantifiers) but it instantiates randomly. So we have three players, Abelard, Eloise and Random (or the Verifier, Falsifier and Randomizer).

But more is still needed. We need a random selection between specific options, be it between heads and tails, 1-6, cards, numbers, or anything else. One way of doing this would be to create a special domain just for the random choices. Я would only instantiate from this domain, and if there are multiple random selections, we will require multiple indexed domains.

Hence, given Di(Heads, Tails),
represents a coin flip since Я randomly instantiates out of the domain containing only Heads and Tails.

I prefer to use an artifact of Independence Friendly logic, the dependence indicator: a forward slash, /. The dependence indicator means that the quantifier only depends on those objects, variables, quantifiers or formulas specified. Hence

Яx/(Heads, Tails)

means that the variable x is randomly instantiated to Heads or Tails, since the only things that Яx is logically aware of are Heads and Tails. Therefore this too represents a coin flip, without having multiple domains.)

Now that we have an instantiation rule for Я we also need a negation rule for it. If some object is not selected at random, then it must have been individually selected. In this case the only other players that could have selected the object are ∀ and ∃. Hence the negation rule for Я is just like the negation rule for the other quantifiers: negating a quantifier means that a different player is responsible for instantiation of the variable. If neither player is responsible, it can be considered random: ¬Яx ↔ (∀x or ∃x). We can leave the basic negation rule for ∀ and ∃ the way it is.

Therefore, given the additions of the new quantifier and domain (or slash notation), we can represent randomness within logic.


See “Propositional Logics for Three” by Tulenheimo and Venema in Dialogues, Logics And Other Strange Things by Cedric Degremont (Editor) College Publications 2008, for a generalized framework for logics with 3 quantifiers. Since the above logic requires either indexed domains or dependence operators, Яandom Logic is a bit different, but it is a good discussion.

Posted in game theory, logic, science. Tagged with , , , , .

IF Logic and Cogito Ergo Sum

(∃x∃x) → ∃x

Descartes Law

If something has informational dependence upon itself, then that thing exists.  For example, thinking that you are thinking is informationally self dependent and therefore a thinking thing (you) exists.

Posted in epistemology, independence friendly logic, logic. Tagged with , .

Rock Paper Scissors

Rock Paper Scissors is a game in which 2 players each choose one of three options: either rock, paper or scissors.  Then the players simultaneously reveal their choices.  Rock beats scissors but loses to paper (rock smashes scissors); Paper beats rock and loses to scissors (paper covers rock); Scissors beats paper but loses to rock (scissors cut paper).  This cyclical payoff scheme (Rock > Scissors, Scissors > Paper, Paper > Rock) can be represented by this rubric:

Child 2
rock paper scissors
Child 1 rock 0,0 -1,1 1,-1
paper 1,-1 0,0 -1,1
scissors -1,1 1,-1 0,0
(ref: Shor, Mikhael, “Rock Paper Scissors,” Dictionary of Game Theory Terms, Game Theory .net,  <http://www.gametheory.net/dictionary/Games/RockPaperScissors.html>  Web accessed: 22 September 2010)

However, if we want to describe the game of Rock Paper Scissors – not just the payoff scheme – how are we to do it?

Ordinary logics have no mechanism for representing simultaneous play.  Therefore Rock Paper Scissors is problematic because there is no way to codify the simultaneous revelation of the players’ choices.

However, let’s treat the simultaneous revelation of the players’ choices as a device to prevent one player from knowing the choice of the other.  If one player were to know the choice of the other, then that player would always have a winning strategy by selecting the option that beats the opponent’s selection.  For example, if Player 1 knew (with absolute certainty) that Player 2 was going to play rock, then Player 1 would play paper, and similarly for the other options.  Since certain knowledge of the opponent’s play trivializes and ruins the game, it is this knowledge that must be prevented.

Knowledge – or lack thereof – of moves can be represented within certain logics.  Ordinarily all previous moves within logic are known, but if we declare certain moves to be independent from others, then those moves can be treated as unknown.  This can be done in Independence Friendly Logic, which allows for explicit dependence relations to be stated.

