please enable javascript

Committing to lunch

Alexander Comerford
July 11th, 2021 · 10 min read

Alice and Bob are close friends that have known each other for years. Recently they decided that they wanted to schedule a weekly Saturday lunch together to catch up on their lives and shoot the shit. They enjoy their discussions as well as the food. Sometimes Alice chooses where to eat, sometimes Bob chooses, and sometimes they just return to their favorite spots. Several weeks go by and the constant debate they always revisit is where are they going to eat next. Sometimes both of them want to choose, and sometimes neither of them wants to choose. Eventually they agreed that in those circumstances, they need a systematic way to decide where they are going to partake in their next meal.

They both agree on two processes to settle their lunch dilemma. The first process is applied when each wants to go to a different restaurant. In this circumstance one of them is randomly chosen to become the decider for that week. The second process is applied when neither party has a preference on a set of places. In this circumstance, a random place is chosen (no decider).

They want the processes to be fair, so sometime during the week, they meet up in person and either flip a coin to see who becomes the decider, or they roll a die to choose a place from a list. Both of these methods work for them. Happy with the coin and the die, Alice and Bob enjoy several more weeks of lunches.

Unfortunately, this doesn't last. Alice and Bob's schedules start to get busy and they don't get a lot of time to see each other in person during the week. Therefore they can never find time to meet up to flip a coin or roll a die. This puts Alice and Bob in a pickle. How will they decide where to go to lunch if they can't meet up?!

They found it most convenient with their busy schedules to communicate online. One of them would have to do the flip/roll and tell the other the result. But this puts both of them in a tough spot. Despite being great friends, they don't trust each other to be honest when it comes to lunch. One of them could easily lie without the other knowing. They both agree that they want a way to ensure the authenticity of a flip/roll when they can't directly see it.

Alice and Bob need a virtual coin/die so when one of them flips/rolls the other can verify that no cheating took place.

This problem could be solved by trusting a third person to flip/roll for them. However they avoid this option to eliminate the risk that they would bribed with an egg-roll.

Overall they need a scheme that satisfies a few requirements

  1. If someone cheats, the other one can prove it
  2. If there is no cheating, then they can prove that the process was fair
  3. Cheating should be hard and therefore disincentivized and unlikely

Creating a virtual coin

Alice and Bob both do some research on finding a solution to their problem. After a few days of digging, Alice comes back to Bob with a solution to the "virtual coin" problem that is fair for the both of them. She proposes a hash-based commitment scheme used by Manuel Blum to build a secure coin flip and sends Bob the paper.

Alice is really excited about this protocol because not only is it simple to execute, but it is hard to cheat under the random oracle model. Bob upon hearing this looks a little puzzled and inquires about the random oracle model. Alice begins to explain ...

What is a random oracle?

The random oracle model is a cryptographic abstraction assuming the existence of a function where given some arbitrary input, the output of the function is 'truly' random from the random oracle's output domain. Additionally if the same input were given to the random oracle, it would respond with the same output every time.

If this sounds familiar, that's because the definition of a random oracle is similar to that of an ideal hash function commonly used when programming. To understand why Alice's commitment scheme is hard to cheat and possible to prove fair, we need to understand what is a hash function.

To learn more about random oracles check out Matthew Green's blog post about it.

What is a hash function?

The simplest definition of a hash function is a function that inputs an arbitrary amount of data, and outputs a fixed sized amount of data. This is very generic and many functions can satisfy this. However in order to make a hash function behave like a random oracle, they need to have some additional 'ideal' properties.

The first property our hash function needs to have is one-way determinism. This means that no matter where or who you are, the same input to a hash function yields the same output and that output is hard to invert. This is important because non-determinism leads to varying results which makes our function inconsistent (which we don't want). It also means that given some arbitrary output from the hash function's range, it is hard to find the input that can produce it.

Alice and Bob both want their results to be consistent and hard to invert. This means they need this one-way determinism property so they both can trust that they can compute the same things.

The second property, that will prevent cheating in Alice's proposed commitment scheme, is called "collision resistance."

