Skip to Content.
Sympa Menu

bluesky - Re: Think cash

bluesky AT lists.ibiblio.org

Subject: Global-Scale Distributed Storage Systems

List archive

Chronological Thread  
  • From: dmolnar <dmolnar AT hcs.harvard.edu>
  • To: Global-Scale Distributed Storage Systems <bluesky AT franklin.oit.unc.edu>
  • Subject: Re: Think cash
  • Date: Fri, 9 Mar 2001 08:50:21 -0500 (EST)




On Fri, 9 Mar 2001 graydon AT venge.net wrote:

> so the people's republic of china, looking at 1 image a minute, could
> completely defeat this system in 6 days. that is a quantitative

Depending on what we're trying to do, slowing down an automated attack may
be enough. For password systems, it's not enough. For using think cash as
an alternative to proofs of work in order to slow down DoS attacks, maybe
it is. Although what always bothers me about these sorts of solutions is
that even checking a challenge introduces overhead...so they don't solve
the problem so much as defray it.

> the audience with enough experience watching people wave their hands (or
> enough experience waving our own hands) will be justifiably skeptical.

Agreed. Which is why we should start looking for related work and start
trying to formalize the idea.

For instance, if we assume that a human has a x% chance of recognizing
pictures, and a computer is known to have a y% chance, and there's a gap,
what can we build from that gap? It seems to me the "right" thing to do
would be to abstract out what the problem is and start talking only in
terms of the human's advantage. (What notion of "advantage" makes sense
here, anyway?) Then we can worry about which problems are actually hard
independently of our think cash protocols.

Maybe we won't find any provable gaps. Maybe the gaps which we do find
will be too small for comfort (I doubt we'll find anything which seems
exponentially easier for a human, for instance).

I heard recently of a project at CMU which tried to leverage human pattern
recognition for passphrases; I'll see if I can find more information about
that.

This idea of a "reflected Turing test" also sounds interesting. I regret I
haven't read Ian's message in enough detail to comment on it fully. Still,
it sounds like it might be a good idea if you can assume that lots of
people are sitting at their computers in some p2p system and willing to
participate. So what kind of model can we build to test whether that's
true?

-David







Archive powered by MHonArc 2.6.24.

Top of Page