So, let’s assume our 2 players, Abelard (∀) and Eloise (∃) each decide which of the three options he or she will play out of the Domain {r, p, s} .  These decisions are made without knowledge of what the other has chosen, i.e. independently of each other.

∀x ∃y/∀x

This means that Abelard chooses a value for x first and then Eloise chooses a value for y.  The /∀x next to y means that the choice of y is made independently from, without knowledge of the value of, x.

R-P-S: ∀x ∃y/∀x (Vxy)

The decisions are then evaluated according to V, which is some encoding of the above rubric like this:

V: x=y → R-P-S &
x=r & y=s → T &
x=r & y=p → F &
x=p & y=r → T &
x=p & y=s → F &
x=s & y=p → T &
x=s & y=r → F

T means Abelard wins; F means Eloise wins.  R-P-S means play more Rock Paper Scissors!

Johan van Benthem, Sujata Ghosh and Fenrong Liu put together a sophisticated and generalized logic for concurrent action:

Posted in game theory, independence friendly logic, logic, philosophy. Tagged with , , , .

The Non-Reducibility & Scientific Explanation Problem

Q: What is a multiple star system?

A: More than one star in a non-reducible mutual relationship spinning around each other.

Q: How did it begin?

A: Well, I guess, the stars were out in space and at some point they became close in proximity.  Then their gravitations caused each other to alter their course and become intertwined.

Q: How did the gravitations cause the courses of the stars to become intertwined?  Gravity does one thing: it changes the shape of space-time; it does not intertwine things.

A: That seems right.  It is not only the gravities that cause this to happen.  It is both the trajectory and mass (gravity) of the stars in relation to each other that caused them to form a multiple star system.

Q: Saying that it is both the trajectories and the masses in relation to each other is not an answer.  That is what is in need of being explained.

A: You are asking the impossible.  I have already said that the relation is non-reducible.  I am not going to go back upon my word in order to reduce the relation into some other relation to explain it to you.  The best that can be done is to describe it as best we can.

Here is the problem: If you have a non-reducible relation (e.g., a 3-body problem or a logical mutual interdependence) then you cannot explain how it came to exist.  Explaining such things would mean that the relation was reducible.  But being unable to explain some scientific phenomenon violates the principle of science: we should be able to explain physical phenomenon.  Then the relation must not be non-reducible or it must have been a preexisting condition going all the way back to the origin of the universe.  Either you have a contradiction or it is unexplainable by definition.

What can we do?  You can hold out for a solution to the 3-body-problem or, alternatively, you can change what counts as explanation.  The latter option is the way to go, though, I am not going into this now.

For now I just want to illustrate that this problem of non-reducibility and explanation is pervasive:

Q: What is a biological symbiotic relationship?

A: More than one organism living in a non-reducible relationship together.

Q: How did it begin?

A: Well, I guess, the organisms were out in nature and at some point they became close in proximity.  Then their features caused each other to alter their evolution and become intertwined.

Q: How did the features cause the courses of their evolution to become intertwined?  Physical features do one thing: they enable an organism to reproduce; they do not intertwine things.

A: That seems right.  It is not only the features that cause this to happen.  It is both the ecosystem and the features of the organisms in relation to each other that caused them to form a symbiosis.

Q: Saying that it is both the place the organisms are living in and their features in relation to each other is not an answer.  That is what is in need of being explained.

A: You are asking the impossible.  I have already said that the relation is non-reducible.  I am not going to go back upon my word in order to reduce the relation into some other relation to explain it to you.  The best that can be done is to describe it as best we can.

As you can see, I am drawing a parallel between a multiple body problem and multiple organisms that live together.  Like the star example above, there is no way to explain the origins of organisms living together.  Even in the most basic case it is impossible.

Posted in biology, epistemology, evolution, independence friendly logic, ontology, philosophy, physics, science. Tagged with , , , , , .

What are Quantifiers?

