Path: utzoo!utgpu!jarvis.csri.toronto.edu!mailrus!tut.cis.ohio-state.edu!ucbvax!SLCS.SLB.COM!7thSon
From: 7thSon@SLCS.SLB.COM (Chris Garrigues)
Newsgroups: comp.protocols.tcp-ip
Subject: Re: Subliminal (RFC1097)
Message-ID: <19890425150305.1.7THSON@GLOWWORM.LispM.SLCS.SLB.COM>
Date: 25 Apr 89 15:03:00 GMT
References: <3956@ficc.uu.net>
Sender: daemon@ucbvax.BERKELEY.EDU
Organization: The Internet
Lines: 40


    Date: 24 Apr 89 15:48:40 GMT
    From: peter%ficc%sugar%texbell@bellcore.com  (Peter da Silva)

    What sort of hardware is needed to implement this? The 1/60 second update
    rate of standard CRTs is too slow for subliminal messages, even ignoring
    considerations of phosphor decay rates. Do you need some sort of RS232-able
    tachistoscope to get the required ~2ms display time?

I was actually implementing it on a Lisp Machine and because this is an
AI machine, I can use various AI techniques to display the message very
quickly on a portion of the screen where I can be reasonably certain the
user isn't looking (by use of an expert system modeling the user's
visual system).

Admittedly, this solution isn't guaranteed to always keep the message at
a subliminal level, but we feel that our model is accurate enough (and
our users are sufficiently asleep), that the majority of them will never
know what hit their retina.

A second option I was considering was to simply display it as a
notification with the string "Don't read this:  " appended to the front,
but after some experimentation, I now feel that this might be too
obvious.

Two techniques which I haven't experimented with are (a) to dither the
message into the background pattern on my window system.  This would
make the message difficult to read, but isn't that the idea? and (b) to
use the audio system of the console to actually say the words very
quietly.  I suspect this would work on people like myself who work with
the radio on, but might be detectable by those who perfer to work in
silence.

I expect to have more success when I port to a color screen because I
can simply write the charaters with the color being almost, but not
quite, identical to the background color.  After all, can *you*
differentiate 16,777,216 different colors?  How to select the colors is
certainly a fascinating area for future research.

Chris