What is collision resistance?

This will be the deepest we will go before we return to Alice and Bob's lunch conundrum.

A hash function hh is said to be 'collision resistant' when it is hard to find two differing inputs aa, bb that produce the same hash value such that h(a)=?h(b)h(a) \stackrel{?}{=} h(b).

In the real world, perfect collision resistance (and therefore random oracles) are impossible to construct due to the pigeonhole principle. However, a lot of research has been done to make the possibility of finding collisions really, really hard. So hard that we can assume that it's not worth the time and effort for any adversary to try and break it.

There are several popular hash functions that have been made (such as BLAKE or Keccack) that are so trusted, they are often the underlying foundation of several other cryptographic systems like HMACs, password storage, and POW blockchains.

Now that we know more about hash functions and collision resistance, we can rejoin Alice and Bob.

The coin flipping protocol

Now that Alice and Bob both understand hash functions and collision resistance, Alice can lay the protocol out to Bob. In this protocol there will be two roles.

  • Guesser: The one who guesses the coin flip
  • Flipper: The one who performs the coin flip

The role of the Guesser will be played by Bob, and the role of the Flipper will be played by Alice. The ground rules / assumptions are:

  1. Cheating with proof will result in an immediate loss for the cheater
  2. Both parties will use the same hash function and agree on a format for exchanging data
  3. Both parties are talking over a secure channel that can't be intercepted or tampered with
  4. If the Guesser correctly predicts the flip, they will become the decider. Otherwise, the Flipper will become the decider

With these ground rules in place, Alice tells Bob the protocol operates in a sequence of steps.

  1. Alice and Bob each creates a one time password (OTP) (call them pap_a and pbp_b)

    alicebobpapb\def\arraystretch{0.9} \begin{array}{c:c} alice & bob \\ \hline p_a & p_b \\[0.1cm] \end{array}
  2. Bob sends Alice his OTP

    alicebobpa,pbpb\def\arraystretch{1} \begin{array}{c:} alice & bob \\ \hline p_a, p_b & \larr p_b \\[0.1cm] \end{array}
  3. Alice creates a commitment that will 'commit' her flip for the rest of the protocol

    • She flips a coin (call it fa{0,1}f_a \isin \{0, 1\})
    • She computes her commitment ca=h(fa,pa,pb)c_a = h(f_a, p_a, p_b) (where hh is the collision resistant hash function)
    • She sends her commitment cac_a to Bob
    alicebobpa,pbpbfa,h(fa,pa,pb)ca\def\arraystretch{1} \begin{array}{c:} alice & bob \\ \hline p_a, p_b & p_b \\[0.1cm] f_a, h(f_a, p_a, p_b) \rarr & c_a \\[0.1cm] \end{array}
  4. Bob guesses the flip (call it gb{0,1}g_b \isin \{0, 1\}) and sends this to Alice

    alicebobpa,pbpbfa,cacagbgb\def\arraystretch{1} \begin{array}{c:} alice & bob \\ \hline p_a, p_b & p_b \\[0.1cm] f_a, c_a & c_a \\[0.1cm] g_b & \larr g_b\\[0.1cm] \end{array}
  5. Alice reveals her flip and OTP by sending them to Bob

    alicebobpa,pbpbfa,cacagbgb(fa,pa)(fa,pa)\def\arraystretch{1} \begin{array}{c:} alice & bob \\ \hline p_a, p_b & p_b \\[0.1cm] f_a, c_a & c_a \\[0.1cm] g_b & g_b \\[0.1cm] (f_a, p_a) \rarr & (f_a, p_a) \\[0.1cm] \end{array}
  6. Bob now verifies that Alice 'committed' to her flip and OTP with the following equality

    h(fa,pa,pb)=?ca\begin{CD} h(f_a, p_a, p_b) \stackrel{?}{=} c_a \end{CD}
  7. If the equality holds true, both Alice and Bob can now agree on the outcome of the toss with the final equality

    fa=?gb\begin{CD} f_a \stackrel{?}{=} g_b \end{CD}

Now that Alice has explained the coin flipping protocol, Bob is immediately skeptical and asks why he should trust that this scheme can't be cheated? Alice then explains why cheating is hard.