What are quantifiers?  Quantifiers have been thought of things that ‘range over’ a set of objects.  For example, if I say

There are people with blue eyes

this statement can be represented as (with the domain restricted to people):


This statement says that there is at least one person with property B, blue eyes. So the ‘Ex’ is doing the work of looking at the people in the domain (all people) and picking out one with blue eyes.  Without this ‘∃x’ we would just have Bx, or x has blue eyes.

This concept of ‘ranging over’ and selecting an individual with a specific property out of the whole group works in the vast majority of applications.  However, I’ve pointed out a few instances in which it makes no sense to think of the domain as a predetermined group of objects, such as in natural language and relativistic situations.  In these cases the domain cannot be defined until something about the people involved are known, if at all; people may have a stock set of responses to questions but can also make new ones up.

So, since the problem resides with a static domain being linked to specific people, I suggest that we find a way to link quantifiers to those people.  This means that if two people are playing a logic game, each person will have their own quantifiers linked to their own domain.  The domains will be associated with the knowledge (or other relevant property) of the people playing the game.

We could index individual quantifiers to show which domain they belong to, but game theory has a mechanism for showing which player is making a move by using negation.  When a negation is reached in a logic game, it signals that it is the other player’s turn to make a move.  I suggest negation should also signal a change in domains, as to mirror the other player’s knowledge.

Using negation to switch the domain that the quantifiers reference is more realistic/ natural treatment of logic: when two people are playing a game, one may know certain things to exist that the other does not.  So using one domain is an unrealistic view of the world because it is only in special instances that two people believe the exact same objects to exist in the world.  Of course there needs to be much overlap for two people to be playing the same game, but having individual domains to represent individual intelligences makes for a more realistic model of reality.

Now that each player in a game has his or her own domain, what is the activity of the quantifier?  It still seems to be ranging over a domain, even if the domain is separate, so the problem raised above has not yet been dealt with.

Besides knowing different things, people think differently too.  The different ways people deal with situations can be described as unique strategies.  Between the strategies people have and their knowledge we have an approximate representation of a person playing a logic game.

If we now consider how quantifiers are used in logic games, whenever we encounter one we have to choose an element of the domain according to a strategy.  This strategy is a set of instructions that will yield a specified result and are separate from the domain. So quantifiers are calls to use a strategy as informed by your domain, your knowledge.  They do not ‘range over’ the domain; it is the strategies a person uses that take the domain and game (perhaps “game-state” is more accurate at this point) as inputs and returns an individual.

The main problem mentioned above can now be addressed: Instead of predetermining sets objects in domains, what we need to predetermine are the players in the game. The players may be defined by a domain of objects and strategies that will be used to play the game, but this only becomes relevant when a quantifier is reached in the game.  Specifying the players is sufficient because each brings his or her own domain and strategies to the game, so nothing is lost, and the domain and strategies do no have to be predefined because they are initially called upon within the game, not before.

I don’t expect this discussion to cause major revisions to the way people go about practicing logic, but I do hope that it provides a more natural way to think about what is going on when dealing with quantifiers and domains, especially when dealing with relativistic or natural language situations.

Posted in epistemology, game theory, logic, philosophy. Tagged with , , , , , , .

Where Does Probability Come From? (and randomness to boot)

I just returned from a cruise to Alaska. It is a wonderful, beautiful place. I zip-lined in a rain forest canopy, hiked above a glacier, kayaked coastal Canada and was pulled by sled-dogs. Anywho, as on many cruises, there was a casino, which is an excellent excuse for me to discuss probability.

What is probability and where does it come from? Definitions are easy enough to find. Google returns:

a measure of how likely it is that some event will occur; a number expressing the ratio of favorable cases to the whole number of cases possible …

So it’s a measure of likelihood. What’s likelihood? Google returns:

The probability of a specified outcome.

Awesome. So ‘probability as likelihood’ is non-explanatory. What about this ‘ratio of favorable cases to the whole number of cases possible’? I’m pretty wary about the word favorable. Let’s modify this definition to read:

a number expressing the ratio of certain cases to the whole number of cases possible.

Nor do I like ‘a number expressing…’ This refers to a particular probability, not probability at large, so let’s go back to using ‘measure’:

a measure of certain cases to the whole number of cases possible.

