[OTR-dev] mod_otr: man in the middle implementation for ejabberd

[Donny Viszneki]

On 3/31/07, Paul Wouters <paul at cypherpunks.ca> wrote:
> On Fri, 30 Mar 2007, Olivier Goffart wrote:
> > I made this module to show that it is not possible to make e2e encryption user
> > friendly.
> I would call OTR pretty userfriendly....

Please distinguish between user-friendly and idiot-friendly. If "it is
not possible to make e2e encryption user-friendly" because you choose
to lump complete morons in with the rest of the computer-using world,
how far are you from saying that password authentication isn't safe?
(Think about it -- even some of the most inane phishing scams can see
quite a bit of success.)

However, I do acknowledge that the idea of the fingerprint is not one
that has experienced a great deal of penetration into the collective
of mainstream computer users. I don't think this is the same as being
user-unfriendly. However in my personal experience, many mainstream
users are impatient when it comes to learning new ideas, and therefore
it may be best to attack the problem of mainstream penetration
head-on. I have a strange idea of how to do that which I'm sure others
have thought of, but I have been sitting on it for the better part of
a decade and so I figure I might as well throw it out there now that
the discussion has begun:

I have been rolling around an idea in my mind for a long time to
improve the utility of fingerprint/checksum mechanisms by making
fingerprints more memorable. What if the output of a hash weren't such
tightly packed, seemingly random data? What if you plugged
fingerprints into a dictionary file and got out a couple of words
instead? What if you plugged it into a clipart library? Or the library
of congress?

Your encryption fingerprint could be the face of George Washington,
five red-and-white striped ponies, or a line from a poem by Edgar
Allen Poe.

The more coherent and meaningful fingerprints/checksums are, the more
memorable they are, and IMHO that makes them more useful. One
challenge would be getting everyone to agree on what fingerprinting
mechanism / service to use. A much more interesting challenge though
would be studying what characteristics make a "humanistic
fingerprinting" mechanism most effective.

With the basic idea out of the way, here are a few elaborations:

Using familiar graphical transformations such as recoloring,
contorting, flipping, or tiling, you can increase the total output
potential of any graphical media library perahps a hundred fold. Maybe
the hash of the file you're looking for can be represented as a
multi-colored and tiled (like a Warhol) George Bush Sr. Or maybe it is
the upside-down silouette of a pine tree masked over the colors of the
rainbow.

Another idea would be to use a "generative" mechanism which would
constitute its own humanistic representation algorithmically without
requiring a multimedia library to draw from. It might produce anything
from something reminiscent of modern art, to techno music authored by
a psyconaut nearing the end of a four-day trip, or even the rorschac
you might get during a psychological evaluation: but it does have the
advantage of potentially being implemented with a small footprint and
wouldn't require access to an enormous media library to draw upon.

As mentioned earlier, I'm sure a lot of research could be invested in
determining which techniques produce the most effective output.

I have more to say on the subject but I just spent 6 minutes typing
that I really should be spending working! If anyone has any comments
feel free to email me. I really think that making fingerprinting a
visual experience would garner more attention from the mainstream as
well as developers for mainstream applications (I'd love to see the
tiled shape of a turtle like a work by Escher the next time a "Could
not verify SSL certificate" dialog pops up in Firefox ;)