Why is the coin flipping protocol hard to cheat?

To figure out why this protocol is hard to cheat, Alice lays out some of the possible ways to cheat, and how the other party can catch them.

Cheating as the Guesser

In Bob's role as the Guesser there isn't much room for cheating. The only way he can cheat (and win) is if he has pre-knowledge of Alice's flip before he guesses (in step 4) which we will assume he doesn't.

Cheating as the Flipper

In Alice's role as the Flipper there are a few places where she can try to cheat: in step 3 (when she sends her commitment) or step 5 (revealing her flip and OTP). Successfully cheating for Alice means that she must be able to convince Bob that he guessed incorrectly no matter which outcome he chooses!

To do this requires Alice, in step 4, to reveal her flip as the opposite of Bob's guess. The problem for her is that this flip (and OTP) must coincide with the commitment she sent in step 3.3 or Bob will know she is cheating. To continue with her ruse, Alice needs to craft a special commitment in step 3 such that either possible flip (and OTP) that she reveals can be successfully verified by Bob without suspicion. This means that Alice needs to satisfy the equality

h(0,pa0,pb)=h(1,pa1,pb)\begin{CD} h(0, p_{a0}, p_b) = h(1, p_{a1}, p_b) \end{CD}

If Alice is able to satisfy this equality then she can take the following measures to ensure she wins no matter how Bob guesses.

  • When Bob guesses "0", then Alice can reveal a flip of "1" and pa1p_{a1} to show Bob she is abiding by her earlier commitment and that he lost

  • When Bob guesses "1", then Alice can reveal a flip of "0" and pa0p_{a0} to show Bob she is abiding by her earlier commitment and that he lost

This is terrible news for Bob as no matter how he guesses he will lose! Then Alice will have a 100% chance of choosing where to go to lunch!

Luckily if we require that the hash function hh being used has the "collision resistance" property, Alice will not be able to find such a commitment that relies on this hash collision and would have to find some other way to cheat.

  • Footnote: You might be wondering why Bob would need to generate an OTP at the start at all. Bob does this so he isn't vulnerable to a pre-image attack.

With the collision resistance property in place, Alice has the incentive to honestly reveal her flip and one time password. If she doesn't, she risks being caught cheating and forfeits the game. Bob has enough assurance that he won't be a victim to foul play.

Alice and Bob both agree on this scheme and when they can't agree where to go to lunch, they bring out the coin flipping protocol, find out who is "the decider," then delegate them with the responsibility of deciding where they go.

Creating a virtual die

Alice and Bob enjoy several more lunches with the aid of their arbiter the "virtual coin." But as their next lunch comes around, Bob gets recommended a top 10 ramen list for the area. Bob loves ramen and recommends to Alice that they should go somewhere on that list for their next lunch. Alice is in the mood for ramen and agrees. They read over the list and can't agree on just one place as all of them look delicious! Even though they have their trusty "virtual coin," they both agree that neither one of them wants to be "the decider" as they don't want to be forced to choose. They would rather have their old die that would "decide" for them.

Alice and Bob's "virtual coin" solves their problem of establishing who is the decider of their lunch spot. However it doesn't solve their problem of choosing a set place from a list like their die did. What they want, just like their "virtual coin," is a "virtual die" with some arbitrary amount of sides. And just like their coin, they want the die to have the same security guarantees.

Alice and Bob do some more research, read a few more papers, and Bob comes back to Alice with a solution to the "virtual die" problem. He proposes a modified scheme of the one Alice introduced.

Instead of a Guesser and a Flipper, both will play the role of a Roller.