We need to be a bit more explicit about what we are measuring:

a measure of the frequency of certain cases to the whole number of cases possible.

OK. I think this isn’t that bad. When we flip a fair coin the probability is the frequency of landing on heads compared to the total cases possible, heads + tails, so 1 out of 2. Pretty good.

But notice the addition of the word fair. Where did it come from, what’s it doing there? Something is said to be fair if that thing shows no favoritism to any person or process. In terms of things that act randomly, this means that the thing acts in a consistently random way. Being consistently random means it is always random, not sometimes random and other times not random. This means that fairness has to do with the distribution of the instances of the cases we are studying. What governs this distribution?

In the case of of a coin, the shape of the coin and the conditions under which it is measured make all the difference in the distribution of heads and tails. The two sides, heads and tails, must be distinguishable, but the coin must be flipped in a way such that no one can know which side will land facing up. The shape of the coin, even with uniform mass distribution, cannot preclude this previous condition. Therefore the source of probability is the interdependence of physical conditions (shape and motion of the coin) and an epistemic notion (independence of knowledge of which side will land up). When the physical conditions and our knowledge of the conditions are dependent upon each other then the situation becomes probabilistic because the conditions preclude our knowing the exact outcome of the situation.

It is now time to recall that people cheat at gambling all the time. A trio of people in March 2004 used a computer and lasers to successfully predict the decaying orbit of a ball spinning on a roulette wheel (and walked out with £1.3 million). This indicates that after a certain point it is possible to predict the outcome of a coin flipping or a roulette ball spinning, so the dependence mentioned above is eventually broken. However this is only possible once the coin is flipping or the roulette ball is rolling, not before the person releases the roulette ball or flips the coin.

With the suggestion that it is the person that determines the outcome we can expand the physical-epistemic dependence to an physical-epistemic-performative one. If I know that I, nor anyone else, can predict the outcome until after I perform a task, then the knowledge of the outcome is dependent upon how I perform that task.

This makes sense because magicians and scam artists train themselves to be able to perform tasks like shuffling and dealing cards in ways that most of us think is random but are not. The rest of us believe that there is a dependence between the physical setup and the outcome that precludes knowing the results, but this is merely an illusion that is exploited.

What about instances in which special training or equipment is unavailable; can we guarantee everyone’s ability to measure the thing in question to be equal? We can: light. Anyone who can see at all sees light that is indistinguishable from the light everyone else sees: it has no haecceity.

This lack of distinguishability, lack of haecceity (thisness), is not merely a property of the photon but a physical characteristic of humans. We have no biology that can distinguish one photon from another of equivalent wavelength. To distinguish something we have to use a smaller feature of the thing to tell it apart from its compatriots. Since we cannot see anything smaller, this is impossible. Nor is there a technology that we could use to augment our abilities: for us to have a technology that would see something smaller than a photon would require us to know that the technology interacted at a deeper level with reality than photons do. But we cannot know that because we are physically limited to using the photon as our minimal measurement device. The act of sight is foundational: we cannot see anything smaller than a photon nor can anything smaller exist in our world.

The way we perceive photons will always be inherently distributed because of this too. We cannot uniquely identify a single photon, and hence we can’t come back and measure the properties of a photon we have previously studied. Therefore the best we will be able to accomplish when studying photons is to measure a group of photons and use a distribution of their properties, making photons inherently probabilistic. Since the act of seeing light is a biological feature of humans, we all have equal epistemological footing in this instance. This means that the epistemic dependence mentioned above can be ignored because it adds nothing to the current discussion. Therefore we can eliminate the epistemic notion from our above dependence, reducing it to a physical-performative interdependence.

Since it is a historical/ evolutionary accident that the photon is the smallest object we can perceive, the photon really is not fundamental to this discussion. Therefore, the interdependence of the physical properties of the smallest things we can perceive and our inherent inability to tell them apart is a source of probability in nature.

