On Mon, Mar 2, 2015 at 1:57 PM, John Rose via AGI <[email protected]> wrote:
> Interestingly, random numbers are related to AGI in many ways and are 
> associated with many AGI mathematical ideas but few things if any are related 
> as much to the quality of the randomness as much as cryptography. Probably 
> because of that limiting Boolean of being able to decrypt something or the 
> goal of preventing that distinct event.

Cryptography plays a critical role in my distributed AGI proposal.
http://mattmahoney.net/agi2.html

Peers need to know when two messages are from the same source. If a
peer earns a reputation for being a reliable source of information
(like Google or your bank), then malicious peers will try to spoof
messages from them. To prevent this, peers sign their messages using a
mutually agreed secret key chosen at random. After an initial exchange
(using e.g. Diffie-Hellman), I send you a message and a signature like
SHA256(message + key). You receive the message, compute the signature,
and compare it to the signature that I sent you. Since nobody else
knows the key, and the hash is not invertible, you know the message
must have come from me.

But if the random number generator is bad, then malicious peers could
guess the keys to create forged messages. The backdoor doesn't need to
be deterministic. It just needs a sufficiently nonuniform distribution
so that the attacker can test likely keys by computing signatures and
comparing them with the one I sent. If the distribution is uniform,
then the attack is not practical because it would take 2^256 tests to
find a match.


-- 
-- Matt Mahoney, [email protected]


-------------------------------------------
AGI
Archives: https://www.listbox.com/member/archive/303/=now
RSS Feed: https://www.listbox.com/member/archive/rss/303/21088071-f452e424
Modify Your Subscription: 
https://www.listbox.com/member/?member_id=21088071&id_secret=21088071-58d57657
Powered by Listbox: http://www.listbox.com

Reply via email to