Bob lays out the protocol as follows

  1. Alice and Bob each create a one time password (OTP) and (with the same names as before). They also each create a random value within some agreed upon range (call them vav_a and vbv_b)

    alicebobpa,vapb,vb\def\arraystretch{0.9} \begin{array}{c:c} alice & bob \\ \hline p_a,v_a & p_b,v_b \\[0.1cm] \end{array}
  2. Both Alice and Bob create commitments for their respective values by computing h(p,v)h(p, v) (call them cac_a and cbc_b), then they exchange them

    alicebobpa,vapb,vbh(pa,va)h(pa,va)\def\arraystretch{0.9} \begin{array}{c:c} alice & bob \\ \hline p_a,v_a & p_b,v_b \\[0.1cm] h(p_a,v_a) \rarr & \larr h(p_a,v_a) \\[0.1cm] \end{array}
  3. Alice and Bob reveal their OTP and random value (pp and vv respectively) to each other

    alicebobpa,vapb,vbca,cbca,cb(pa,va)(pb,vb)\def\arraystretch{0.9} \begin{array}{c:c} alice & bob \\ \hline p_a,v_a & p_b,v_b \\[0.1cm] c_a,c_b & c_a,c_b \\[0.1cm] (p_a,v_a) \rarr & \larr (p_b,v_b) \\[0.1cm] \end{array}
  4. Alice and Bob verify that the other has 'committed' to their random value

    h(p,v)=?c\begin{CD} h(p, v) \stackrel{?}{=} c \end{CD}
  5. Alice and Bob both compute h(va,vb)(modn)h(v_a, v_b) \pmod n as the output of the roll (where nn is the number of restaurants)

Now that Bob has explained the die rolling protocol, Alice is suspicious if this new protocol is just as hard to cheat as the coin flipping protocol. Bob explains why the die rolling protocol is hard to cheat.

Why is the die rolling protocol hard to cheat?

Bob explains that this protocol is hard to cheat for the same reasons the coin flipping protocol. This is because they both use the same underlying commitment mechanism.

In order for a Roller (we'll say Bob in this case) to successfully cheat, he would need to successfully manipulate the output without Alice being able to catch him.

To do this, Bob would first need to wait to see Alice's revealed value in step 5.

Next, using Alice's revealed value, Bob can determine the output of the roll by performing steps 6 & 7 without communicating to Alice. He can then manipulate vbv_b to find his desired output (call this new value vbv_b^{\prime}). Once he finds his desired output, Bob now needs to find a new OTP (call this pbp_b^{\prime}) such that the following equality is met.

cb=?h(pb,vb)\begin{CD} c_b \stackrel{?}{=} h(p_b^{\prime}, v_b^{\prime}) \end{CD}

If Bob can satisfy this equality then he can reveal a convincing OTP and value in step 5 that can be verified by Alice and confince her of his desired output.

Unfortunately for Bob (or fortunately for Alice), the hash function hh they previously agreed upon is collision resistant. This means Alice can trust that finding a collision is hard and that Bob won't manipulate the output of the roll.

Alice and Bob agree on this new virtual die scheme so when they jointly need to decide on a single spot from a set of restaurants, they break out the die rolling protocol.

With the virtual coin and virtual die both of Alice and Bobs lunch conundrums are solved. They enjoy many more lunches at ease knowing they have a trustless way to agree on a place to eat.

Committing to lunch

Hearing the phrase "flipping a coin over the telephone" at first seems paradoxical and impossible. But once you frame the problem in the form of commitments, suddenly it doesn't seem impossible at all.

Commitment schemes / protocols are an eye opening concept that are just the tip of the cryptography iceberg as they are the basis for even more incredible concepts like zero knowledge proofs and secure computation. Although Alice and Bob's construction of commitments is simple, they use it to reach a trustless agreement which is an incredibly powerful tool.

Cryptography always keeps me in a consistent state of awe as whole collections of seemingly paradoxical problems are defeated again and again with elegant logical solutions. I hope commitments can be your gateway to learning more about this amazing world.

More posts from The Art of Abstraction

The Rainbow Hats Puzzle

ROYGBV

April 27th, 2024 · 4 min read

Lagrange's Theorem

Groups in groups in groups

July 16th, 2023 · 5 min read