Re: biological systems and cryptography
On Tue, Dec 31, 2002 at 12:23:51PM -0800, Tim May wrote:
On Tuesday, December 31, 2002, at 11:41 AM, Michael Cardenas wrote:
How do you all see the future use of biologically based systems affecting cryptography in general?
By biologically based systems I mean machine learning, genetic algorithms, chips that learn (like Carver Mead's work), neural networks, vecor support machines, associative memory, etc.
Strong crypto is, ipso facto, resistant to all of the above. For the obvious reason that the specific solution to a cipher is like a Dirac delta function (a spike) rising above a featureless plain, this in terms of the usual hill-climbing or landscape-learning models which all of the above use in one form or another.
People do break cyphers, by finding weaknesses in them. Are you saying that you think that current cyphers are unbreakable? Also, what about using biological systems to create strong cyphers, not to break them?
Cryptanalysis of weak crypto, in terms of mundane things like passphrase guessing, finding images tagged with stego code, etc., already in some cases makes use of these tools. Bob Baldwin's Crytpographer's Workbench used learning algorithms a long time ago.
Strong math wins out over weak crypto any day, and attempting to brute force a cipher with even a swimming pool full of Adleman machines will not work: if a 400-digit number takes, for instance, a million Pentium 4 years to brute force factor, then how long does a 600-digit number take?
(And using larger RSA moduli is of course trivial...)
Homework: Using the estimates Schneier, Diffie, Hellman, and others have made for the number of computer operations to break ciphers of various kinds, describe a reasonable cipher and modulus or key length which will take more energy than there is in the entire universe to break. The answer, in terms of how small the key or modulus is, may surprise you.
It seems that all of these analyses assume that an instruction is a single mathematical operation in a turing machine. What if each operation was something else? I refuse to believe that the human mind is just a turing machine. -- michael cardenas | lead software engineer, lindows.com hyperpoem.net | GNU/Linux software developer people.debian.org/~mbc | encrypted email preferred "It is as hard to see one's self as to look backwards without turning around." - Henry David Thoreau [demime 0.97c removed an attachment of type application/pgp-signature]
On Wed, 1 Jan 2003, Michael Cardenas wrote:
People do break cyphers, by finding weaknesses in them. Are you saying that you think that current cyphers are unbreakable?
People break cyphers by 1) cryptoanalysis (mostly brain, a bit of muscle) 2) brute force (no brain at all, pure muscle) So far we've been talking purely brute force here. It is easy to see that a current cypher requiring a 10^6..10^9 computer-years brute force can be broken in realtime if attacked by a massively parallel molecular electronics computer common several decades downstream. It is trivial to design cyphers running on today's hardware which would be safe from that attack. No one is doing that because you want speedy encryption on today's software. And secrets become stale quick. The quantum computer is a dark horse, because no one really knows how much quantum parallelism you can extract from a given pile of molecules. Plus, not all algorithms can be mapped to a QM machine. It also seems that entanglement is energy bound, but the field is still moving far too quickly to say anything meaningful.
Also, what about using biological systems to create strong cyphers, not to break them?
Molecules are molecules. Solvated linear biopolymers are lousy computers in any case. So are 5-qubit QM machines which require an NMR machine, for that matter. Whether classical, or QM, it has better be solid state, and preferrably not require mK environment to run.
It seems that all of these analyses assume that an instruction is a single mathematical operation in a turing machine. What if each operation was something else? I refuse to believe that the human mind is just a turing machine.
The human mind is not a Turing machine. But a Turing machine is an all purpose computational device, so in principle it can simulate relevant aspects of physical system evolution. Including the spiking, diffusion, genomic activity networks in each cell and god knows what else in the physical system residing between your ears. Meaning, a suitably constructed machine could be intelligent. It can be even you if you keep your and its state synched.
At 08:55 PM 1/1/03 -0800, Michael Cardenas wrote:
On Tue, Dec 31, 2002 at 12:23:51PM -0800, Tim May wrote: ...
Strong crypto is, ipso facto, resistant to all of the above. For the obvious reason that the specific solution to a cipher is like a Dirac delta function (a spike) rising above a featureless plain, this in terms of the usual hill-climbing or landscape-learning models which all of the above use in one form or another.
People do break cyphers, by finding weaknesses in them. Are you saying that you think that current cyphers are unbreakable?
Well, there's a difference between a system to recover plaintext given ciphertext (which ought not to work for any decent cipher, given a hill-climbing sort of approach), and a system to help a human work out the right way to cryptanalyze a system. Hill-climbing techniques make sense when analyzing a component of a cipher, say. (I know people have done stuff like this in various places, but I'm away from my library, so you'll have to look it up yourself.)
Also, what about using biological systems to create strong cyphers, not to break them?
This ought to just be the other side of using these systems to do analysis. If you can find an especially good way to partition the set of texts for a partitioning attack, you can use that to decide how to design your cipher to resist the attacks, for example. --
michael cardenas | lead software engineer, lindows.com hyperpoem.net | GNU/Linux software developer people.debian.org/~mbc | encrypted email preferred
--John Kelsey, kelsey.j@ix.netcom.com
participants (3)
-
Eugen Leitl -
John Kelsey -
Michael Cardenas