"Verification Problem" for Quantum Computers
In his book, Labyrinths of Reason, William Poundstone
talks of the “Oracle of the Maze”, an oracle who can answer any question
immediately. Even if he is a true oracle, convincing others isn’t easy. Why?
“Answer a question that no one else
possibly could, and he is accused of fabricating; answer a question whose
answer is known or knowable, and he is accused of cheating.”
So is there no way
out for the oracle? Aha, there is: “a difficult
question whose answer, once stated, can be verified easily”. The solution to an insanely big and twisted maze is an
example, ergo, the name Poundstone gave: Oracle of the Maze.
This topic is of
practical relevance (no, not maze solving; checking if an oracle is really an
oracle) in… quantum computing, writes
Erica Klarreich:
“Once a quantum computer can perform
computations a classical computer can’t, how will we know if it has done them
correctly?”
See the parallel
with the Oracle of the Maze? Is there “any ironclad guarantee that it really
has done what it claimed”? Worse, unlike the solution to a maze, many problems
a quantum computer could solve cannot even be checked using classical methods! This then is the “verification
problem”.
Researchers had
proven that a verifier can check a quantum computer if the verifier can measure
quantum bits. But to read a quantum bit, you needed a quantum computer. But
trying to use a quantum computer to check the answer of a quantum computer is
circular…
Enter Urmila
Mahadev, a seventh-year graduate student at Berkeley (it probably helped that
“I was never thinking of graduation, because my goal was never graduation”).
Here’s the simplified gist of her solution (sorry, it’s still very technical/quantum):
1)
The
quantum computer first calculates the answer;
2)
It
then entangles the answer with
another piece of quantum information that the computer does not know but for which the verifier
himself knows the options to;
3)
At
that stage, making the computer read the quantum bit(s) from the entangled
entity would solve the circularity problem. Because the verifier already knows
the possible options to #2, he can use that part as a check for the validity of
the answer the computer found to #1.
In principle, this
works. But the hard part is #2, building a “secret state” that is known only to
the verifier but not the computer. Mahadev with her doctoral advisor, Umesh
Vazirani, and Paul Christiano found a way to solve #2. The concept is called
the “trapdoor function”. Why the name? Because the function ensures that “it’s
impossible for the quantum computer to cheat significantly without leaving
unmistakable traces of its duplicity”, as Klarreich puts it.
While this is a
huge step in solving the “verification problem”, it still has two
limitations/risks:
A)
Step
#3 above is just a very educated guess: what are the odds that a quantum
computer guessed the bits of an unknown piece of information by sheer chance?
The lower the odds, the more likely it really works…
B)
The
trapdoor function itself. It needs to be something even a quantum computer
can’t crack. While we believe we know a few such functions exist, what if we
were wrong and a quantum computer could solve them? Of course, that would only
mean we need to find another trapdoor function, not that Mahadev’s concept is
wrong.
In any case,
Mahadev’s protocol won’t be feasible for at least five years, but hey, it’s not
like we have quantum computers around the corner either. But it makes sense to
have a way to check the correctness of one whenever it’s ready.
Comments
Post a Comment