This is a source of natural randomness as well: once we know the probability of some property that we cannot measure directly, the lack of haecceity means that we will not be able to predict when we will measure an individual with said property. Therefore the order in which we measure the property will inherently be random. [Assume the contradiction: the order in which we measure the property is not random, but follows some pattern. Then there exists some underlying structure that governs the appearance of the property. However, since we are already at the limit of what can be measured, no such thing can exist. Hence the order in which we measure the property is random.]


If I were Wittgenstein I might have said:

Consider a situation in which someone asks, “How much light could you see?” Perhaps a detective is asking a hostage about where he was held. But then the answer is, “I didn’t look.” —— And this would make no sense.

hmmmm…. I did really mean to get back to gambling.

Posted in biology, epistemology, evolution, fitness, independence friendly logic, logic, measurement, mind, philosophy, physics, Relativity, science, Special Relativity, technology. Tagged with , , , , .

The Monty Hall Problem

[check out my more recent Monty Redux for, perhaps, a clearer exposition]

The Monty Hall Problem illustrates an unusual phenomenon of changing probabilities based upon someone else’s knowledge. On the game-show Let’s Make a Deal the host, Monty Hall, asks the contestant to choose one of three possibilities – Door One, Two or Three – with one door leading to a prize and the other two leading to goats. After the contestant selects a door, another door is opened, one with a goat behind it. At this point the contestant is allowed to switch the previously selected door with the remaining (unopened) door.

Common intuition is that this choice does not present any advantage because the probability of selecting the correct door is set at 1/3 at the beginning. Each door has this 1 out of 3 chance of having a prize behind it, so changing which door you select has no effect on the outcome.

In hindsight, this intuition is wrong. If you initially selected the first goat and then switch when you get a chance, you win. If you selected the second goat and switch, you win. If you selected the prize and switch, you lose. Therefore if you switch, you win 2 out of 3, whereas if you do not switch you win only 1/3 of the time.

So what has gone horribly wrong here:

  1. Why is most everyone’s intuition faulty in this situation?
  2. How does switching doors make any difference?
  3. When did the 1/3 probability turn into a 2/3 probability?

At the beginning of the game you have a 2 out of 3 chance of losing. Likewise the game show has a 2 out of 3 chance of winning (not giving you a prize) at the beginning of the game. Both of these probabilities do not depend upon which door the prize is behind, but only upon the set-up of a prize behind only one of three doors. For instance, an outside service (not the game show) could have set everything up such that both you and the game show would be kept in the dark: there would still be 2 goats and a prize, but neither you nor the game show would know which door led to the prize.

Now imagine that it is the game show that is playing the game. The game show is trying to win by selecting a goat. From this perspective, whichever door that was chosen is good: this door has a 2 out of 3 probability of being a winner (being a goat). Therefore when given the opportunity to change (after the outside service opens a door and shows a goat), there is no reason to do so.

Of course you, the contestant, are the one making the selection, and you do not want a goat. However, if you imagined yourself in the position of the game show at the beginning, as trying to select a goat, you would reasonably assume that, just as the game show did, you were successful in choosing a goat. When given the choice to switch, now that the other goat has been removed, it seemingly makes sense to change your selection.

In this case the easiest way to view the situation is in terms of how to lose, or by considering all the possible outcomes (as mentioned above). Though this is a guess, it seems that our first blush reaction to this problem is always to view it in terms of winning and this is the reason we do not immediately recognize the benefit in switching. We start out with a 1/3 chance of winning and switching doors doesn’t immediately seem to increase this percentage.

To answer how switching doors makes a difference we need to look more closely at the doors. The door that was initially selected has a 1 out of 3 chance of being a prize, and this does not change. If you were to play many times and ignore changing doors, then you would win 33.3% of the time. At the outset the other two doors each have the exact same chance of being a winner, 1 out of 3. So the other two doors combined have a 2 out of 3 chance of containing a winning door.

Now the game show changes the number of doors available from 3 to 2, with one door guaranteed to contain a prize. If you were presented this situation without knowledge of the previous process, then you would rightly put the chance of selecting the prize at 1 out of 2, 50%.

However, you know something about the setup: The door that was initially selected had a probability of having a prize behind it set at 1 out of 3. The thing behind the other door, though, has been selected from a stacked deck: Whatever is behind the door was selected from a group of objects with a 2 out of 3 chance of containing a prize (1/3 + 1/3). You know that the odds on this door are stacked in your favor because the game show knowingly reveals the goat: In the 2/3 case in which you have previously selected a goat, the prize is behind one of the other two doors. When the game-show reveals (and removes) a goat, it guarantees that the prize is behind the last door. Therefore switching doors at the end is equivalent to combining and selecting the probability associated with the two doors not initially selected.

If the game show did not knowingly reveal the goat, you would not be able to take advantage of the stacked deck. Imagine that you select the first door and then another door is opened randomly, revealing a goat. By randomly eliminating this door (and not looking behind the unselected doors) the door that was initially selected becomes unrelated to the present choice: Only by looking behind the unselected doors does the initial selection become fixed in reference to the other doors. Since no one looked behind the doors, some bored, but not malicious, demon could have come and switched whatever was behind the selected and remaining door and neither you nor the game-show would be able to tell. Therefore switching doors when a goat is randomly revealed provides no advantage because the initial selection cannot be related to the probable location of the prize.

Only when the contestant can fix the probable locations of the prize because the location of the prize is known by the game-show, is it possible to assign interdependent probabilities on the location of the prize and the previous selection made. The odds are then tilted in the contestant’s favor by switching away from the low probability initial selection to the door that has the combination of remaining probabilities.

The logic of this needs to be represented game-theoretically with the different quantifiers representing different players of a game of incomplete information. The game would run* like this:

Domain={prize, goat, goat}

Contestant Game Show
1. ∃x∃y∃z∀a/x,y,z∃b∀c/x,y,z(a=x & b=y & c=z)
2. ∃y∃z∀a/x,y,z∃b∀c/x,y,z(a=g & b=y & c=z)
3. ∃z∀a/x,y,z∃b∀c/x,y,z(a=g & b=g & c=z)
4. ∀a/x,y,z∃b∀c/x,y,z(a=g & b=g & c=p)
5. ∃b∀c/x,y,z(p=g & b=g & c=p)
6. ∀c/x,y,z(p=g & g=g & c=p)
7. ∀d∀c/x,y,z(d=g & g=g & c=p)
8. ∀c/x,y,z(g=g & g=g & c=p)
9. (g=g & g=g & p=p)

Line 1 is the initial setup of the prize game: the goal is for the contestant to make his or her placement of the prize and goats match the game show’s placement. Whatever is on the left side of an = will be what the contestant thinks is behind a door and what is on the right of an = will be what the game show puts behind the door, such that each = represents a door. If the formula is satisfied then the contestant will have successfully guessed the location of the prize.

Lines 2, 3 and 4 represent the results of the Game Show placing the prize and goats. Line 5 is the result of the first move of the contestant choosing where he or she thinks the prize is: the ‘a/x,y,z’ means that whatever placed in spot a has to be done independently, i.e. without knowledge, of what x or y or z is. Then the game show reveals a goat behind one of the doors not selected by the contestant. Line 7 represents the choice that is given to the contestant to switch his or her initial placement of where the prize is. Line 8 is the important step: since the contestant does not know what is behind the doors (c/x,y,z) it looks as if there is no advantage to switching. However, the contestant does know that when making a choice to reveal a goat in line 6 that at this point the game show had to know what was behind every door. This means that c is dependent upon b which was depended upon x, y, and z. With this knowledge the contestant can figure out that there is an advantage to switching because the selection of b in line 6 fixed the locations of the prize & goats and in doing so fixed the odds. Since the odds were intially stacked against the contestant, switching to the only remaining door flips the odds in the contestant’s favor, and is done so in this example. Line 9 shows that all the contestant’s choices match up with what the game show has placed behind the doors and hence she or he has won the prize.

* To do a better representation would require keeping the gameshow from not placing a prize anywhere by using a line like ‘x≠y or x≠z’. For graphical brevity I left it out.

Posted in game theory, independence friendly logic, logic, measurement, philosophy. Tagged with , , , .